eval

cross_peginsertionside/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4236675d6537384de34ae6b135dbc3704e16feea0e1de27cb46723ceb0caab6a
+size 93175179

cross_peginsertionside/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: PegInsertionSide-v1
+env_cfg:
+  name: maniskill
+  task: PegInsertionSide-v1
+  dataset_repo_id: johnMinelli/ManiSkill_PegInsertionSide-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

cross_stackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eda55ea9e1f8307ee644bd07811ecfa905b19768cf8af8da8825ea015ad98c21
+size 93175118

cross_stackcube/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: StackCube-v1
+env_cfg:
+  name: maniskill
+  task: StackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_StackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 30
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

cross_tworobotstackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41ca41133e0f04e72bf55945157c3adf1191e52a018134a4c8155b07d80df57e
+size 93219135

cross_tworobotstackcube/params/agent.yaml

@@ -0,0 +1,99 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: TwoRobotStackCube-v1
+env_cfg:
+  name: maniskill
+  task: TwoRobotStackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_TwoRobotStackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.image: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 18
+    observation.privileged: !!python/tuple
+    - 40
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 16
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.image: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

cross_unitreeg1transportbox/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a72cbc84bd0d33c7bb3dbb7a3858e05a7014b8b93a369c03699909f5c335a3a
+size 93255978

cross_unitreeg1transportbox/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: UnitreeG1TransportBox-v1
+env_cfg:
+  name: maniskill
+  task: UnitreeG1TransportBox-v1
+  dataset_repo_id: johnMinelli/ManiSkill_UnitreeG1TransportBox-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.head_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 25
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 25
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.head_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

lstm_peginsertionside/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:481b2a2417fd1ac3aa8691f7ba906efe587dd0fc67e9231a509b4d883d828714
+size 42676885

lstm_peginsertionside/params/agent.yaml

@@ -0,0 +1,94 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: LSTMPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: PegInsertionSide-v1
+env_cfg:
+  name: maniskill
+  task: PegInsertionSide-v1
+  dataset_repo_id: johnMinelli/ManiSkill_PegInsertionSide-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

lstm_stackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3863c1d750feb2d79aa18c36d2d7c895f24d9e9dd48b1d6c45747f98ec459665
+size 42676824

lstm_stackcube/params/agent.yaml

@@ -0,0 +1,94 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: LSTMPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: StackCube-v1
+env_cfg:
+  name: maniskill
+  task: StackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_StackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 30
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

lstm_tworobotstackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60c1de3094b962f8eaf3c7d9faf96e59e9eb92431850cc59d3798c9f5b4b6bfd
+size 41653831

lstm_tworobotstackcube/params/agent.yaml

@@ -0,0 +1,89 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: LSTMPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: TwoRobotStackCube-v1
+env_cfg:
+  name: maniskill
+  task: TwoRobotStackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_TwoRobotStackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.image: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 18
+    observation.privileged: !!python/tuple
+    - 40
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 16
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.image: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

lstm_unitreeg1transportbox/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f2f44037adaebbebfa4cfd8f3ccdf07d068616a1475763bb49bc53964af02be
+size 42724986

lstm_unitreeg1transportbox/params/agent.yaml

@@ -0,0 +1,94 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: LSTMPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: UnitreeG1TransportBox-v1
+env_cfg:
+  name: maniskill
+  task: UnitreeG1TransportBox-v1
+  dataset_repo_id: johnMinelli/ManiSkill_UnitreeG1TransportBox-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.head_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 25
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 25
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.head_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

none_peginsertionside/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91f462488160c8089ccd2e7071f566291b1a24f0fba9748a749682b9797d7ffc
+size 93175181

none_peginsertionside/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: false
+  ca_att: false
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: PegInsertionSide-v1
+env_cfg:
+  name: maniskill
+  task: PegInsertionSide-v1
+  dataset_repo_id: johnMinelli/ManiSkill_PegInsertionSide-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

none_stackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c991aac3d9207510a2a021c9d6c3b93c17ea6a015c3bfb393830eee59ee9d277
+size 93175121

none_stackcube/params/agent.yaml

@@ -0,0 +1,98 @@
+seed: 42
+device: cuda:0
+n_epochs: 200
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: TransformerPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_flag: false
+  ca_flag: false
+  sta_flag: false
+  sta_residual: false
+  sta_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Agent
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: StackCube-v1
+env_cfg:
+  name: maniskill
+  task: StackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_StackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 30
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

none_tworobotstackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38bef25c60cb946966c5c56ea2d881437e4e2dcf3bb82aaa3da0b21e62c08c3f
+size 93219134

none_tworobotstackcube/params/agent.yaml

@@ -0,0 +1,99 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: false
+  ca_att: false
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: TwoRobotStackCube-v1
+env_cfg:
+  name: maniskill
+  task: TwoRobotStackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_TwoRobotStackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.image: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 18
+    observation.privileged: !!python/tuple
+    - 40
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 16
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.image: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

none_unitreeg1transportbox/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0274a2dee4c158384796bc95b982f4194839e004951a7ea39759a56c933889f5
+size 93255981

none_unitreeg1transportbox/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: false
+  ca_att: false
+  army_att: false
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: UnitreeG1TransportBox-v1
+env_cfg:
+  name: maniskill
+  task: UnitreeG1TransportBox-v1
+  dataset_repo_id: johnMinelli/ManiSkill_UnitreeG1TransportBox-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.head_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 25
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 25
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.head_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_peginsertionside/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2e4814f84dab65674c120a46a2fb252580538f57f43bc8323849a7fc62348c41
+size 93175183

sta_peginsertionside/params/agent.yaml

@@ -0,0 +1,98 @@
+seed: 42
+device: cuda:0
+n_epochs: 200
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: TransformerPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_flag: true
+  ca_flag: true
+  sta_flag: true
+  sta_residual: false
+  sta_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: true
+preprocessor_class: Lerobot_2_Agent
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: PegInsertionSide-v1
+env_cfg:
+  name: maniskill
+  task: PegInsertionSide-v1
+  dataset_repo_id: johnMinelli/ManiSkill_PegInsertionSide-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_peginsertionside_noisy/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95149197c474d978c9c540c6011aa373f3ca750655d23c686527480e48a0fefc
+size 93175182

sta_peginsertionside_noisy/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: true
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: PegInsertionSide-v1
+env_cfg:
+  name: maniskill
+  task: PegInsertionSide-v1
+  dataset_repo_id: johnMinelli/ManiSkill_PegInsertionSide-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_stackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d714678729d554c791038f8a62d6099edfdedcebec87417f3abf67820b69eeb0
+size 93175117

sta_stackcube/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: StackCube-v1
+env_cfg:
+  name: maniskill
+  task: StackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_StackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 30
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_stackcube_noisy/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef31875454bfd56a639d112ee74b16f7336ae05e5e5c70c0cfa1abf25b3dd3f0
+size 93175116

sta_stackcube_noisy/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: true
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: StackCube-v1
+env_cfg:
+  name: maniskill
+  task: StackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_StackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 30
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_tworobotstackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da45ef75d11ce0f92c8a193a6662a5d5cd1aa139f07a89f89fe92a1e46fd977c
+size 93219132

sta_tworobotstackcube/params/agent.yaml

@@ -0,0 +1,99 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: TwoRobotStackCube-v1
+env_cfg:
+  name: maniskill
+  task: TwoRobotStackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_TwoRobotStackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.image: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 18
+    observation.privileged: !!python/tuple
+    - 40
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 16
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.image: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_tworobotstackcube_noisy/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74b77a37d4f2c2ddeb23580ae48e1e89788c86e082666010733917800b5406ec
+size 93219131

sta_tworobotstackcube_noisy/params/agent.yaml

@@ -0,0 +1,99 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: true
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: TwoRobotStackCube-v1
+env_cfg:
+  name: maniskill
+  task: TwoRobotStackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_TwoRobotStackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.image: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 18
+    observation.privileged: !!python/tuple
+    - 40
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 16
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.image: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_unitreeg1transportbox/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba139bc6eef6088de70c5d67f1e97da247865c75db47bff8f76448b5b8cd8c76
+size 93255978

sta_unitreeg1transportbox/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: UnitreeG1TransportBox-v1
+env_cfg:
+  name: maniskill
+  task: UnitreeG1TransportBox-v1
+  dataset_repo_id: johnMinelli/ManiSkill_UnitreeG1TransportBox-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.head_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 25
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 25
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.head_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

sta_unitreeg1transportbox_noisy/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:275345602402e5826857b67fd9f5d1a43f66841a28b683c3bcf3bada3ebb01be
+size 93255977

sta_unitreeg1transportbox_noisy/params/agent.yaml

@@ -0,0 +1,104 @@
+seed: 50
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: ARMYPolicy
+policy_kwargs:
+  dim_decoder: 512
+  n_decoder_layers: 4
+  n_decoder_heads: 8
+  use_rmsnorm: true
+  pre_norm: true
+  sa_att: true
+  ca_att: true
+  army_att: true
+  army_residual: false
+  army_norm: math
+  dropout: 0.0
+  dropout_residual: 0.0
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: true
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: UnitreeG1TransportBox-v1
+env_cfg:
+  name: maniskill
+  task: UnitreeG1TransportBox-v1
+  dataset_repo_id: johnMinelli/ManiSkill_UnitreeG1TransportBox-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.head_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 25
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 25
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.head_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

tcn_peginsertionside/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed19d3a5e08c7de93c8525848d6be466c083504400fa90e115ebc4280152ab56
+size 39797489

tcn_peginsertionside/params/agent.yaml

@@ -0,0 +1,94 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: TCNPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: PegInsertionSide-v1
+env_cfg:
+  name: maniskill
+  task: PegInsertionSide-v1
+  dataset_repo_id: johnMinelli/ManiSkill_PegInsertionSide-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

tcn_stackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98e6b3af802bae37423955954a602d833351925e09bbe7d44e7bacbd7da70c2c
+size 39797427

tcn_stackcube/params/agent.yaml

@@ -0,0 +1,94 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: TCNPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: StackCube-v1
+env_cfg:
+  name: maniskill
+  task: StackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_StackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.hand_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 9
+    observation.privileged: !!python/tuple
+    - 30
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 8
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.hand_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

tcn_tworobotstackcube/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb7c9a7d875b7552477079ed88cf8f96a99d91a41414b63d8e78ec5c3b3ab2c3
+size 38774435

tcn_tworobotstackcube/params/agent.yaml

@@ -0,0 +1,89 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: TCNPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: TwoRobotStackCube-v1
+env_cfg:
+  name: maniskill
+  task: TwoRobotStackCube-v1
+  dataset_repo_id: johnMinelli/ManiSkill_TwoRobotStackCube-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.image: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 18
+    observation.privileged: !!python/tuple
+    - 40
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 16
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.image: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false

tcn_unitreeg1transportbox/best_model.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f64c9e6df0228a0b67418c62536ba2aa6a495c39728cc20fbade9832494b9ea5
+size 39845585

tcn_unitreeg1transportbox/params/agent.yaml

@@ -0,0 +1,94 @@
+seed: 42
+device: cuda:0
+n_epochs: 50
+batch_size: 16
+gradient_accumulation_steps: 1
+mixed_precision: 'no'
+learning_rate: 8.0e-05
+lr_scheduler: constant
+ent_coef: 0.01
+l2_coef: 0.01
+weight_decay: 0
+max_grad_norm: 1.0
+agent_class: sl_agent.SL
+policy: TCNPolicy
+policy_kwargs:
+  dim_model: 512
+  final_ffn: true
+  action_horizon: 16
+  optimizer_kwargs:
+    eps: 1.0e-08
+  noise: false
+preprocessor_class: Lerobot_2_Army
+preprocessor_kwargs:
+  resize_images: !!python/tuple
+  - 94
+  - 94
+  normalize_images: true
+  tokenizer_kwargs:
+    model_template: qwen2-chat
+    num_image_token: 9
+    padding_max_length: 60
+    padding_side: right
+  prompt_generator: FixedPromptGenerator("Insert the peg into the hole.")
+  observation_modes:
+    observation.state: min_max
+  action_modes:
+    action: min_max
+  overwrite_stats: false
+  stats: null
+  data_key_map:
+    observation.images.base_camera: cam1
+    observation.images.hand_camera: cam2
+    observation.images.head_camera: cam2
+    observation.images.image_side_1: cam1
+    observation.images.image_side_2: cam2
+    observation.images.image_wrist_1: cam3
+    observation.images.image_wrist_2: cam4
+    observation.top: cam1
+    image: cam1
+    wrist_image: cam2
+    observation.state: state
+    state: state
+    action: action
+    actions: action
+  task: UnitreeG1TransportBox-v1
+env_cfg:
+  name: maniskill
+  task: UnitreeG1TransportBox-v1
+  dataset_repo_id: johnMinelli/ManiSkill_UnitreeG1TransportBox-v1_recovery
+  episodes: null
+  dataset_revision: v2.0
+  dataset_root: null
+  video_backend: pyav
+  input_shapes:
+    observation.images.base_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.images.head_camera: !!python/tuple
+    - 480
+    - 640
+    - 3
+    observation.state: !!python/tuple
+    - 25
+    observation.privileged: !!python/tuple
+    - 27
+  output_shapes:
+    action: !!python/tuple
+    - 16
+    - 25
+  normalization:
+    observation_modes:
+      observation.state: min_max
+    action_modes:
+      action: min_max
+  fps: 20
+  delta_timestamps:
+    expert_mask: '[i / 20 for i in range(16)]'
+    observation.images.base_camera: '[i / 20 for i in range(16)]'
+    observation.images.head_camera: '[i / 20 for i in range(16)]'
+    observation.state: '[i / 20 for i in range(16)]'
+    action: '[i / 20 for i in range(16)]'
+  image_transforms:
+    enable: false