image_encoders: encoder_cfg: type: crop_resnet34 resize_shape: - 224 - 224 crop_shape: null pretrained: true action_representation: policy_translator: _target_: srl_il.algo.base_algo.PolicyTranslatorRelative2Absolute translate_rot: false in_ee_frame: false action_pose_key: actions_arm current_pose_key: qpos_arm r6d_repr: true action_arm_dim: 9 data_augment: outname: actions_arm inputname: - actions_arm - qpos_arm type: abs_action_pose_2_rela translate_rot: false in_ee_frame: false r6d_repr: true action_arm_normalizer: type: augmentor_stats min_max: true max_data_points: 30000 batch_size: 64 seed: 43 debugrun: false dataset_name: mixedcubes-02+03+06 wandb_cfg: project: rwr2025task2 run_name: diffusion_policy_20251206-232055_mixedcubes-02+03+06 entity: rwr25 tags: - diffusion_policy - resnet34 - relative_no_rot_6d - mixedcubes-02+03+06 mode: online output_dir: runs/diffusion_policy_20251206-232055_mixedcubes-02+03+06 pipeline: _target_: srl_il.pipeline.imitation_learning.ImitationLearningPipeline dataset_cfg: data: _target_: srl_il.dataset.faive_dataset_zarr.faive_train_val_test data_directory: /home/ubuntu/srl_il/Data/mixedcubes-02+03+06/ data_filter_keys: [] rgb_keys: - oakd_wrist_view/color - oakd_side_view/color rgb_resize_to: - 224 - 224 rgb_crop_dict: oakd_wrist_view: top: 0 left: 0 bottom: null right: null oakd_side_view: top: 0 left: 0 bottom: null right: null zarr_cfgs: n_reader_workers: 1 n_writer_workers: 4 cache_root: /home/ubuntu/srl_il/Data/cache dataset_name: mixedcubes-02+03+06 overwrite: false test_fraction: 0.15 val_fraction: 0.15 window_size_train: 21 window_size_test: 21 keys_traj: - - actions_arm - qpos_arm - 1 - null - - actions_hand - actions_hand - 1 - null - - qpos_arm - qpos_arm - 0 - 1 - - qpos_hand - qpos_hand - 0 - 1 - - side_view_images - oakd_side_view/color - 0 - 1 - - wrist_view_images - oakd_wrist_view/color - 0 - 1 keys_global: [] pad_before: false pad_after: false pad_type: near random_seed: 43 batch_size: 64 pin_memory: true num_workers: 4 persistent_workers: true algo_cfg: _target_: srl_il.algo.diffusion_policy.DiffusionPolicyTrainer algo_cfg: device: cuda target_dims: actions_arm: 9 actions_hand: 17 T_target: 20 network_is_causal: true network_group_keys: - qpos - side_view_images - wrist_view_images network_cfg: d_model: 512 nhead: 8 num_blocks: 6 dim_feedforward: 2048 dropout: 0.1 scheduler_cfg: scheduler_type: ddim num_train_timesteps: 100 num_inference_steps: 16 beta_schedule: squaredcos_cap_v2 clip_sample: true beta_start: 0.0001 beta_end: 0.02 trainer_cfg: loss_params: null optimizer_cfg: network: optm_cls: torch.optim.Adam lr: 0.0002 weight_decay: 0.0001 obs_encoder: optm_cls: torch.optim.Adam lr: 0.0002 weight_decay: 0.0001 obs_encoder_cfg: output_dim: 512 obs_groups_cfg: qpos: datakeys: - qpos_arm - qpos_hand encoder_cfg: type: lowdim_concat input_dim_total: 24 posemb_cfg: type: none side_view_images: datakeys: - side_view_images backbone_instance_name: side_img_backbone encoder_cfg: type: crop_resnet34 resize_shape: - 224 - 224 crop_shape: null pretrained: true posemb_cfg: type: none wrist_view_images: datakeys: - wrist_view_images backbone_instance_name: wrist_img_backbone encoder_cfg: type: crop_resnet34 resize_shape: - 224 - 224 crop_shape: null pretrained: true posemb_cfg: type: none group_emb_cfg: type: each_group_learned policy_cfg: policy_bs: 1 policy_translator: _target_: srl_il.algo.base_algo.PolicyTranslatorRelative2Absolute translate_rot: false in_ee_frame: false action_pose_key: actions_arm current_pose_key: qpos_arm r6d_repr: true policy_aggregator_cfg: type: temporal_aggr update_every: 20 k: 0.01 policy_obs_list: - - qpos_arm - 1 - - qpos_hand - 1 - - side_view_images - 1 - - wrist_view_images - 1 lr_scheduler_cfg: network: type: diffusers name: cosine params: num_warmup_steps: 1000 num_training_steps: 500000 step_with_metrics: false obs_encoder: type: diffusers name: cosine params: num_warmup_steps: 1000 num_training_steps: 500000 step_with_metrics: false training_cfg: num_epochs: 500 num_steps_per_epoch: 100 num_eval_steps_per_epoch: 10 num_test_steps_per_epoch: 10 steps_saving: 50 rollout: enabled: false visualization: enabled: true num_samples: 1 every_n_epoch: 1 visualizations: - type: action_projector video: false visualization_prefix: wrist_position - type: joints video: false visualization_prefix: hand_joints normalizer_cfg: actions_arm: type: augmentor_stats min_max: true max_data_points: 30000 actions_hand: type: dataset_stats min_max: true dataname: actions_hand qpos_arm: type: dataset_stats dataname: qpos_arm qpos_hand: type: dataset_stats dataname: qpos_hand side_view_images: type: hardcode mean: 0.0 std: 1.0 wrist_view_images: type: hardcode mean: 0.0 std: 1.0 data_augmentation_cfg: data_augments: - outname: actions_arm inputname: - actions_arm - qpos_arm type: abs_action_pose_2_rela translate_rot: false in_ee_frame: false r6d_repr: true - outname: qpos_hand type: gaussian_noise mean: 0.0 std: 0.01 - outname: wrist_view_images type: color_jitter brightness: 0.1 contrast: 0.05 saturation: 0.05 hue: 0.05 - outname: side_view_images type: color_jitter brightness: 0.1 contrast: 0.05 saturation: 0.05 hue: 0.05 sim_env_cfg: {} visualizer_cfg: action_projector: action_key: actions_arm img_views: side_view_images: extrinsics_key: null intrinsics_key: null wrist_view_images: extrinsics_key: null intrinsics_key: null visualizations: - reference_frame: null viz_args: keypoints_indexes: - 0 joints: action_key: actions_hand visualizations: - viz_args: null joints_indexes: - 0 - 1 - 2 - 3 - 4 - 5 - 6 - 7 - 8 - 9 - 10 - 11 - 12 - 13 - 14 - 15 - 16 accelerator_cfg: enabled: false