{ "per_task": [ { "task_group": "nexarm", "task_id": 0, "metrics": { "sum_rewards": [ 19.999741732082292, 19.9993749535682, 0.06517700312502606, 19.99918795441979, 19.999354785835855, 19.999305042391867, 19.99909444219284, 19.999436107739697, 20.00095343345978, 19.999353066623705, 19.99943955503608, 19.99858888041281, 0.3067627018883404, 0.47690094443258346, 19.999417557086495, 19.999019444517383, 19.99967162477491, 19.99886806614837, 19.99954142102483, 19.99945694381435, 19.99985162186975, 19.998493250340513, 19.998624170770473, 19.999152790034874, 19.999665839034755, 19.999831809627494, 19.99937823709292, 19.999756012829824, 19.99892343164552, 20.000706207496105, 19.998840280959566, 19.999456002190986, 19.99900940590086, 20.000036151128796, 20.00020383920245, 19.998818211190137, 0.06005205916227785, 20.0002179365243, 0.23741693254659801, 19.998953935165947, 19.99967361350237, 19.999820768445836, 19.999807071330597, 0.19664303896830715, 19.99983531862073, 19.999608240975867, 19.999810022688152, 19.99983114472411, 19.999972144703797, 19.999817461781358 ], "max_rewards": [ 10.0, 10.0, 0.029326789678992937, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 0.02948354137911824, 0.039734686012591, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0, 1.5702163283493002, 10.0, 0.027526364659749558, 10.0, 10.0, 10.0, 10.0, 0.038515615317530116, 10.0, 10.0, 10.0, 10.0, 10.0, 10.0 ], "successes": [ true, true, false, true, true, true, true, true, true, true, true, true, false, false, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, false, true, false, true, true, true, true, false, true, true, true, true, true, true ], "video_paths": [ "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_0.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_1.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_2.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_3.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_4.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_5.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_6.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_7.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_8.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_9.mp4" ], "predicted_video_paths": [], "stages": [ { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": false, "nested_middle": false, "released_middle": false, "picked_top": false, "nested_top": false, "released_top": false }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": false, "nested_middle": false, "released_middle": false, "picked_top": false, "nested_top": false, "released_top": false }, { "picked_middle": false, "nested_middle": false, "released_middle": false, "picked_top": false, "nested_top": false, "released_top": false }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": false, "picked_top": false, "nested_top": false, "released_top": false }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": false, "nested_middle": false, "released_middle": false, "picked_top": false, "nested_top": false, "released_top": false }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": false, "nested_middle": false, "released_middle": false, "picked_top": false, "nested_top": false, "released_top": false }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true }, { "picked_middle": true, "nested_middle": true, "released_middle": true, "picked_top": true, "nested_top": true, "released_top": true } ], "pc_picked_middle": 90.0, "pc_nested_middle": 90.0, "pc_released_middle": 88.0, "pc_picked_top": 88.0, "pc_nested_top": 88.0, "pc_released_top": 88.0 }, "n_episodes": 50, "n_success": 44, "pc_success": 88.0, "pc_success_ci95": [ 76.19518261679701, 94.38239984906772 ] } ], "per_group": { "nexarm": { "avg_sum_reward": 17.62641705222061, "avg_max_reward": 8.834696066507945, "pc_success": 88.0, "n_episodes": 50, "n_success": 44, "pc_success_ci95": [ 76.19518261679701, 94.38239984906772 ], "video_paths": [ "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_0.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_1.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_2.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_3.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_4.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_5.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_6.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_7.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_8.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_9.mp4" ], "predicted_video_paths": [], "pc_picked_middle": 90.0, "pc_nested_middle": 90.0, "pc_released_middle": 88.0, "pc_picked_top": 88.0, "pc_nested_top": 88.0, "pc_released_top": 88.0 } }, "overall": { "avg_sum_reward": 17.62641705222061, "avg_max_reward": 8.834696066507945, "pc_success": 88.0, "n_episodes": 50, "n_success": 44, "pc_success_ci95": [ 76.19518261679701, 94.38239984906772 ], "eval_s": 359.3926601409912, "eval_ep_s": 7.1878532123565675, "video_paths": [ "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_0.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_1.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_2.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_3.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_4.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_5.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_6.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_7.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_8.mp4", "/home/marinelab/code/plan-bind-d2/experiments/261028-plan-bind-d2-latency/outputs/full/eval/planbind_L0_none/videos/nexarm_0/eval_episode_9.mp4" ], "predicted_video_paths": [], "pc_picked_middle": 90.0, "pc_nested_middle": 90.0, "pc_released_middle": 88.0, "pc_picked_top": 88.0, "pc_nested_top": 88.0, "pc_released_top": 88.0 } }