Add files using upload-large-folder tool

.gitattributes

@@ -473,3 +473,51 @@ bmk18y9z/episode_rosbags/episode_6_2025-01-07-22-54-12.bag filter=lfs diff=lfs m
 4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/11_eval_11_6141e16e285aea41414e.mp4 filter=lfs diff=lfs merge=lfs -text
 4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/12_eval_12_212b177b80b2034deb4f.mp4 filter=lfs diff=lfs merge=lfs -text
 4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/9_eval_9_c7576684fb563a572588.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/2_eval_2_537ec4e4ec2357120fef.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/3_eval_3_5243c77fb64415fcde38.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/4_eval_4_123736b8ecd81d086c47.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/18_eval_18_fd5b9c7ce55fe4462130.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/13_eval_13_49e329d2ff2e935251c1.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/14_eval_14_b1f7698aaf2c3d6e6abf.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/16_eval_16_a93b394f6e07a47f91ce.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/6_eval_6_718698df80de1dc3b1e7.mp4 filter=lfs diff=lfs merge=lfs -text
+c9fefav0/episode_rosbags/episode_0_2025-01-06-21-06-27.bag filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/17_eval_17_e673279ceb422c546539.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/15_eval_15_b314c0181db144f9c4bb.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/7_eval_7_0c0abd0712ae602c07c3.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/0_eval_0_ae65028e83f44b9ac5b1.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/19_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/3_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/12_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+c9fefav0/episode_rosbags/episode_4_2025-01-06-21-09-26.bag filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/7_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/8_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/4_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/13_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/9_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/11_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/18_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/15_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/6_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/1_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/5_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/16_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/2_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/17_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/10_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/0_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/eval_video/14_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+3d3i07nt/episode_rosbags/episode_2_2025-01-06-21-02-35.bag filter=lfs diff=lfs merge=lfs -text
+4fmihp69/episode_rosbags/episode_3_2025-01-07-22-09-09.bag filter=lfs diff=lfs merge=lfs -text
+4fmihp69/episode_rosbags/episode_4_2025-01-07-22-09-52.bag filter=lfs diff=lfs merge=lfs -text
+bkletf5j/wandb/run-20250106_205004-bkletf5j/files/media/videos/eval/3_eval_3_8f993ed9f4bf0eae5cf1.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/wandb/run-20250106_205004-bkletf5j/files/media/videos/eval/0_eval_0_d5629c4a8c29fc9b2a6b.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/wandb/run-20250106_205004-bkletf5j/files/media/videos/eval/4_eval_4_6a8935b2358637140fbc.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/wandb/run-20250106_205004-bkletf5j/files/media/videos/eval/1_eval_1_b0cb5720b5b83600f051.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/wandb/run-20250106_205004-bkletf5j/files/media/videos/eval/2_eval_2_4cf0e0208d2fc5ae5937.mp4 filter=lfs diff=lfs merge=lfs -text
+4fmihp69/episode_rosbags/episode_2_2025-01-07-22-08-27.bag filter=lfs diff=lfs merge=lfs -text
+bkletf5j/eval_video/3_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/eval_video/4_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/eval_video/1_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/eval_video/2_eval.mp4 filter=lfs diff=lfs merge=lfs -text
+bkletf5j/eval_video/0_eval.mp4 filter=lfs diff=lfs merge=lfs -text

3d3i07nt/episode_rosbags/episode_2_2025-01-06-21-02-35.bag

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db0b689d49eea41c1743f5492b49ed351384554397594bb2fb35b8906956e607
+size 1413121843

4fmihp69/.hydra/config.yaml

@@ -0,0 +1,350 @@
+root_dir: /home/${oc.env:USER}/fish_leon
+nstep: 3
+seed: 41
+dataset_shuffle_seed: ${seed}
+device: cuda
+save_video: true
+save_buffer: true
+use_tb: true
+baseline: false
+use_wandb: true
+eval: true
+process_contact_features: ${eval}
+obs_type: pixels
+use_color: true
+use_depth: true
+use_masks: false
+mask_list:
+- EE_obj_mask
+mask_representation: channels
+crop_hw:
+- 144
+- 144
+crop_down_offset: 48
+color_crop_type: null
+depth_crop_type: null
+segmask_crop_type: null
+add_crop_binary_mask: false
+add_coord_conv_map: false
+use_context_color: false
+use_context_depth: false
+use_context_segmask: false
+context_color_crop_type: null
+context_depth_crop_type: null
+context_segmask_crop_type: null
+context_add_crop_binary_mask: false
+context_add_coord_conv_map: false
+use_contact_map: false
+use_sdf_maps: false
+use_normals_maps: false
+which_objects: both
+max_contact_prob: 0.1
+max_depth: 2.0
+grasped_dtc_max_value: 0.105
+env_dtc_max_value: 0.425
+grasped_normals_mask_max_dtc_value: 0.105
+env_normals_mask_max_dtc_value: 0.425
+clamp_dtc: true
+dtc_adaptive_normalization: false
+mask_normals_within_sdf: true
+adaptive_normals_mask: true
+learnable_contact_preprocess_params: false
+contact_model_name: local_multitask_outhd64all_home_crop_h144w144d48_ctxt_seed_183386_epoch_9
+contact_estimation_model_ckpt_path: ~/fish_leon/contact_estimation/artifacts/175604_2/checkpoints/epoch=09-val_loss=0.00.ckpt
+num_eval: 20
+debug_timestamps: false
+open_loop: false
+action_trajectories: true
+stop_after_action: false
+interpolation_frequency: 25
+policy_frequency: 5
+wait_for_new_camera_frames: true
+random_start: false
+eval_starts: ${root_dir}/FISH/eval_starts/${suite.name}_${obs_type}/${task_name}
+train_demo_idxs_list_or_num: null
+num_valid_demos: null
+val_num_groups: 3
+name_of_expert_demo: 112_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act
+expert_dataset_dirpath: ${root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${name_of_expert_demo}
+expert_dataset: ${expert_dataset_dirpath}/demos.zarr
+action_key: ${oc.if_else:${action_trajectories}, 'action_trajectory_${interpolation_frequency}hz',
+  'action'}
+semantic_demo_grouping_name: semantic_demo_grouping.yaml
+semantic_demo_grouping: ${expert_dataset_dirpath}/${semantic_demo_grouping_name}
+expert_dataset_config: ${expert_dataset_dirpath}/demo_config.yaml
+bc_regularize: false
+bc_weight_type: qfilter
+load_checkpoint: ${agent.load_checkpoint}
+wandb_run_id: '1002_0'
+true_action_history: false
+wandb_notes: null
+checkpoint_epoch: 12000
+load_residual_weight: false
+checkpoint_root_dir: /home/${oc.env:USER}/fish_leon/FISH
+checkpoint_weight_dir: ${checkpoint_root_dir}/exp_local/${suite.name}_${obs_type}/${task_name}/${wandb_run_id}
+residual_weight: ${root_dir}/FISH/weights/${suite.name}_${obs_type}/${task_name}/weight.pt
+experiment_dir: ./exp_local/${suite.name}_${obs_type}/${task_name}/${wandb_run_id}
+final_experiment_dir: ${experiment_dir}/${oc.generate_run_id:}
+agent:
+  _target_: agent.diffusion_policy.DiffusionPolicyAgent
+  name: diffusion_policy
+  load_checkpoint: ${eval}
+  device: ${device}
+  n_obs_steps: ${.config.policy_cfg.n_obs_steps}
+  suite_name: ${suite.name}
+  obs_type: ${obs_type}
+  enable_arm: ${eval}
+  enable_camera: ${eval}
+  use_tb: ${use_tb}
+  desired_image_shape:
+  - 13
+  - 180
+  - 240
+  orig_cam_shape:
+  - 3
+  - 240
+  - 320
+  config:
+    _target_: agent.diffusion_policy.DiffusionPolicyAgentConfig
+    compile: false
+    device: ${device}
+    cam_resize_shape: ${agent.desired_image_shape}
+    orig_cam_shape: ${agent.orig_cam_shape}
+    policy_frequency: ${policy_frequency}
+    interpolation_frequency: ${interpolation_frequency}
+    policy_cfg:
+      _target_: lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig
+      n_obs_steps: 1
+      horizon: 36
+      n_action_steps: ${agent.config.policy_cfg.horizon}
+      input_shapes:
+        observation.image: ${agent.config.cam_resize_shape}
+        context_observation.image: ${agent.config.cam_resize_shape}
+        observation.state:
+        - 8
+        observation.action_history:
+        - 7
+      output_shapes:
+        action:
+        - 7
+      input_normalization_modes:
+        observation.image: mean_std
+        observation.state: min_max
+        observation.action_history: min_max
+      output_normalization_modes:
+        action: min_max
+      vision_backbone: resnet18
+      pretrained_backbone_weights: null
+      transforms:
+      - _target_: torchaug.transforms.RandomAffine
+        degrees:
+        - -5
+        - 5
+        translate:
+        - 0.05
+        - 0.05
+        batch_transform: true
+        num_chunks: -1
+        batch_inplace: true
+      - _target_: torchaug.transforms.RandomColorJitter
+        brightness: 0.3
+        contrast: 0.4
+        saturation: 0.5
+        hue: 0.08
+        batch_transform: true
+        num_chunks: -1
+        batch_inplace: true
+      use_group_norm: true
+      spatial_softmax_num_keypoints: 32
+      action_history_encoder_config:
+        _target_: lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig
+        in_channels: 7
+        out_channels: 32
+        history_length: ${agent.config.policy_cfg.n_action_steps}
+        kernel_size: ${agent.config.policy_cfg.kernel_size}
+        downsample_kernel_size: 3
+        downsample_stride: 2
+        downsample_padding: 1
+      down_dims:
+      - 256
+      - 512
+      - 1024
+      kernel_size: 5
+      n_groups: 8
+      diffusion_step_embed_dim: 128
+      use_film_scale_modulation: true
+      noise_scheduler_type: DDIM
+      beta_schedule: squaredcos_cap_v2
+      beta_start: 0.0001
+      beta_end: 0.02
+      prediction_type: epsilon
+      clip_sample: true
+      clip_sample_range: 1.0
+      num_train_timesteps: 50
+      num_inference_steps: 10
+      do_mask_loss_for_padding: false
+    train_cfg:
+      _target_: utils.TrainConfig
+      lr: 0.0001
+      lr_scheduler: cosine
+      lr_warmup_steps: 500
+      adam_betas:
+      - 0.95
+      - 0.999
+      adam_eps: 1.0e-08
+      adam_weight_decay: 1.0e-06
+      grad_clip_norm: 10
+      offline_steps: ${num_train_frames_diffusion}
+      use_amp: true
+    observation_cfg:
+      _target_: agent.encoder.VisualFeatureSet
+      use_depth: ${use_depth}
+      use_color: ${use_color}
+      mask_input_dict:
+        _target_: agent.encoder.MaskInputDict
+        enable: ${use_masks}
+        representation: ${mask_representation}
+        mask_list: ${mask_list}
+      crop_input_config:
+        _target_: agent.encoder.CropInputConfig
+        color_crop_type: ${color_crop_type}
+        depth_crop_type: ${depth_crop_type}
+        segmask_crop_type: ${segmask_crop_type}
+        crop_hw: ${crop_hw}
+        crop_down_offset: ${crop_down_offset}
+        add_crop_binary_mask: ${add_crop_binary_mask}
+        add_coord_conv_map: ${add_coord_conv_map}
+      context_input_config:
+        _target_: agent.encoder.ContextInputConfig
+        use_color: ${use_context_color}
+        use_depth: ${use_context_depth}
+        mask_input_dict:
+          _target_: agent.encoder.MaskInputDict
+          enable: ${use_context_segmask}
+          representation: ${mask_representation}
+          mask_list: ${mask_list}
+        crop_input_config:
+          _target_: agent.encoder.CropInputConfig
+          color_crop_type: ${context_color_crop_type}
+          depth_crop_type: ${context_depth_crop_type}
+          segmask_crop_type: ${context_segmask_crop_type}
+          crop_hw: ${crop_hw}
+          crop_down_offset: ${crop_down_offset}
+          add_crop_binary_mask: ${context_add_crop_binary_mask}
+          add_coord_conv_map: ${context_add_coord_conv_map}
+      mask_soft_approx_scheduler_config:
+        _target_: agent.encoder.MaskSoftApproxSchedulerConfig
+        num_steps: 40000
+        initial_value: 10.0
+        final_value: 1000.0
+        interpolation_scheme: constant
+      use_contact_map: ${use_contact_map}
+      use_sdf_maps: ${use_sdf_maps}
+      use_normals_maps: ${use_normals_maps}
+      which_objects: ${which_objects}
+      grasped_dtc_max_value: ${grasped_dtc_max_value}
+      env_dtc_max_value: ${env_dtc_max_value}
+      grasped_normals_mask_max_dtc_value: ${grasped_normals_mask_max_dtc_value}
+      env_normals_mask_max_dtc_value: ${env_normals_mask_max_dtc_value}
+      clamp_dtc: ${clamp_dtc}
+      max_contact_prob: ${max_contact_prob}
+      mask_normals_within_sdf: ${mask_normals_within_sdf}
+      dtc_adaptive_normalization: ${dtc_adaptive_normalization}
+      adaptive_normals_mask: ${adaptive_normals_mask}
+      max_depth: ${max_depth}
+      image_shape: ${agent.desired_image_shape}
+      learnable_contact_preprocess_params: ${learnable_contact_preprocess_params}
+      learning_rate: ${agent.config.train_cfg.lr}
+      weight_decay: 0.0
+      contact_model_name: ${contact_model_name}
+      zero_centered: false
+suite:
+  suite: frankagym
+  name: frankagym
+  frame_stack: ${agent.n_obs_steps}
+  action_repeat: 1
+  discount: 0.99
+  hidden_dim: 1024
+  num_train_frames: 2010
+  num_seed_frames: 260
+  num_train_epochs: 5000
+  validate_every_epochs: 100
+  validate_diffusion_on_action_loss_every_epochs: 500
+  train_eval_diffusion_on_action_loss_every_epochs: 500
+  check_topk_every_epochs: 10
+  save_snapshot_every_epochs: 5000
+  eval_every_frames: 2000
+  num_eval_episodes: 5
+  save_snapshot: true
+  wait_for_user_to_start_episode: true
+  task_make_fn:
+    _target_: suite.frankagym.make
+    name: ${task_name}
+    height: 240
+    width: 320
+    frame_stack: ${suite.frame_stack}
+    action_repeat: ${suite.action_repeat}
+    seed: ${seed}
+    enable_arm: ${agent.enable_arm}
+    enable_gripper: ${enable_gripper}
+    start_with_gripper_open: ${start_with_gripper_open}
+    enable_camera: ${agent.enable_camera}
+    path_to_depth_extrinsics: ${path_to_depth_extrinsics}
+    contact_estimation_model_ckpt_path: ${contact_estimation_model_ckpt_path}
+    x_limit: ${x_limit}
+    y_limit: ${y_limit}
+    z_limit: ${z_limit}
+    device: ${device}
+    interpolation_frequency: ${interpolation_frequency}
+    policy_frequency: ${policy_frequency}
+    debug_timestamps: ${debug_timestamps}
+    stop_after_action: ${stop_after_action}
+    open_loop: ${open_loop}
+    wait_for_new_camera_frames: ${wait_for_new_camera_frames}
+    action_key: ${action_key}
+    action_trajectory_horizon: ${agent.config.policy_cfg.horizon}
+    action_trajectories: ${action_trajectories}
+    path_to_zarr_dataset: ${expert_dataset}
+    agent_policy_cfg: ???
+    true_action_history: ${true_action_history}
+num_train_frames_bc: 50000
+num_train_frames_drq: 1100000
+stddev_schedule_drq: linear(1.0,0.1,100000)
+task_name: FrankaInsertion-v1
+num_train_frames_vinn: 25000
+num_train_frames_diffusion: 1000000
+num_train_epochs_bc: 5000
+num_train_epochs_diffusion: 5000
+validate_every_epochs_bc: 5
+validate_every_epochs_diffusion: 25
+validate_diffusion_on_action_loss_every_epochs: 50
+train_eval_diffusion_on_action_loss_every_epochs: 500
+check_topk_every_epochs: 5
+check_topk_every_epochs_diffusion: ${validate_diffusion_on_action_loss_every_epochs}
+save_snapshot_every_epochs_diffusion: 5000
+x_limit:
+- 0.2
+- 0.7
+y_limit:
+- -0.4
+- 0.4
+z_limit:
+- -0.05
+- 0.55
+home_displacement:
+- 0.55
+- 0.0
+- 0.55
+- 180.0
+- 0.0
+- 0.0
+enable_gripper: true
+start_with_gripper_open: true
+offset_mask:
+- 1
+- 1
+- 1
+- 1
+- 1
+- 1
+path_to_depth_extrinsics: ~/fish_leon/FISH/cfgs/camera_poses/camera_poses_L515/20240904-122305/color_tf_world.npy

4fmihp69/.hydra/hydra.yaml

@@ -0,0 +1,169 @@
+hydra:
+  run:
+    dir: ${final_experiment_dir}
+  sweep:
+    dir: ${final_experiment_dir}
+    subdir: ${hydra.job.num}
+  launcher:
+    submitit_folder: ${final_experiment_dir}/.slurm
+    timeout_min: 60
+    cpus_per_task: null
+    gpus_per_node: null
+    tasks_per_node: 1
+    mem_gb: null
+    nodes: 1
+    name: ${hydra.job.name}
+    stderr_to_stdout: false
+    _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.LocalLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - agent=diffusion
+    - suite=frankagym
+    - suite/frankagym_task@_global_=insertion
+  job:
+    name: eval_robot
+    chdir: true
+    override_dirname: agent=diffusion,suite/frankagym_task@_global_=insertion,suite=frankagym
+    id: ???
+    num: ???
+    config_name: config_eval
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.1'
+    cwd: /home/leonmkim/fish_leon/FISH
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/leonmkim/fish_leon/FISH/cfgs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4fmihp69
+    choices:
+      suite: frankagym
+      suite/frankagym_task@_global_: insertion
+      agent: diffusion
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: submitit_local
+      hydra/output: default
+  verbose: false

4fmihp69/.hydra/overrides.yaml

@@ -0,0 +1,3 @@
+- agent=diffusion
+- suite=frankagym
+- suite/frankagym_task@_global_=insertion

4fmihp69/episode_rosbags/episode_2_2025-01-07-22-08-27.bag

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:043761c26ac246cb92cfbae7e064fd8672f3f268a46a6105d061ca36f4b0d20c
+size 1411056075

4fmihp69/episode_rosbags/episode_3_2025-01-07-22-09-09.bag

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b24d39303ac5a858f58b6d42916ba22dbfb88407b11a9c5201329b098070c501
+size 1407832609

4fmihp69/episode_rosbags/episode_4_2025-01-07-22-09-52.bag

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11d8c765f1dca02888e208646b35a256943ec9952c29d3b8327193fae3a9bc1a
+size 1411111409

4fmihp69/eval_video/0_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae65028e83f44b9ac5b1292e9a19a482040bef9b85dc8452c1003e498adcf200
+size 1203236

4fmihp69/eval_video/10_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e9fbf635a50e60a4cb3bc19311a779d79512d4e50a6bdec2a27735c3a018973
+size 1317315

4fmihp69/eval_video/11_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6141e16e285aea41414e683be227f5ddaab91e58c74c8fe9a6908e9a5ab511ed
+size 1325039

4fmihp69/eval_video/12_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:212b177b80b2034deb4f20821b47755bf2b9d214e66cc422750d97dfd12630bc
+size 1365265

4fmihp69/eval_video/13_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49e329d2ff2e935251c138c38bd869e83a370b6a20c29a6f3f6f4fb5241a5926
+size 1214289

4fmihp69/eval_video/14_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1f7698aaf2c3d6e6abf7855ee3e636573dce1a6d4641acb7f13c8ce0412b52c
+size 1410819

4fmihp69/eval_video/15_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b314c0181db144f9c4bbd755580f55f4fa904dbd7bf96ae9ebbdf5aa6aea7161
+size 1504211

4fmihp69/eval_video/16_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a93b394f6e07a47f91ce5c6436e2ac6f677aa64e46d4e4c1331cad98e6429f67
+size 1441574

4fmihp69/eval_video/17_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e673279ceb422c54653970040c48bf551dbc8cea9639629ae5c5f872e26f2b86
+size 1438873

4fmihp69/eval_video/18_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd5b9c7ce55fe446213004b70d71e620113f558086949526d140680e6e08433b
+size 1143394

4fmihp69/eval_video/19_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b857c64ab339c17adb153080a0a92442b3ece13d773e460489d0645e2a50a415
+size 1247835

4fmihp69/eval_video/1_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5acd6aafe960785647d8b9e25843b0bfe650a9f5b093f48cab02c8bb3ca18dc
+size 1200266

4fmihp69/eval_video/2_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:537ec4e4ec2357120fef5cc7f4edbf14515d4b2e6bdfab19c7eeeacbe708bc89
+size 1203819

4fmihp69/eval_video/3_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5243c77fb64415fcde38b5e587dfe83a26bc73991f69876d74ed1cbc67f2c743
+size 1150313

4fmihp69/eval_video/4_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:123736b8ecd81d086c472194678ae8960360473b3f54800e2d9f897b23c89739
+size 1255949

4fmihp69/eval_video/5_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdc3dec2e2f95a3c767dc580906d7f9f320e106e08f6a4b27752d14fc60dd841
+size 1148520

4fmihp69/eval_video/6_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:718698df80de1dc3b1e7aea8932c6797bc8d2289782b524986f54f2abbffc39c
+size 1163347

4fmihp69/eval_video/7_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c0abd0712ae602c07c33b31081a76a7241f86938b9d37db8c91daabb3f3f779
+size 1201147

4fmihp69/eval_video/8_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8cc872cbcaee100937a0d3546df9603678e05465b5928a259b9ed6b4d1ae545
+size 1257988

4fmihp69/eval_video/9_eval.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7576684fb563a572588fe00177b0da5477339e45aaf0f59ca72100987f63cfa
+size 1299376

4fmihp69/tb/events.out.tfevents.1736305571.leonmkim-ROG-Strix-G15CS-G15CS.2240043.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca15f9bcf722c39eaf851a1c07befd8105ae64c156772bc2d5af22a4b8608749
+size 22784

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/code/FISH/eval_robot.py

@@ -0,0 +1,605 @@
+#%%
+import warnings
+import os
+
+os.environ['MKL_SERVICE_FORCE_INTEL'] = '1'
+os.environ['MUJOCO_GL'] = 'egl'
+from pathlib import Path
+#%%
+import hydra
+import numpy as np
+import torch
+
+import utils
+from utils import get_feature_dirname_from_configs
+
+from video import VideoRecorder
+import pickle
+import time
+import threading
+import shutil
+from logger import Logger
+
+import wandb 
+from omegaconf import OmegaConf, open_dict
+
+from replay_buffer_robot import RosbagEvalReplayBufferStorage
+from lerobot.common.utils.utils import _relative_path_between
+
+torch.backends.cudnn.benchmark = True
+warnings.filterwarnings('ignore', category=DeprecationWarning)
+
+# import specs for replay buffer
+from dm_env import specs
+
+import sys, signal
+import yaml
+
+import binomial_cis as bc
+
+# get path of current file
+current_path = os.path.dirname(os.path.realpath(__file__))
+sys.path.append(os.path.join(current_path, os.pardir))
+# from contact_estimation.src.utils.viz_utils import normalized_surface_normal_to_rgb, depth_map_to_im, grasped_env_dtc_map_to_im, contact_prob_map_to_im, desaturate_color_image, masked_overlay_im_list
+
+def make_agent(obs_spec, action_spec, cfg):
+	cfg.obs_shape = obs_spec['pixels'].shape
+	dataset_statistics = None # this will be loaded from the checkpoint
+	try:
+		cfg.action_shape = action_spec.shape
+	except:
+		pass
+	return hydra.utils.instantiate(cfg, dataset_statistics)
+
+class Workspace:
+	def __init__(self, cfg):
+		self.work_dir = Path.cwd()
+		print(f'workspace: {self.work_dir}')
+		
+		signal.signal(signal.SIGINT, self.signal_handler)
+		
+		self.cfg = cfg
+		self.loading_uncompiled_checkpoint_with_compile = False
+		self.loading_compiled_checkpoint_with_no_compile = False
+
+		snapshot_path = Path(self.cfg.checkpoint_weight_dir) / f'snapshot_{self.cfg.checkpoint_epoch}.pt' 
+		self.load_checkpoint_conf(snapshot_path=snapshot_path)
+
+		# load config for action trajectories
+		utils.set_seed_everywhere(self.cfg.seed)
+		self.device = torch.device(self.cfg.device)
+		self.setup()
+
+		# self.agent = make_agent(self.eval_env.observation_spec(),
+		# 						self.eval_env.action_spec(), self.cfg.agent)
+		self.timer = utils.Timer()
+		# self._global_step = 0
+		self._global_episode = 0
+		self._global_epoch = 0
+		self.num_episode_successes = 0
+
+		self.alpha_range = [.01, .025, .05, .1]
+
+		# Need to convert hydra config to primitive container for wandb https://docs.wandb.ai/guides/integrations/hydra
+		with open_dict(self.cfg):
+			self.cfg.feature_type = get_feature_dirname_from_configs(
+				hydra.utils.instantiate(self.cfg.agent.config.observation_cfg), 
+				self.cfg.agent.config.policy_cfg.input_shapes,
+				hydra.utils.instantiate(self.cfg.agent.config.policy_cfg.action_history_encoder_config) if 'observation.action_history' in self.cfg.agent.config.policy_cfg.input_shapes else None,
+			)
+
+		wandb_config = OmegaConf.to_container(
+			self.cfg, resolve=True, throw_on_missing=True
+		)
+		# must be called before any tf summary writer is created
+		if self.cfg.use_wandb:
+			# get the run id from the final_experiment_dir directory
+			run_id = os.path.basename(os.path.normpath(self.cfg.final_experiment_dir))
+			wandb.init(project='extrinsic_contact_downstream', entity='serialexperimentsleon', job_type='eval', sync_tensorboard=self.cfg.use_tb, config=wandb_config, id=run_id)
+
+		self.logger = Logger(self.work_dir, use_tb=self.cfg.use_tb, use_wandb=self.cfg.use_wandb)
+
+		# if not self.loading_uncompiled_checkpoint_with_compile and self.cfg.agent.config.compile:
+		# 	self.agent.compile_modules()
+			
+		# self.load_checkpoint(snapshot_path=snapshot_path)
+
+		# if self.loading_uncompiled_checkpoint_with_compile: # need to call compile after loading the checkpoint
+		# 	self.agent.compile_modules()
+
+		print(f"loaded agent with feature_type: {self.cfg.feature_type}")
+
+	def check_for_key_press(self):
+		while self.continue_keypress_thread:
+			inp = input("Press 'r' to restart current episode, 'n' to stop current episode and skip to next, 'q' to break entire eval\n")
+			if inp == 'n':
+				self.preempt_episode = True
+				print("preempting episode")
+			elif inp in ['', '0', '1']: # enter key
+				if inp in ['0', '1']:
+					self.num_episode_successes += int(inp)
+				self.proceed_after_env_reset_event.set()
+				print("proceeding to start episode!")
+			elif inp == 'q':
+				self.proceed_after_env_reset_event.set()
+				self.preempt_episode = True
+				self.exit_eval = True
+				self.continue_keypress_thread = False # will stop the keypress thread
+				print("quitting eval")
+				break
+			elif inp == 'r':
+				print('restarting episode')
+				self.preempt_episode = True
+				self.restart_episode = True
+			else:
+				print("Invalid key press, try again")
+		
+		# self.keypress_input_thread.join() # wait for the keypress thread to finish
+
+	def signal_handler(self, signal, frame):
+		print("\nprogram exiting gracefully")
+		self.proceed_after_env_reset_event.set()
+		self.preempt_episode = True
+		self.exit_eval = True
+		self.continue_keypress_thread = False # will stop the keypress thread
+		self.keypress_input_thread.join() # wait for the keypress thread to finish
+		video_filepath = self.video_recorder.save()
+		# get the video file and convert to video tensor to log 
+		self.logger.log_video('eval/video', video_filepath, self.global_step)
+		wandb.finish()
+		sys.exit(0)
+		
+	def setup(self):
+		# create envs
+		self.eval_env = hydra.utils.call(self.cfg.suite.task_make_fn)
+		# expert_demo_config_path = os.path.join(os.path.dirname(self.cfg.expert_dataset), 'demo_config.yaml')
+		# self.expert_demo_config = yaml.load(open(expert_demo_config_path, 'r'), Loader=yaml.FullLoader)
+		# self.eval_env._env.action_trans_norm = expert_demo_config['max_translation_action_norm']
+		# self.eval_env._env.action_rot_norm = expert_demo_config['max_rotation_action_norm']
+		# self.eval_env._env.action_period = expert_demo_config['sample_period']
+		# print(f"setting max_translation_action_norm to {expert_demo_config['max_translation_action_norm']} and sample_period to {expert_demo_config['sample_period']}")
+		# print(f"setting max_rotation_action_norm to {expert_demo_config['max_rotation_action_norm']}")
+
+		# self.eval_env.set_demo_params(self.cfg.expert_dataset)
+
+		# Turn off random start
+		self.eval_env.random_start = False
+
+		# create replay buffer
+		# data_specs = [
+		# 	{
+		# 		'observation': self.eval_env.observation_spec(),
+		# 	},
+		# 	# self.eval_env.observation_spec()['features'],
+		# 	self.eval_env.action_spec(),
+		# 	specs.Array(self.eval_env.action_spec().shape, self.eval_env.action_spec().dtype, 'vinn_action'), 
+		# 	specs.Array((1, ), np.float32, 'reward'),
+		# 	specs.Array((1, ), np.float32, 'discount'),
+		# ]
+
+		# self.eval_replay_storage = ZarrEvalReplayBufferStorage(data_specs, self.work_dir / 'eval_buffer', debug_timestamps=self.cfg.debug_timestamps, save_buffer=self.cfg.save_buffer, debug_info_data_specs=self.eval_env.debug_info_data_specs, camera_info_dict=self.eval_env.get_camera_info_dict())
+		self.eval_replay_storage = RosbagEvalReplayBufferStorage(self.work_dir)
+
+		self.video_recorder = VideoRecorder(
+			self.work_dir if self.cfg.save_video else None,
+			ros_enabled=True,
+			fps=self.cfg.agent.config.policy_frequency,
+			)
+		
+		print('workspace setup complete')
+		
+	@property
+	def global_step(self):
+		# return self._global_step
+		return self.eval_env.get_global_step()
+
+	@property
+	def global_episode(self):
+		return self._global_episode
+
+	@property
+	def global_frame(self):
+		return self.global_step * self.cfg.action_repeat
+
+	@property
+	def global_epoch(self):
+		return self._global_epoch
+	
+	def reset(self, eval_idx):
+		if not self.eval_env.enable_arm:
+			return np.array([0,0,0], dtype=np.float32)
+		self.eval_env.arm_refresh(reset=False)
+		# Set start position
+		try:
+			self.eval_env.set_position(self.start_pos[eval_idx])
+		except:
+			self.eval_env.arm.set_position(self.start_pos[eval_idx])
+			if self.eval_env.arm.keep_gripper_closed:
+				self.eval_env.arm.close_gripper_fully()
+			else:
+				self.eval_env.arm.open_gripper_fully()
+		time.sleep(0.1)
+		time_step = self.eval_env.step(np.zeros(self.eval_env.action_spec().shape[0], dtype=np.float32),
+									   np.zeros(self.eval_env.action_spec().shape[0], dtype=np.float32))	
+		return time_step
+
+	def eval(self):
+		# before evals start, prompt user for name of grasped object and the left book of the slot location
+		grasped_obj_name = input("Enter the name of the grasped object: ")
+		left_book_slot = input("Enter the left book slot location: ")
+		# update wandb config
+		if self.cfg.use_wandb:
+			wandb.config.update({'grasped_obj_name': grasped_obj_name, 'left_book_slot': left_book_slot})
+
+		self.preempt_episode = False
+		self.exit_eval = False
+		self.restart_episode = False
+
+		self.continue_keypress_thread = True
+		self.proceed_after_env_reset_event = threading.Event()
+		self.keypress_input_thread = threading.Thread(target=self.check_for_key_press)
+		self.keypress_input_thread.start()
+
+		# # Set model to eval mode
+		# self.agent.train(False)
+		
+		eval_until_episode = utils.Until(self.cfg.num_eval)
+
+		self.use_action_history = False
+		# if "dp" in repr(self.agent) and "observation.action_history" in self.cfg.agent.config.policy_cfg.input_shapes:
+		if "observation.action_history" in self.cfg.agent.config.policy_cfg.input_shapes:
+			self.use_action_history = True
+
+		# self.eval_replay_storage._new_eval_step(0)
+
+		# if 'vinn' in repr(self.agent) or 'openloop' in repr(self.agent):
+		# 	with open(self.cfg.expert_dataset, 'rb') as f:
+		# 		if self.cfg.obs_type == 'pixels':
+		# 			self.expert_demo, _, self.expert_action, self.expert_reward = pickle.load(f)
+		# 		elif self.cfg.obs_type == 'features':
+		# 			_, self.expert_demo, self.expert_action, self.expert_reward = pickle.load(f)
+				
+		# 		if self.cfg.action_trajectories:
+		# 			with open(self.cfg.expert_action_trajectories, 'rb') as f:
+		# 				self.expert_action = pickle.load(f)
+
+		# 		if isinstance(self.cfg.train_demo_idxs_list_or_num, int):
+		# 			if self.cfg.train_demo_idxs_list_or_num == -1:
+		# 				self.cfg.train_demo_idxs_list_or_num = len(self.expert_demo)
+		# 			train_demo_idxs_list_or_num = list(range(self.cfg.train_demo_idxs_list_or_num))
+				
+		# 		self.expert_demo = self.expert_demo[train_demo_idxs_list_or_num]
+		# 		self.expert_action = self.expert_action[train_demo_idxs_list_or_num]
+		# 		self.expert_reward = self.expert_reward[train_demo_idxs_list_or_num]
+		# 		# if self.cfg.action_plans:
+		# 		# 	self.expert_action_plans = self.expert_action_plans[self.cfg.train_demo_idxs_list_or_num]
+		# 		# self.expert_demo = self.expert_demo[:self.cfg.num_demos]
+		# 		# self.expert_action = self.expert_action[:self.cfg.num_demos]
+		# 		# self.expert_reward = self.expert_reward[:self.cfg.num_demos]
+
+		# 		self.expert_demo = np.concatenate(self.expert_demo, axis=0)
+		# 		self.expert_rgb_obs = np.ascontiguousarray(np.transpose(self.expert_demo, (0,2,3,1))[:, :,:,:3].astype(np.uint8))
+		# 		self.expert_action = np.concatenate(self.expert_action, axis=0)
+				
+		# 		self.agent.save_representations(self.expert_demo, self.expert_action, 128, config=self.expert_demo_config)
+		
+		# Get start points
+		if self.cfg.random_start:
+			eval_starts = Path(self.cfg.eval_starts) / 'starts.pkl'
+			if eval_starts.exists():
+				with eval_starts.open('rb') as f:
+					self.start_pos = pickle.load(f)
+			else:
+				eval_starts = Path(self.cfg.eval_starts)
+				eval_starts.mkdir(parents=True, exist_ok=True)
+				
+				# Generate start points
+				self.start_pos = []
+				try:
+					for _ in range(self.cfg.num_eval):
+						self.start_pos.append(self.eval_env.get_random_pos())
+				except:
+					for _ in range(self.cfg.num_eval):
+						self.start_pos.append(self.eval_env.arm.get_random_pos())
+
+				# Save start points for the task
+				eval_starts = eval_starts / 'starts.pkl'
+				with eval_starts.open('wb') as f:
+					pickle.dump(self.start_pos, f)
+
+		time_step = self.eval_env.reset()
+		# replay_thread = None
+		while eval_until_episode(self.global_episode) and not self.exit_eval:
+			# self.video_recorder.init(self.eval_env, video_filename=f'{self.global_episode}_eval.mp4')
+			print(f"Starting episode {self.global_episode}")
+			time_step = self.eval_env.reset() #Leon: need to call reset twice in case objects are trapped
+			self.video_recorder.init(self.eval_env, video_filename=f'{self.global_episode}_eval.mp4')
+			# x = input("Press Enter to continue... after reseting env")
+			print("Press Enter to continue... after reseting env. To rate prev episode, press 0 for failure and 1 for success")
+			self.proceed_after_env_reset_event.clear() # clear the event flag
+			self.proceed_after_env_reset_event.wait() # blocking wait for the event flag to be set
+			if self.global_episode > 0:
+				self.logger.log_metrics({'num_success': self.num_episode_successes}, self.global_step, 'eval', episode=self.global_episode)
+				self.logger.log_metrics({'success_rate': self.num_episode_successes/self.global_episode}, self.global_step, 'eval', episode=self.global_episode)
+
+				# log confidence intervals for success rate
+				k = self.num_episode_successes # number of successes
+				n = self.global_episode # number of trials
+
+				table_columns = []
+				table_data = []
+				for alpha in self.alpha_range:
+					lb = bc.binom_ci(k, n, alpha, 'lb')
+					ub = bc.binom_ci(k, n, alpha, 'ub')
+
+					self.logger.log_metrics({f'success_rate_lb_{alpha}': lb}, self.global_step, 'eval', episode=self.global_episode)
+					self.logger.log_metrics({f'success_rate_ub_{alpha}': ub}, self.global_step, 'eval', episode=self.global_episode)
+
+			time_step = self.eval_env.reset()
+			# debug_info_dict = self.eval_env.debug_info_dict
+			# if replay_thread is not None:
+			# 	# wait for the last replay thread to finish
+			# 	replay_thread.join()
+
+			# self.eval_replay_storage.add(time_step._replace(observation=time_step.observation[self.cfg.obs_type]), debug_info_dict)
+			# replay_thread = threading.Thread(target=self.eval_replay_storage.add, args=(time_step._replace(observation=time_step.observation[self.cfg.obs_type]), debug_info_dict))
+			# replay_thread = threading.Thread(target=self.eval_replay_storage.add, args=(time_step, debug_info_dict))
+
+			# replay_thread.start()
+			if self.cfg.random_start:
+				time_step = self.reset(self.global_episode)
+				time.sleep(2) #5)
+			# if 'vinn' in repr(self.agent):
+			# 	self.agent.reset()
+			# 	# self.agent.buffer.reset()
+			# 	# if self.cfg.open_loop:
+			# 	# 	self.agent.current_step = 0
+			# if 'openloop' in repr(self.agent):
+			# 	self.agent.curr_step = 0
+			# at start of each episode, provide zero action for policies that use action history
+			# shape should be (T_o, T_a, action_dim)
+
+			# while not time_step.last() and not self.preempt_episode:
+			self.video_recorder.ros_start_recording()
+			self.eval_replay_storage.start_episode()
+			self.eval_env.start_policy_timer()
+			while not self.eval_env.episode_done() and not self.preempt_episode:
+				# with torch.no_grad(), utils.eval_mode(self.agent):
+				# 	# if self.cfg.agent.provide_topk:
+				# 	# 	action, vinn_action, topk = self.agent.act(
+				# 	# 		time_step.observation['pixels'],
+				# 	# 		self.global_step,
+				# 	# 		eval_mode=True)
+				# 	# elif self.cfg.agent.provide_obs:
+				# 	# 	action, vinn_action, obs = self.agent.act(
+				# 	# 		time_step.observation['pixels'],
+				# 	# 		self.global_step,
+				# 	# 		eval_mode=True)
+				# 	# else:
+				# 	action, vinn_action = self.agent.act(
+				# 		time_step.observation,
+				# 		self.global_step,
+				# 		eval_mode=True,
+				# 		obs_timestamp=time_step.observation['timestamp'],
+				# 		obs_seq=time_step.observation['seq'],
+				# 		action_history=action_history,
+				# 		action_history_start_timestamp=action_history_start_timestamp,
+				# 		)
+					# DONT WAIT FOR POLICY TO GET AN ACTION 
+					# we dont want to slow down grabbing obs and passing to sam/contact features
+
+				self.eval_env.run_policy_threads() # this just does a rospy sleep
+					
+				# if self.use_action_history:
+				# 	action_history_start_timestamp = time_step.observation['timestamp']
+				# 	# action_history = action[:self.cfg.agent.config.policy_cfg.action_history_encoder_config.history_length, ...]		
+				# 	# add n_obs_steps dimension to action_history, for now we assume n_obs_steps = 1
+				# 	# TODO: handle n_obs_steps > 1
+				# 	action_history = action[np.newaxis, ...]
+
+				# time_step = self.eval_env.step(action, vinn_action) # obs, reward after action has been taken
+				# debug_info_dict = self.eval_env.debug_info_dict
+				
+				# time_step = self.eval_env.ros_step()
+				
+				# replay_thread.join()
+
+				# time how long it takes to execute the step
+				# time_before_add = time.perf_counter()
+				# self.eval_replay_storage.add(time_step._replace(observation=time_step.observation[self.cfg.obs_type]), debug_info_dict)
+				# use thread to call the add function in a separate thread
+				# replay_thread = threading.Thread(target=self.eval_replay_storage.add, args=(time_step._replace(observation=time_step.observation[self.cfg.obs_type]), debug_info_dict))
+
+				# replay_thread = threading.Thread(target=self.eval_replay_storage.add, args=(time_step, debug_info_dict))
+				# replay_thread.start()
+				
+				# print(f"Time to add to replay buffer: {time.perf_counter() - time_before_add}")
+
+				# self.video_recorder.record(self.eval_env)
+				# self._global_step += 1
+			
+			self.eval_env.stop_policy_timer()
+			
+			if self.restart_episode:
+				# means we should delete the current episode and start again
+				self.restart_episode = False
+				self.eval_replay_storage.reset_current_episode()
+				self.video_recorder.reset_current_episode()
+
+			else:
+				self.eval_replay_storage.store_current_episode()
+				video_filepath = self.video_recorder.save()
+				self.logger.log_video(f"eval/{video_filepath.name.rstrip('.mp4')}", video_filepath, self.global_step)
+				self._global_episode += 1
+			
+			self.preempt_episode = False # reset preempt_episode flag
+		
+		# self.video_recorder.save(f'{episode}_eval.mp4')
+		# get the video file and convert to video tensor to log 
+
+		self.eval_env.reset()
+		
+		print("Evaluation finished. To wrap up, rate prev episode, press 0 for failure and 1 for success")
+		self.proceed_after_env_reset_event.clear() # clear the event flag
+		self.proceed_after_env_reset_event.wait() # blocking wait for the event flag to be set
+		if self.global_episode > 0:
+			# self.logger.log_metrics({'num_success': self.num_episode_successes}, self.global_step, 'eval', episode=self.global_episode)
+			self.logger.log_metrics({'num_success': self.num_episode_successes}, self.global_step, 'eval', episode=self.global_episode)
+			self.logger.log_metrics({'success_rate': self.num_episode_successes/self.global_episode}, self.global_step, 'eval', episode=self.global_episode)
+			
+			# log confidence intervals for success rate
+			k = self.num_episode_successes # number of successes
+			n = self.global_episode # number of trials
+
+			table_columns = ['success_rate']
+			table_data = [self.num_episode_successes/self.global_episode]
+			for alpha in self.alpha_range:
+				lb = bc.binom_ci(k, n, alpha, 'lb')
+				ub = bc.binom_ci(k, n, alpha, 'ub')
+
+				self.logger.log_metrics({f'success_rate_lb_{alpha}': lb}, self.global_step, 'eval', episode=self.global_episode)
+				self.logger.log_metrics({f'success_rate_ub_{alpha}': ub}, self.global_step, 'eval', episode=self.global_episode)
+
+				table_columns.extend([f'success_rate_lb_{alpha}', f'success_rate_ub_{alpha}'])
+				table_data.extend([lb, ub])
+
+			table_data = [table_data]
+
+			# seperately log as a table
+			wandb.log({
+				"eval/success_rate_ci": wandb.Table(data=table_data, columns=table_columns)
+			})
+
+			# also accumulate eval metrics across previous eval runs 
+			# TODO: change wandb init to resume from an existing run!!!
+			run_filter={
+				"jobType": "eval",
+				"config.wandb_run_id": self.cfg.wandb_run_id, 
+				"summary_metrics.episode": {"$gte": 5},
+				"config.checkpoint_epoch": self.cfg.checkpoint_epoch,
+				"state": "finished",
+				# "config.grasped_obj_name": grasped_obj_name,
+				# "config.left_book_slot": left_book_slot,
+			}
+
+			api = wandb.Api()
+			filtered_runs = api.runs("serialexperimentsleon/extrinsic_contact_downstream", filters=run_filter)
+			total_num_successes = self.num_episode_successes
+			total_num_episodes = self.global_episode
+			list_of_historical_run_ids = []
+			if len(filtered_runs) > 0:
+				for filtered_run in filtered_runs:
+					total_num_successes += filtered_run.summary_metrics['eval/num_success']
+					# total_num_episodes += filtered_run.summary_metrics['episode']
+					total_num_episodes += filtered_run.config['num_eval']
+					list_of_historical_run_ids.append(filtered_run.id)
+
+			wandb.summary['total_num_successes'] = total_num_successes
+			wandb.summary['total_num_episodes'] = total_num_episodes
+			wandb.summary['total_success_rate'] = total_num_successes/total_num_episodes
+				
+			# log the accumulated metrics as a table
+			total_table_columns = ['total_num_successes', 'total_num_episodes', 'total_success_rate']
+			total_table_data = [total_num_successes, total_num_episodes, total_num_successes/total_num_episodes]
+			self.logger.log_metrics({'total_success_rate': total_num_successes/total_num_episodes}, self.global_step, 'eval', episode=total_num_episodes)
+			
+			for alpha in self.alpha_range:
+				lb = bc.binom_ci(total_num_successes, total_num_episodes, alpha, 'lb')
+				ub = bc.binom_ci(total_num_successes, total_num_episodes, alpha, 'ub')
+				total_table_columns.extend([f'total_success_rate_lb_{alpha}', f'total_success_rate_ub_{alpha}'])
+				total_table_data.extend([lb, ub])
+				wandb.summary[f'total_success_rate_lb_{alpha}'] = lb
+				wandb.summary[f'total_success_rate_ub_{alpha}'] = ub
+
+				self.logger.log_metrics({f'total_success_rate_lb_{alpha}': lb}, self.global_step, 'eval', episode=total_num_episodes)
+				self.logger.log_metrics({f'total_success_rate_ub_{alpha}': ub}, self.global_step, 'eval', episode=total_num_episodes)
+
+			total_table_data = [total_table_data]
+			wandb.log({
+				'eval/total_success_rate_ci': wandb.Table(data=total_table_data, columns=total_table_columns)
+			})
+
+		self.continue_keypress_thread = False # will stop the keypress thread
+		self.keypress_input_thread.join() # wait for the keypress thread to finish
+
+	def load_checkpoint_conf(self, snapshot_path):
+		config_path = snapshot_path.parent / 'config.yaml'
+		if not config_path.exists():
+			raise FileNotFoundError(f'No snapshot conf found at {config_path}')
+		else:
+			# load the omegaconf config
+			hydra.core.global_hydra.GlobalHydra.instance().clear()
+			hydra.initialize(
+				str(_relative_path_between(Path(config_path).absolute().parent, Path(__file__).absolute().parent)),
+			)
+			cfg = hydra.compose(Path(config_path).stem)
+			from deepdiff import DeepDiff
+			from omegaconf import open_dict
+			diff = DeepDiff(OmegaConf.to_container(cfg), OmegaConf.to_container(self.cfg)) # old, new
+			# import re
+			overwriteable_keys = [f"root{overwritable_key}" for overwritable_key in ["['use_wandb']", "['path_to_depth_extrinsics']", "['eval']", "['root_dir']", "['wandb_notes']", "['agent']['config']['train_cfg']['use_amp']", "['agent']['config']['compile']", "['agent']['config']['policy_cfg']['num_inference_steps']"]]
+			if "values_changed" in diff:
+				# top_k_checkpoints, wandb_notes, agent.config.train_cfg.use_amp, save_snapshot_every_epochs_diffusion, check_topk_every_epochs_diffusion, validate_diffusion_on_action_loss_every_epochs, train_eval_diffusion_on_action_loss_every_epochs, validate_every_epochs_diffusion
+				# for keys above, overwrite the old config with the new config
+				for k, v in diff['values_changed'].items():
+					# replace any keys that are under "root['suite']"
+					if k in overwriteable_keys or k.startswith("root['suite']"):
+						print(f"Found changed key {k} with value {v}. Overwriting old checkpoint config")
+						if k == "root['agent']['config']['compile']":
+							if diff['values_changed'][k]['new_value']:
+								self.loading_uncompiled_checkpoint_with_compile = True
+							elif not diff['values_changed'][k]['new_value']:
+								# raise ValueError("Cannot load a compiled checkpoint without compile")
+								self.loading_compiled_checkpoint_with_no_compile = True
+						exec(f"{k.replace('root[', 'cfg[')} = {k.replace('root[', 'self.cfg[')}")
+			# for any new values, update the old checkpoint config
+			if "dictionary_item_added" in diff:
+				for new_key in diff['dictionary_item_added']: # this is a list
+					# if new_key == "root['suite']['task_make_fn']['observation_cfg']":
+					if new_key == "root['suite']['task_make_fn']['agent_policy_cfg']":
+						# pass the agents observation_cfg to the suite task_make_fn
+						with open_dict(cfg): # to allow addition of non-existing keys
+							# cfg.suite.task_make_fn.observation_cfg = cfg.agent.config.observation_cfg
+							cfg.suite.task_make_fn.agent_policy_cfg = cfg.agent.config
+						continue
+					elif "['agent']['config']['policy_cfg']['input_shapes']" in new_key:
+						# skip adding the new key if it is the input_shapes of the policy_cfg
+						continue
+					else:
+						print(f"Found new key {new_key} with value {eval(new_key.replace('root[', 'self.cfg['))}. Adding to checkpoint config")
+					# eval(new_key.replace('root', 'cfg')) = eval(new_key.replace('root', 'self.cfg'))
+					if new_key == "root['agent']['config']['compile']":
+						if self.cfg.agent.config.compile:
+							self.loading_uncompiled_checkpoint_with_compile = True
+					
+					with open_dict(cfg):
+						exec(f"{new_key.replace('root[', 'cfg[')}={new_key.replace('root[', 'self.cfg[')}")
+			self.cfg = cfg
+
+	def load_checkpoint(self, snapshot_path, bc=False):
+		print(f'resuming {repr(self.agent)}: {snapshot_path}')
+		with snapshot_path.open('rb') as f:
+			payload = torch.load(f)
+		agent_payload = {}
+		for k, v in payload.items():
+			if k not in self.__dict__:
+				agent_payload[k] = v
+			elif k == '_global_epoch':
+				self._global_epoch = v
+				print(f'loaded epoch: {v}')
+				if self.cfg.use_wandb:
+					# add to config of wandb
+					wandb.config.update({'epoch': v})
+
+		# self.agent.load_snapshot_eval(agent_payload, bc)
+
+@hydra.main(config_path='cfgs', config_name='config_eval')
+def main(cfg):
+	from eval_robot import Workspace as W
+	root_dir = Path.cwd()
+	workspace = W(cfg)
+
+	workspace.eval()
+
+if __name__ == '__main__':
+	main()

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/table/eval/success_rate_ci_20_97faedb494e69b73017e.table.json

@@ -0,0 +1 @@
+{"columns": ["success_rate", "success_rate_lb_0.01", "success_rate_ub_0.01", "success_rate_lb_0.025", "success_rate_ub_0.025", "success_rate_lb_0.05", "success_rate_ub_0.05", "success_rate_lb_0.1", "success_rate_ub_0.1"], "data": [[0.3, 0.1003983021581674, 0.558511674392442, 0.1215232685965274, 0.502513885497544, 0.16696532077297443, 0.48316777497866625, 0.17897007993323794, 0.4645809233259468]]}

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/table/eval/total_success_rate_ci_21_54cfd2c4c0340bc54565.table.json

@@ -0,0 +1 @@
+{"columns": ["total_num_successes", "total_num_episodes", "total_success_rate", "total_success_rate_lb_0.01", "total_success_rate_ub_0.01", "total_success_rate_lb_0.025", "total_success_rate_ub_0.025", "total_success_rate_lb_0.05", "total_success_rate_ub_0.05", "total_success_rate_lb_0.1", "total_success_rate_ub_0.1"], "data": [[41, 105, 0.3904761904761905, 0.286629945082423, 0.5046236664047531, 0.30276687447201045, 0.4824152737891081, 0.3192752711836344, 0.46708489955129906, 0.3268393464732785, 0.4542280621918193]]}

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/0_eval_0_ae65028e83f44b9ac5b1.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae65028e83f44b9ac5b1292e9a19a482040bef9b85dc8452c1003e498adcf200
+size 1203236

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/13_eval_13_49e329d2ff2e935251c1.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49e329d2ff2e935251c138c38bd869e83a370b6a20c29a6f3f6f4fb5241a5926
+size 1214289

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/14_eval_14_b1f7698aaf2c3d6e6abf.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1f7698aaf2c3d6e6abf7855ee3e636573dce1a6d4641acb7f13c8ce0412b52c
+size 1410819

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/15_eval_15_b314c0181db144f9c4bb.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b314c0181db144f9c4bbd755580f55f4fa904dbd7bf96ae9ebbdf5aa6aea7161
+size 1504211

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/16_eval_16_a93b394f6e07a47f91ce.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a93b394f6e07a47f91ce5c6436e2ac6f677aa64e46d4e4c1331cad98e6429f67
+size 1441574

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/17_eval_17_e673279ceb422c546539.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e673279ceb422c54653970040c48bf551dbc8cea9639629ae5c5f872e26f2b86
+size 1438873

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/18_eval_18_fd5b9c7ce55fe4462130.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd5b9c7ce55fe446213004b70d71e620113f558086949526d140680e6e08433b
+size 1143394

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/2_eval_2_537ec4e4ec2357120fef.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:537ec4e4ec2357120fef5cc7f4edbf14515d4b2e6bdfab19c7eeeacbe708bc89
+size 1203819

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/3_eval_3_5243c77fb64415fcde38.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5243c77fb64415fcde38b5e587dfe83a26bc73991f69876d74ed1cbc67f2c743
+size 1150313

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/4_eval_4_123736b8ecd81d086c47.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:123736b8ecd81d086c472194678ae8960360473b3f54800e2d9f897b23c89739
+size 1255949

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/6_eval_6_718698df80de1dc3b1e7.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:718698df80de1dc3b1e7aea8932c6797bc8d2289782b524986f54f2abbffc39c
+size 1163347

4fmihp69/wandb/run-20250107_220610-4fmihp69/files/media/videos/eval/7_eval_7_0c0abd0712ae602c07c3.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c0abd0712ae602c07c33b31081a76a7241f86938b9d37db8c91daabb3f3f779
+size 1201147

4fmihp69/wandb/run-20250107_220610-4fmihp69/logs/debug.log

@@ -0,0 +1,31 @@
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Current SDK version is 0.17.5
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Configure stats pid to 2240043
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Loading settings from /home/leonmkim/.config/wandb/settings
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Loading settings from /home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4fmihp69/wandb/settings
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Loading settings from environment variables: {}
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Applying setup settings: {'_disable_service': False}
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Inferring run settings from compute environment: {'program_relpath': 'FISH/eval_robot.py', 'program_abspath': '/home/leonmkim/fish_leon/FISH/eval_robot.py', 'program': '/home/leonmkim/fish_leon/FISH/eval_robot.py'}
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_setup.py:_flush():76] Applying login settings: {}
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_init.py:_log_setup():529] Logging user logs to /home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4fmihp69/wandb/run-20250107_220610-4fmihp69/logs/debug.log
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_init.py:_log_setup():530] Logging internal logs to /home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4fmihp69/wandb/run-20250107_220610-4fmihp69/logs/debug-internal.log
+2025-01-07 22:06:10,448 INFO    MainThread:2240043 [wandb_init.py:init():569] calling init triggers
+2025-01-07 22:06:10,449 INFO    MainThread:2240043 [wandb_init.py:init():576] wandb.init called with sweep_config: {}
+config: {'root_dir': '/home/leonmkim/fish_leon', 'replay_buffer_size': 150000, 'replay_buffer_num_workers': 2, 'nstep': 3, 'batch_size': 128, 'seed': 0, 'dataset_shuffle_seed': 0, 'device': 'cuda', 'save_video': True, 'save_train_video': True, 'use_tb': True, 'use_wandb': True, 'wandb_run_id': '1002_0', 'wandb_notes': '1002_0_req_1365_0', 'eval': True, 'true_action_history': False, 'train_pad_after': 4, 'process_contact_features': True, 'obs_type': 'pixels', 'use_color': True, 'use_depth': True, 'use_masks': True, 'mask_list': ['EE_obj_mask'], 'mask_representation': 'channels', 'crop_hw': [144, 144], 'crop_down_offset': 48, 'color_crop_type': None, 'depth_crop_type': None, 'segmask_crop_type': None, 'add_crop_binary_mask': False, 'add_coord_conv_map': False, 'use_context_color': False, 'use_context_depth': False, 'use_context_segmask': False, 'context_color_crop_type': None, 'context_depth_crop_type': None, 'context_segmask_crop_type': None, 'context_add_crop_binary_mask': False, 'context_add_coord_conv_map': False, 'use_contact_map': False, 'use_sdf_maps': False, 'use_normals_maps': False, 'which_objects': 'both', 'max_contact_prob': 0.1, 'max_depth': 2.0, 'grasped_dtc_max_value': 0.2, 'env_dtc_max_value': 0.4, 'grasped_normals_mask_max_dtc_value': 0.2, 'env_normals_mask_max_dtc_value': 0.4, 'clamp_dtc': True, 'dtc_adaptive_normalization': False, 'mask_normals_within_sdf': True, 'adaptive_normals_mask': True, 'learnable_contact_preprocess_params': True, 'contact_model_name': 'local_multitask_outhd64all_home_crop_h144w144d48_mask_ctxtmask_seed_220979_epoch_9', 'contact_estimation_model_ckpt_path': '~/fish_leon/contact_estimation/artifacts/175604_2/checkpoints/epoch=09-val_loss=0.00.ckpt', 'encoder_type': 'small', 'debug_timestamps': False, 'open_loop': False, 'action_trajectories': True, 'stop_after_action': False, 'interpolation_frequency': 25, 'policy_frequency': 5, 'wait_for_new_camera_frames': True, 'baseline': False, 'train_demo_idxs_list_or_num': -1, 'log_train_every_steps': 25, 'name_of_expert_demo': '120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act', 'expert_dataset_dirpath': '/home/leonmkim/fish_leon/FISH/expert_demos/frankagym/FrankaInsertion-v1/120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act', 'store_dataset_in_memory': False, 'expert_dataset': '/home/leonmkim/fish_leon/FISH/expert_demos/frankagym/FrankaInsertion-v1/120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act/demos.zarr', 'action_key': 'action_trajectory_25hz', 'semantic_demo_grouping_name': 'semantic_demo_grouping.yaml', 'semantic_demo_grouping': '/home/leonmkim/fish_leon/FISH/expert_demos/frankagym/FrankaInsertion-v1/120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act/semantic_demo_grouping.yaml', 'include_groups_list': 'all', 'expert_dataset_config': '/home/leonmkim/fish_leon/FISH/expert_demos/frankagym/FrankaInsertion-v1/120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act/demo_config.yaml', 'name_of_valid_demo': '120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act', 'valid_dataset_dir': '/home/leonmkim/fish_leon/FISH/expert_demos/frankagym/FrankaInsertion-v1/120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act/demos.zarr', 'valid_demo_idxs_list_or_num': None, 'val_num_groups': 5, 'load_bc': True, 'checkpoint_epoch_list': [99, 199, 299, 399, 499, 599, 699, 799, 899, 999, 1249, 1499, 1749, 1999, 2999, 3999, 4999, 5999, 6999, 7999, 8999, 9999], 'snapshot_root_dir': '/mnt/grasp_high_usage/leonmkim/contact_estimation/FISH', 'save_snapshot': True, 'save_last_snapshot': True, 'save_snapshot_when_done': True, 'top_k_checkpoints': 5, 'save_snapshot_link_to_weights_dir': 'deprecated', 'bc_regularize': False, 'bc_weight_type': 'qfilter', 'experiment_dir': './exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0', 'agent': {'_target_': 'agent.diffusion_policy.DiffusionPolicyAgent', 'name': 'diffusion_policy', 'load_checkpoint': True, 'device': 'cuda', 'n_obs_steps': 1, 'suite_name': 'frankagym', 'obs_type': 'pixels', 'enable_arm': True, 'enable_camera': True, 'use_tb': True, 'desired_image_shape': [13, 180, 240], 'orig_cam_shape': [3, 240, 320], 'config': {'_target_': 'agent.diffusion_policy.DiffusionPolicyAgentConfig', 'compile': False, 'device': 'cuda', 'cam_resize_shape': [13, 180, 240], 'orig_cam_shape': [3, 240, 320], 'policy_frequency': 5, 'interpolation_frequency': 25, 'policy_cfg': {'_target_': 'lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig', 'n_obs_steps': 1, 'horizon': 36, 'n_action_steps': 36, 'output_shapes': {'action': [7]}, 'input_normalization_modes': {'observation.image': 'mean_std', 'observation.state': 'min_max', 'observation.action_history': 'min_max'}, 'output_normalization_modes': {'action': 'min_max'}, 'vision_backbone': 'resnet18', 'pretrained_backbone_weights': None, 'transforms': [{'_target_': 'torchaug.transforms.RandomAffine', 'degrees': [-5, 5], 'translate': [0.05, 0.05], 'batch_transform': True, 'num_chunks': -1, 'batch_inplace': True}, {'_target_': 'torchaug.transforms.RandomColorJitter', 'brightness': 0.3, 'contrast': 0.4, 'saturation': 0.5, 'hue': 0.08, 'batch_transform': True, 'num_chunks': -1, 'batch_inplace': True}], 'use_group_norm': True, 'spatial_softmax_num_keypoints': 32, 'action_history_encoder_config': {'_target_': 'lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig', 'in_channels': 7, 'out_channels': 32, 'history_length': 6, 'kernel_size': 5, 'downsample_kernel_size': 3, 'downsample_stride': 2, 'downsample_padding': 1}, 'down_dims': [256, 512, 1024], 'kernel_size': 5, 'n_groups': 8, 'diffusion_step_embed_dim': 128, 'use_film_scale_modulation': True, 'noise_scheduler_type': 'DDIM', 'beta_schedule': 'squaredcos_cap_v2', 'beta_start': 0.0001, 'beta_end': 0.02, 'prediction_type': 'epsilon', 'clip_sample': True, 'clip_sample_range': 1.0, 'num_train_timesteps': 50, 'num_inference_steps': 10, 'do_mask_loss_for_padding': False, 'input_shapes': {'observation.image': [13, 180, 240], 'context_observation.image': [13, 180, 240], 'observation.state': [8], 'observation.action_history': [7]}}, 'train_cfg': {'_target_': 'utils.TrainConfig', 'lr': 0.0001, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 500, 'adam_betas': [0.95, 0.999], 'adam_eps': 1e-08, 'adam_weight_decay': 1e-06, 'grad_clip_norm': 10, 'offline_steps': 1000000, 'use_amp': True}, 'observation_cfg': {'_target_': 'agent.encoder.VisualFeatureSet', 'use_depth': True, 'use_color': True, 'mask_input_dict': {'_target_': 'agent.encoder.MaskInputDict', 'enable': True, 'representation': 'channels', 'mask_list': ['EE_obj_mask']}, 'crop_input_config': {'_target_': 'agent.encoder.CropInputConfig', 'color_crop_type': None, 'depth_crop_type': None, 'segmask_crop_type': None, 'crop_hw': [144, 144], 'crop_down_offset': 48, 'add_crop_binary_mask': False, 'add_coord_conv_map': False}, 'context_input_config': {'_target_': 'agent.encoder.ContextInputConfig', 'use_color': False, 'use_depth': False, 'mask_input_dict': {'_target_': 'agent.encoder.MaskInputDict', 'enable': False, 'representation': 'channels', 'mask_list': ['EE_obj_mask']}, 'crop_input_config': {'_target_': 'agent.encoder.CropInputConfig', 'color_crop_type': None, 'depth_crop_type': None, 'segmask_crop_type': None, 'crop_hw': [144, 144], 'crop_down_offset': 48, 'add_crop_binary_mask': False, 'add_coord_conv_map': False}}, 'mask_soft_approx_scheduler_config': {'_target_': 'agent.encoder.MaskSoftApproxSchedulerConfig', 'num_steps': 40000, 'initial_value': 10.0, 'final_value': 1000.0, 'interpolation_scheme': 'cosine'}, 'use_contact_map': False, 'use_sdf_maps': False, 'use_normals_maps': False, 'which_objects': 'both', 'grasped_dtc_max_value': 0.2, 'env_dtc_max_value': 0.4, 'grasped_normals_mask_max_dtc_value': 0.2, 'env_normals_mask_max_dtc_value': 0.4, 'clamp_dtc': True, 'max_contact_prob': 0.1, 'mask_normals_within_sdf': True, 'dtc_adaptive_normalization': False, 'adaptive_normals_mask': True, 'max_depth': 2.0, 'image_shape': [13, 180, 240], 'learnable_contact_preprocess_params': True, 'learning_rate': 0.0001, 'weight_decay': 0.0, 'contact_model_name': 'local_multitask_outhd64all_home_crop_h144w144d48_mask_ctxtmask_seed_220979_epoch_9', 'zero_centered': False}}}, 'suite': {'suite': 'frankagym', 'name': 'frankagym', 'frame_stack': 1, 'action_repeat': 1, 'discount': 0.99, 'hidden_dim': 1024, 'num_train_frames': 2010, 'num_seed_frames': 260, 'num_train_epochs': 5000, 'validate_every_epochs': 100, 'validate_diffusion_on_action_loss_every_epochs': 500, 'train_eval_diffusion_on_action_loss_every_epochs': 500, 'check_topk_every_epochs': 10, 'save_snapshot_every_epochs': 5000, 'eval_every_frames': 2000, 'num_eval_episodes': 5, 'save_snapshot': True, 'wait_for_user_to_start_episode': True, 'task_make_fn': {'_target_': 'suite.frankagym.make', 'name': 'FrankaInsertion-v1', 'height': 240, 'width': 320, 'frame_stack': 1, 'action_repeat': 1, 'seed': 0, 'enable_arm': True, 'enable_gripper': True, 'start_with_gripper_open': True, 'enable_camera': True, 'path_to_depth_extrinsics': '~/fish_leon/FISH/cfgs/camera_poses/camera_poses_L515/20240904-122305/color_tf_world.npy', 'contact_estimation_model_ckpt_path': '~/fish_leon/contact_estimation/artifacts/175604_2/checkpoints/epoch=09-val_loss=0.00.ckpt', 'x_limit': [0.2, 0.7], 'y_limit': [-0.4, 0.4], 'z_limit': [-0.05, 0.55], 'device': 'cuda', 'interpolation_frequency': 25, 'policy_frequency': 5, 'debug_timestamps': False, 'stop_after_action': False, 'open_loop': False, 'wait_for_new_camera_frames': True, 'action_key': 'action_trajectory_25hz', 'action_trajectory_horizon': 36, 'action_trajectories': True, 'path_to_zarr_dataset': '/home/leonmkim/fish_leon/FISH/expert_demos/frankagym/FrankaInsertion-v1/120_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act/demos.zarr', 'agent_policy_cfg': {'_target_': 'agent.diffusion_policy.DiffusionPolicyAgentConfig', 'compile': False, 'device': 'cuda', 'cam_resize_shape': [13, 180, 240], 'orig_cam_shape': [3, 240, 320], 'policy_frequency': 5, 'interpolation_frequency': 25, 'policy_cfg': {'_target_': 'lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig', 'n_obs_steps': 1, 'horizon': 36, 'n_action_steps': 36, 'output_shapes': {'action': [7]}, 'input_normalization_modes': {'observation.image': 'mean_std', 'observation.state': 'min_max', 'observation.action_history': 'min_max'}, 'output_normalization_modes': {'action': 'min_max'}, 'vision_backbone': 'resnet18', 'pretrained_backbone_weights': None, 'transforms': [{'_target_': 'torchaug.transforms.RandomAffine', 'degrees': [-5, 5], 'translate': [0.05, 0.05], 'batch_transform': True, 'num_chunks': -1, 'batch_inplace': True}, {'_target_': 'torchaug.transforms.RandomColorJitter', 'brightness': 0.3, 'contrast': 0.4, 'saturation': 0.5, 'hue': 0.08, 'batch_transform': True, 'num_chunks': -1, 'batch_inplace': True}], 'use_group_norm': True, 'spatial_softmax_num_keypoints': 32, 'action_history_encoder_config': {'_target_': 'lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig', 'in_channels': 7, 'out_channels': 32, 'history_length': 6, 'kernel_size': 5, 'downsample_kernel_size': 3, 'downsample_stride': 2, 'downsample_padding': 1}, 'down_dims': [256, 512, 1024], 'kernel_size': 5, 'n_groups': 8, 'diffusion_step_embed_dim': 128, 'use_film_scale_modulation': True, 'noise_scheduler_type': 'DDIM', 'beta_schedule': 'squaredcos_cap_v2', 'beta_start': 0.0001, 'beta_end': 0.02, 'prediction_type': 'epsilon', 'clip_sample': True, 'clip_sample_range': 1.0, 'num_train_timesteps': 50, 'num_inference_steps': 10, 'do_mask_loss_for_padding': False, 'input_shapes': {'observation.image': [13, 180, 240], 'context_observation.image': [13, 180, 240], 'observation.state': [8], 'observation.action_history': [7]}}, 'train_cfg': {'_target_': 'utils.TrainConfig', 'lr': 0.0001, 'lr_scheduler': 'cosine', 'lr_warmup_steps': 500, 'adam_betas': [0.95, 0.999], 'adam_eps': 1e-08, 'adam_weight_decay': 1e-06, 'grad_clip_norm': 10, 'offline_steps': 1000000, 'use_amp': True}, 'observation_cfg': {'_target_': 'agent.encoder.VisualFeatureSet', 'use_depth': True, 'use_color': True, 'mask_input_dict': {'_target_': 'agent.encoder.MaskInputDict', 'enable': True, 'representation': 'channels', 'mask_list': ['EE_obj_mask']}, 'crop_input_config': {'_target_': 'agent.encoder.CropInputConfig', 'color_crop_type': None, 'depth_crop_type': None, 'segmask_crop_type': None, 'crop_hw': [144, 144], 'crop_down_offset': 48, 'add_crop_binary_mask': False, 'add_coord_conv_map': False}, 'context_input_config': {'_target_': 'agent.encoder.ContextInputConfig', 'use_color': False, 'use_depth': False, 'mask_input_dict': {'_target_': 'agent.encoder.MaskInputDict', 'enable': False, 'representation': 'channels', 'mask_list': ['EE_obj_mask']}, 'crop_input_config': {'_target_': 'agent.encoder.CropInputConfig', 'color_crop_type': None, 'depth_crop_type': None, 'segmask_crop_type': None, 'crop_hw': [144, 144], 'crop_down_offset': 48, 'add_crop_binary_mask': False, 'add_coord_conv_map': False}}, 'mask_soft_approx_scheduler_config': {'_target_': 'agent.encoder.MaskSoftApproxSchedulerConfig', 'num_steps': 40000, 'initial_value': 10.0, 'final_value': 1000.0, 'interpolation_scheme': 'cosine'}, 'use_contact_map': False, 'use_sdf_maps': False, 'use_normals_maps': False, 'which_objects': 'both', 'grasped_dtc_max_value': 0.2, 'env_dtc_max_value': 0.4, 'grasped_normals_mask_max_dtc_value': 0.2, 'env_normals_mask_max_dtc_value': 0.4, 'clamp_dtc': True, 'max_contact_prob': 0.1, 'mask_normals_within_sdf': True, 'dtc_adaptive_normalization': False, 'adaptive_normals_mask': True, 'max_depth': 2.0, 'image_shape': [13, 180, 240], 'learnable_contact_preprocess_params': True, 'learning_rate': 0.0001, 'weight_decay': 0.0, 'contact_model_name': 'local_multitask_outhd64all_home_crop_h144w144d48_mask_ctxtmask_seed_220979_epoch_9', 'zero_centered': False}}, 'true_action_history': False}}, 'num_train_frames_bc': 50000, 'num_train_frames_drq': 1100000, 'stddev_schedule_drq': 'linear(1.0,0.1,100000)', 'task_name': 'FrankaInsertion-v1', 'num_train_frames_vinn': 25000, 'num_train_frames_diffusion': 1000000, 'num_train_epochs_bc': 5000, 'num_train_epochs_diffusion': 15000, 'validate_every_epochs_bc': 5, 'validate_every_epochs_diffusion': 250, 'validate_diffusion_on_action_loss_every_epochs': 250, 'train_eval_diffusion_on_action_loss_every_epochs': 250, 'check_topk_every_epochs': 5, 'check_topk_every_epochs_diffusion': 250, 'save_snapshot_every_epochs_diffusion': 1500, 'x_limit': [0.2, 0.7], 'y_limit': [-0.4, 0.4], 'z_limit': [-0.05, 0.55], 'home_displacement': [0.55, 0.0, 0.55, 180.0, 0.0, 0.0], 'enable_gripper': True, 'start_with_gripper_open': True, 'offset_mask': [1, 1, 1, 1, 1, 1], 'path_to_depth_extrinsics': '~/fish_leon/FISH/cfgs/camera_poses/camera_poses_L515/20240904-122305/color_tf_world.npy', 'feature_type': '180x240_1_RGB_D_2.0_msk_channels_EE_obj_mask_acthst_hst6_out32_dwnkrnl3_dwnstrd2_dwnpd1', 'save_buffer': True, 'num_eval': 20, 'random_start': False, 'eval_starts': '/home/leonmkim/fish_leon/FISH/eval_starts/frankagym_pixels/FrankaInsertion-v1', 'num_valid_demos': None, 'load_checkpoint': True, 'checkpoint_epoch': 12000, 'load_residual_weight': False, 'checkpoint_root_dir': '/home/leonmkim/fish_leon/FISH', 'checkpoint_weight_dir': '/home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0', 'residual_weight': '/home/leonmkim/fish_leon/FISH/weights/frankagym_pixels/FrankaInsertion-v1/weight.pt', 'final_experiment_dir': './exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4fmihp69'}
+2025-01-07 22:06:10,449 INFO    MainThread:2240043 [wandb_init.py:init():619] starting backend
+2025-01-07 22:06:10,449 INFO    MainThread:2240043 [wandb_init.py:init():623] setting up manager
+2025-01-07 22:06:10,456 INFO    MainThread:2240043 [backend.py:_multiprocessing_setup():105] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
+2025-01-07 22:06:10,457 INFO    MainThread:2240043 [wandb_init.py:init():631] backend started and connected
+2025-01-07 22:06:10,472 INFO    MainThread:2240043 [wandb_init.py:init():720] updated telemetry
+2025-01-07 22:06:10,479 INFO    MainThread:2240043 [wandb_init.py:init():753] communicating run to backend with 90.0 second timeout
+2025-01-07 22:06:10,762 INFO    MainThread:2240043 [wandb_run.py:_on_init():2435] communicating current version
+2025-01-07 22:06:10,803 INFO    MainThread:2240043 [wandb_run.py:_on_init():2444] got version response upgrade_message: "wandb version 0.19.2 is available!  To upgrade, please run:\n $ pip install wandb --upgrade"
+
+2025-01-07 22:06:10,803 INFO    MainThread:2240043 [wandb_init.py:init():804] starting run threads in backend
+2025-01-07 22:06:11,143 INFO    MainThread:2240043 [wandb_run.py:_console_start():2413] atexit reg
+2025-01-07 22:06:11,144 INFO    MainThread:2240043 [wandb_run.py:_redirect():2255] redirect: wrap_raw
+2025-01-07 22:06:11,144 INFO    MainThread:2240043 [wandb_run.py:_redirect():2320] Wrapping output streams.
+2025-01-07 22:06:11,144 INFO    MainThread:2240043 [wandb_run.py:_redirect():2345] Redirects installed.
+2025-01-07 22:06:11,146 INFO    MainThread:2240043 [wandb_init.py:init():847] run started, returning control to user process
+2025-01-07 22:06:11,146 INFO    MainThread:2240043 [wandb_run.py:_tensorboard_callback():1544] tensorboard callback: /home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4fmihp69/tb, True
+2025-01-07 22:06:16,468 INFO    MainThread:2240043 [wandb_run.py:_config_callback():1382] config_cb None None {'grasped_obj_name': 'greece', 'left_book_slot': 'twodim'}
+2025-01-07 22:24:00,006 WARNING MsgRouterThr:2240043 [router.py:message_loop():77] message_loop has been closed

4r9mcmxe/.hydra/config.yaml

@@ -0,0 +1,350 @@
+root_dir: /home/${oc.env:USER}/fish_leon
+nstep: 3
+seed: 41
+dataset_shuffle_seed: ${seed}
+device: cuda
+save_video: true
+save_buffer: true
+use_tb: true
+baseline: false
+use_wandb: true
+eval: true
+process_contact_features: ${eval}
+obs_type: pixels
+use_color: true
+use_depth: true
+use_masks: false
+mask_list:
+- EE_obj_mask
+mask_representation: channels
+crop_hw:
+- 144
+- 144
+crop_down_offset: 48
+color_crop_type: null
+depth_crop_type: null
+segmask_crop_type: null
+add_crop_binary_mask: false
+add_coord_conv_map: false
+use_context_color: false
+use_context_depth: false
+use_context_segmask: false
+context_color_crop_type: null
+context_depth_crop_type: null
+context_segmask_crop_type: null
+context_add_crop_binary_mask: false
+context_add_coord_conv_map: false
+use_contact_map: false
+use_sdf_maps: false
+use_normals_maps: false
+which_objects: both
+max_contact_prob: 0.1
+max_depth: 2.0
+grasped_dtc_max_value: 0.105
+env_dtc_max_value: 0.425
+grasped_normals_mask_max_dtc_value: 0.105
+env_normals_mask_max_dtc_value: 0.425
+clamp_dtc: true
+dtc_adaptive_normalization: false
+mask_normals_within_sdf: true
+adaptive_normals_mask: true
+learnable_contact_preprocess_params: false
+contact_model_name: local_multitask_outhd64all_home_crop_h144w144d48_ctxt_seed_183386_epoch_9
+contact_estimation_model_ckpt_path: ~/fish_leon/contact_estimation/artifacts/175604_2/checkpoints/epoch=09-val_loss=0.00.ckpt
+num_eval: 5
+debug_timestamps: false
+open_loop: false
+action_trajectories: true
+stop_after_action: false
+interpolation_frequency: 25
+policy_frequency: 5
+wait_for_new_camera_frames: true
+random_start: false
+eval_starts: ${root_dir}/FISH/eval_starts/${suite.name}_${obs_type}/${task_name}
+train_demo_idxs_list_or_num: null
+num_valid_demos: null
+val_num_groups: 3
+name_of_expert_demo: 112_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act
+expert_dataset_dirpath: ${root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${name_of_expert_demo}
+expert_dataset: ${expert_dataset_dirpath}/demos.zarr
+action_key: ${oc.if_else:${action_trajectories}, 'action_trajectory_${interpolation_frequency}hz',
+  'action'}
+semantic_demo_grouping_name: semantic_demo_grouping.yaml
+semantic_demo_grouping: ${expert_dataset_dirpath}/${semantic_demo_grouping_name}
+expert_dataset_config: ${expert_dataset_dirpath}/demo_config.yaml
+bc_regularize: false
+bc_weight_type: qfilter
+load_checkpoint: ${agent.load_checkpoint}
+wandb_run_id: '1002_0'
+true_action_history: false
+wandb_notes: null
+checkpoint_epoch: 12000
+load_residual_weight: false
+checkpoint_root_dir: /home/${oc.env:USER}/fish_leon/FISH
+checkpoint_weight_dir: ${checkpoint_root_dir}/exp_local/${suite.name}_${obs_type}/${task_name}/${wandb_run_id}
+residual_weight: ${root_dir}/FISH/weights/${suite.name}_${obs_type}/${task_name}/weight.pt
+experiment_dir: ./exp_local/${suite.name}_${obs_type}/${task_name}/${wandb_run_id}
+final_experiment_dir: ${experiment_dir}/${oc.generate_run_id:}
+agent:
+  _target_: agent.diffusion_policy.DiffusionPolicyAgent
+  name: diffusion_policy
+  load_checkpoint: ${eval}
+  device: ${device}
+  n_obs_steps: ${.config.policy_cfg.n_obs_steps}
+  suite_name: ${suite.name}
+  obs_type: ${obs_type}
+  enable_arm: ${eval}
+  enable_camera: ${eval}
+  use_tb: ${use_tb}
+  desired_image_shape:
+  - 13
+  - 180
+  - 240
+  orig_cam_shape:
+  - 3
+  - 240
+  - 320
+  config:
+    _target_: agent.diffusion_policy.DiffusionPolicyAgentConfig
+    compile: false
+    device: ${device}
+    cam_resize_shape: ${agent.desired_image_shape}
+    orig_cam_shape: ${agent.orig_cam_shape}
+    policy_frequency: ${policy_frequency}
+    interpolation_frequency: ${interpolation_frequency}
+    policy_cfg:
+      _target_: lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig
+      n_obs_steps: 1
+      horizon: 36
+      n_action_steps: ${agent.config.policy_cfg.horizon}
+      input_shapes:
+        observation.image: ${agent.config.cam_resize_shape}
+        context_observation.image: ${agent.config.cam_resize_shape}
+        observation.state:
+        - 8
+        observation.action_history:
+        - 7
+      output_shapes:
+        action:
+        - 7
+      input_normalization_modes:
+        observation.image: mean_std
+        observation.state: min_max
+        observation.action_history: min_max
+      output_normalization_modes:
+        action: min_max
+      vision_backbone: resnet18
+      pretrained_backbone_weights: null
+      transforms:
+      - _target_: torchaug.transforms.RandomAffine
+        degrees:
+        - -5
+        - 5
+        translate:
+        - 0.05
+        - 0.05
+        batch_transform: true
+        num_chunks: -1
+        batch_inplace: true
+      - _target_: torchaug.transforms.RandomColorJitter
+        brightness: 0.3
+        contrast: 0.4
+        saturation: 0.5
+        hue: 0.08
+        batch_transform: true
+        num_chunks: -1
+        batch_inplace: true
+      use_group_norm: true
+      spatial_softmax_num_keypoints: 32
+      action_history_encoder_config:
+        _target_: lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig
+        in_channels: 7
+        out_channels: 32
+        history_length: ${agent.config.policy_cfg.n_action_steps}
+        kernel_size: ${agent.config.policy_cfg.kernel_size}
+        downsample_kernel_size: 3
+        downsample_stride: 2
+        downsample_padding: 1
+      down_dims:
+      - 256
+      - 512
+      - 1024
+      kernel_size: 5
+      n_groups: 8
+      diffusion_step_embed_dim: 128
+      use_film_scale_modulation: true
+      noise_scheduler_type: DDIM
+      beta_schedule: squaredcos_cap_v2
+      beta_start: 0.0001
+      beta_end: 0.02
+      prediction_type: epsilon
+      clip_sample: true
+      clip_sample_range: 1.0
+      num_train_timesteps: 50
+      num_inference_steps: 10
+      do_mask_loss_for_padding: false
+    train_cfg:
+      _target_: utils.TrainConfig
+      lr: 0.0001
+      lr_scheduler: cosine
+      lr_warmup_steps: 500
+      adam_betas:
+      - 0.95
+      - 0.999
+      adam_eps: 1.0e-08
+      adam_weight_decay: 1.0e-06
+      grad_clip_norm: 10
+      offline_steps: ${num_train_frames_diffusion}
+      use_amp: true
+    observation_cfg:
+      _target_: agent.encoder.VisualFeatureSet
+      use_depth: ${use_depth}
+      use_color: ${use_color}
+      mask_input_dict:
+        _target_: agent.encoder.MaskInputDict
+        enable: ${use_masks}
+        representation: ${mask_representation}
+        mask_list: ${mask_list}
+      crop_input_config:
+        _target_: agent.encoder.CropInputConfig
+        color_crop_type: ${color_crop_type}
+        depth_crop_type: ${depth_crop_type}
+        segmask_crop_type: ${segmask_crop_type}
+        crop_hw: ${crop_hw}
+        crop_down_offset: ${crop_down_offset}
+        add_crop_binary_mask: ${add_crop_binary_mask}
+        add_coord_conv_map: ${add_coord_conv_map}
+      context_input_config:
+        _target_: agent.encoder.ContextInputConfig
+        use_color: ${use_context_color}
+        use_depth: ${use_context_depth}
+        mask_input_dict:
+          _target_: agent.encoder.MaskInputDict
+          enable: ${use_context_segmask}
+          representation: ${mask_representation}
+          mask_list: ${mask_list}
+        crop_input_config:
+          _target_: agent.encoder.CropInputConfig
+          color_crop_type: ${context_color_crop_type}
+          depth_crop_type: ${context_depth_crop_type}
+          segmask_crop_type: ${context_segmask_crop_type}
+          crop_hw: ${crop_hw}
+          crop_down_offset: ${crop_down_offset}
+          add_crop_binary_mask: ${context_add_crop_binary_mask}
+          add_coord_conv_map: ${context_add_coord_conv_map}
+      mask_soft_approx_scheduler_config:
+        _target_: agent.encoder.MaskSoftApproxSchedulerConfig
+        num_steps: 40000
+        initial_value: 10.0
+        final_value: 1000.0
+        interpolation_scheme: constant
+      use_contact_map: ${use_contact_map}
+      use_sdf_maps: ${use_sdf_maps}
+      use_normals_maps: ${use_normals_maps}
+      which_objects: ${which_objects}
+      grasped_dtc_max_value: ${grasped_dtc_max_value}
+      env_dtc_max_value: ${env_dtc_max_value}
+      grasped_normals_mask_max_dtc_value: ${grasped_normals_mask_max_dtc_value}
+      env_normals_mask_max_dtc_value: ${env_normals_mask_max_dtc_value}
+      clamp_dtc: ${clamp_dtc}
+      max_contact_prob: ${max_contact_prob}
+      mask_normals_within_sdf: ${mask_normals_within_sdf}
+      dtc_adaptive_normalization: ${dtc_adaptive_normalization}
+      adaptive_normals_mask: ${adaptive_normals_mask}
+      max_depth: ${max_depth}
+      image_shape: ${agent.desired_image_shape}
+      learnable_contact_preprocess_params: ${learnable_contact_preprocess_params}
+      learning_rate: ${agent.config.train_cfg.lr}
+      weight_decay: 0.0
+      contact_model_name: ${contact_model_name}
+      zero_centered: false
+suite:
+  suite: frankagym
+  name: frankagym
+  frame_stack: ${agent.n_obs_steps}
+  action_repeat: 1
+  discount: 0.99
+  hidden_dim: 1024
+  num_train_frames: 2010
+  num_seed_frames: 260
+  num_train_epochs: 5000
+  validate_every_epochs: 100
+  validate_diffusion_on_action_loss_every_epochs: 500
+  train_eval_diffusion_on_action_loss_every_epochs: 500
+  check_topk_every_epochs: 10
+  save_snapshot_every_epochs: 5000
+  eval_every_frames: 2000
+  num_eval_episodes: 5
+  save_snapshot: true
+  wait_for_user_to_start_episode: true
+  task_make_fn:
+    _target_: suite.frankagym.make
+    name: ${task_name}
+    height: 240
+    width: 320
+    frame_stack: ${suite.frame_stack}
+    action_repeat: ${suite.action_repeat}
+    seed: ${seed}
+    enable_arm: ${agent.enable_arm}
+    enable_gripper: ${enable_gripper}
+    start_with_gripper_open: ${start_with_gripper_open}
+    enable_camera: ${agent.enable_camera}
+    path_to_depth_extrinsics: ${path_to_depth_extrinsics}
+    contact_estimation_model_ckpt_path: ${contact_estimation_model_ckpt_path}
+    x_limit: ${x_limit}
+    y_limit: ${y_limit}
+    z_limit: ${z_limit}
+    device: ${device}
+    interpolation_frequency: ${interpolation_frequency}
+    policy_frequency: ${policy_frequency}
+    debug_timestamps: ${debug_timestamps}
+    stop_after_action: ${stop_after_action}
+    open_loop: ${open_loop}
+    wait_for_new_camera_frames: ${wait_for_new_camera_frames}
+    action_key: ${action_key}
+    action_trajectory_horizon: ${agent.config.policy_cfg.horizon}
+    action_trajectories: ${action_trajectories}
+    path_to_zarr_dataset: ${expert_dataset}
+    agent_policy_cfg: ???
+    true_action_history: ${true_action_history}
+num_train_frames_bc: 50000
+num_train_frames_drq: 1100000
+stddev_schedule_drq: linear(1.0,0.1,100000)
+task_name: FrankaInsertion-v1
+num_train_frames_vinn: 25000
+num_train_frames_diffusion: 1000000
+num_train_epochs_bc: 5000
+num_train_epochs_diffusion: 5000
+validate_every_epochs_bc: 5
+validate_every_epochs_diffusion: 25
+validate_diffusion_on_action_loss_every_epochs: 50
+train_eval_diffusion_on_action_loss_every_epochs: 500
+check_topk_every_epochs: 5
+check_topk_every_epochs_diffusion: ${validate_diffusion_on_action_loss_every_epochs}
+save_snapshot_every_epochs_diffusion: 5000
+x_limit:
+- 0.2
+- 0.7
+y_limit:
+- -0.4
+- 0.4
+z_limit:
+- -0.05
+- 0.55
+home_displacement:
+- 0.55
+- 0.0
+- 0.55
+- 180.0
+- 0.0
+- 0.0
+enable_gripper: true
+start_with_gripper_open: true
+offset_mask:
+- 1
+- 1
+- 1
+- 1
+- 1
+- 1
+path_to_depth_extrinsics: ~/fish_leon/FISH/cfgs/camera_poses/camera_poses_L515/20240904-122305/color_tf_world.npy

4r9mcmxe/.hydra/hydra.yaml

@@ -0,0 +1,169 @@
+hydra:
+  run:
+    dir: ${final_experiment_dir}
+  sweep:
+    dir: ${final_experiment_dir}
+    subdir: ${hydra.job.num}
+  launcher:
+    submitit_folder: ${final_experiment_dir}/.slurm
+    timeout_min: 60
+    cpus_per_task: null
+    gpus_per_node: null
+    tasks_per_node: 1
+    mem_gb: null
+    nodes: 1
+    name: ${hydra.job.name}
+    stderr_to_stdout: false
+    _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.LocalLauncher
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+    params: null
+  help:
+    app_name: ${hydra.job.name}
+    header: '${hydra.help.app_name} is powered by Hydra.
+
+      '
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help
+
+      '
+    template: '${hydra.help.header}
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (group=option)
+
+
+      $APP_CONFIG_GROUPS
+
+
+      == Config ==
+
+      Override anything in the config (foo.bar=value)
+
+
+      $CONFIG
+
+
+      ${hydra.help.footer}
+
+      '
+  hydra_help:
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+    hydra_help: ???
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  env: {}
+  mode: RUN
+  searchpath: []
+  callbacks: {}
+  output_subdir: .hydra
+  overrides:
+    hydra:
+    - hydra.mode=RUN
+    task:
+    - agent=diffusion
+    - suite=frankagym
+    - suite/frankagym_task@_global_=insertion
+  job:
+    name: eval_policy
+    chdir: true
+    override_dirname: agent=diffusion,suite/frankagym_task@_global_=insertion,suite=frankagym
+    id: ???
+    num: ???
+    config_name: config_eval
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.1'
+    cwd: /home/leonmkim/fish_leon/FISH
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /home/leonmkim/fish_leon/FISH/cfgs
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /home/leonmkim/fish_leon/FISH/exp_local/frankagym_pixels/FrankaInsertion-v1/1002_0/4r9mcmxe
+    choices:
+      suite: frankagym
+      suite/frankagym_task@_global_: insertion
+      agent: diffusion
+      hydra/env: default
+      hydra/callbacks: null
+      hydra/job_logging: default
+      hydra/hydra_logging: default
+      hydra/hydra_help: default
+      hydra/help: default
+      hydra/sweeper: basic
+      hydra/launcher: submitit_local
+      hydra/output: default
+  verbose: false

4r9mcmxe/.hydra/overrides.yaml

@@ -0,0 +1,3 @@
+- agent=diffusion
+- suite=frankagym
+- suite/frankagym_task@_global_=insertion

4r9mcmxe/eval_policy.log

@@ -0,0 +1,15 @@
+[2025-01-07 21:43:24,810][py.warnings][WARNING] - /home/leonmkim/fish_leon/FISH/eval_policy.py:428: UserWarning: 
+The version_base parameter is not specified.
+Please specify a compatability version level, or None.
+Will assume defaults for version 1.1
+  @hydra.main(config_path='cfgs', config_name='config_eval')
+
+[2025-01-07 21:43:24,813][py.warnings][WARNING] - /home/leonmkim/fish_leon/FISH/eval_policy.py:365: UserWarning: 
+The version_base parameter is not specified.
+Please specify a compatability version level, or None.
+Will assume defaults for version 1.1
+  hydra.initialize(
+
+[2025-01-07 21:43:28,293][py.warnings][WARNING] - /home/leonmkim/fish_leon/FISH/eval_policy.py:414: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
+  payload = torch.load(f)
+