Add files using upload-large-folder tool

code/reveal_vla_bimanual/train/configs/rlbench_lift_ball_backbone_only_clip_current.yaml

@@ -0,0 +1,92 @@
+experiment_name: rlbench_lift_ball_backbone_only_clip_current
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: null
+init_strict: false
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball]
+  train_episodes: [0]
+  val_episodes: [1]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+
+optim:
+  epochs: 10
+  batch_size: 8
+  num_workers: 0
+  lr: 0.0002
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_lift_ball_backbone_only_clip_current_3train.yaml

@@ -0,0 +1,92 @@
+experiment_name: rlbench_lift_ball_backbone_only_clip_current_3train
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: null
+init_strict: false
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball]
+  train_episodes: [0, 1, 2]
+  val_episodes: [3]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+
+optim:
+  epochs: 12
+  batch_size: 8
+  num_workers: 2
+  lr: 0.0002
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_lift_ball_backbone_only_clip_current_4train.yaml

@@ -0,0 +1,92 @@
+experiment_name: rlbench_lift_ball_backbone_only_clip_current_4train
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: null
+init_strict: false
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball]
+  train_episodes: [0, 1, 2, 3]
+  val_episodes: [4]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+
+optim:
+  epochs: 10
+  batch_size: 8
+  num_workers: 2
+  lr: 0.0002
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_lift_ball_backbone_only_clip_current_wide.yaml

@@ -0,0 +1,94 @@
+experiment_name: rlbench_lift_ball_backbone_only_clip_current_wide
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: /workspace/outputs/rlbench_current/rlbench_lift_ball_backbone_only_clip_current_3train/checkpoint_best.pt
+init_strict: true
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball]
+  train_episodes: [0, 1, 2, 3, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]
+  val_episodes: [32, 33, 34, 35, 36, 37, 38, 39]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+  max_train_samples: 4096
+  max_val_samples: 2048
+
+optim:
+  epochs: 8
+  batch_size: 16
+  num_workers: 4
+  lr: 0.0001
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_lift_ball_backbone_only_clip_step1.yaml

@@ -0,0 +1,95 @@
+experiment_name: rlbench_lift_ball_backbone_only_clip_step1
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: /workspace/outputs/rlbench_current/rlbench_lift_ball_backbone_only_clip_current_wide/checkpoint_stable.pt
+init_strict: true
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball]
+  train_episodes: [0, 1, 2, 3, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]
+  val_episodes: [32, 33, 34, 35, 36, 37, 38, 39]
+  resolution: 224
+  chunk_horizon: 8
+  supervise_action_steps: 1
+  proprio_dim: 32
+  history_steps: 2
+  max_train_samples: 4096
+  max_val_samples: 2048
+
+optim:
+  epochs: 8
+  batch_size: 16
+  num_workers: 4
+  lr: 0.00005
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_push_box_backbone_only_clip_step1.yaml

@@ -0,0 +1,95 @@
+experiment_name: rlbench_push_box_backbone_only_clip_step1
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: /workspace/outputs/rlbench_current/rlbench_subset3_backbone_only_clip_current_common23/checkpoint_stable.pt
+init_strict: true
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_push_box]
+  train_episodes: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79]
+  val_episodes: [80, 81, 82, 83, 84, 85, 86, 87, 88, 89]
+  resolution: 224
+  chunk_horizon: 8
+  supervise_action_steps: 1
+  proprio_dim: 32
+  history_steps: 2
+  max_train_samples: 12288
+  max_val_samples: 2048
+
+optim:
+  epochs: 6
+  batch_size: 32
+  num_workers: 8
+  lr: 0.00005
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_subset3_backbone_only_clip_current.yaml

@@ -0,0 +1,94 @@
+experiment_name: rlbench_subset3_backbone_only_clip_current
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: null
+init_strict: false
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball, bimanual_push_box, bimanual_dual_push_buttons]
+  train_episodes: [0, 1, 2, 3]
+  val_episodes: [4]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+  max_train_samples: 2048
+  max_val_samples: 512
+
+optim:
+  epochs: 6
+  batch_size: 8
+  num_workers: 2
+  lr: 0.0002
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_subset3_backbone_only_clip_current_common23.yaml

@@ -0,0 +1,94 @@
+experiment_name: rlbench_subset3_backbone_only_clip_current_common23
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: /workspace/outputs/rlbench_current/rlbench_subset3_backbone_only_clip_current_valid9/checkpoint_best.pt
+init_strict: true
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball, bimanual_push_box, bimanual_dual_push_buttons]
+  train_episodes: [0, 1, 2, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]
+  val_episodes: [26, 27, 28, 29]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+  max_train_samples: 8192
+  max_val_samples: 2048
+
+optim:
+  epochs: 8
+  batch_size: 16
+  num_workers: 4
+  lr: 0.0001
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/configs/rlbench_subset3_backbone_only_clip_current_valid9.yaml

@@ -0,0 +1,94 @@
+experiment_name: rlbench_subset3_backbone_only_clip_current_valid9
+output_dir: /workspace/outputs/rlbench_current
+device: cuda
+seed: 17
+init_checkpoint: /workspace/outputs/rlbench_current/rlbench_lift_ball_backbone_only_clip_current_3train/checkpoint_best.pt
+init_strict: true
+
+data:
+  dataset_root: /workspace/data/rlbench2
+  tasks: [bimanual_lift_ball, bimanual_push_box, bimanual_dual_push_buttons]
+  train_episodes: [0, 1, 10, 11, 12, 13, 14, 15]
+  val_episodes: [16]
+  resolution: 224
+  chunk_horizon: 8
+  proprio_dim: 32
+  history_steps: 2
+  max_train_samples: 4096
+  max_val_samples: 1024
+
+optim:
+  epochs: 8
+  batch_size: 16
+  num_workers: 4
+  lr: 0.0001
+  weight_decay: 0.0001
+
+trainer:
+  policy_type: backbone_only
+  use_bf16: true
+  grad_clip_norm: 1.0
+  freeze_backbone: true
+  gradient_checkpointing: false
+  plan_during_train: false
+  plan_during_eval: false
+  support_mode_conditioning: true
+
+policy:
+  backbone:
+    model_name: openai/clip-vit-base-patch32
+    hidden_dim: 512
+    max_text_tokens: 32
+    freeze_backbone: true
+    gradient_checkpointing: false
+    use_dummy_backbone: false
+  fusion:
+    hidden_dim: 512
+    num_cameras: 3
+    num_layers: 4
+    num_heads: 8
+    ff_dim: 2048
+    dropout: 0.1
+    proprio_dim: 32
+    proprio_tokens: 1
+  memory:
+    hidden_dim: 512
+    history_steps: 2
+    num_layers: 1
+    dropout: 0.1
+  decoder:
+    hidden_dim: 512
+    num_heads: 8
+    num_layers: 4
+    ff_dim: 2048
+    dropout: 0.1
+    chunk_size: 8
+    action_dim: 14
+    num_candidates: 8
+  reveal_head:
+    hidden_dim: 512
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+    belief_map_size: 32
+    predict_belief_map: true
+  world_model:
+    hidden_dim: 512
+    action_dim: 14
+    num_support_modes: 3
+    num_approach_templates: 32
+    rollout_horizon: 5
+  planner:
+    hidden_dim: 512
+    num_candidates: 8
+    action_dim: 14
+    utility_margin: 0.1
+
+loss_weights:
+  action: 1.0
+  support_mode: 0.0
+  corridor: 0.0
+  persistence: 0.0
+  disturbance: 0.0
+  world_model: 0.0
+  belief: 0.0

code/reveal_vla_bimanual/train/losses.py

@@ -309,7 +309,11 @@ def compute_total_loss(
 ) -> dict[str, Tensor]:
     weights = weights or LossWeights()
     losses = {
-        "action": chunk_bc_loss(model_output["action_mean"], batch["action_chunk"]),
+        "action": chunk_bc_loss(
+            model_output["action_mean"],
+            batch["action_chunk"],
+            mask=batch.get("action_mask"),
+        ),
     }
     total = weights.action * losses["action"]
 

code/reveal_vla_bimanual/train/run_rlbench_experiment.py

@@ -74,6 +74,7 @@ def main() -> None:
         chunk_size=int(cfg.data.chunk_horizon),
         proprio_dim=int(cfg.data.proprio_dim),
         history_steps=int(cfg.data.get("history_steps", 2)),
+        supervise_action_steps=cfg.data.get("supervise_action_steps"),
         max_samples=cfg.data.get("max_train_samples"),
     )
     val_dataset = RLBenchOfflineChunkDataset(
@@ -84,6 +85,7 @@ def main() -> None:
         chunk_size=int(cfg.data.chunk_horizon),
         proprio_dim=int(cfg.data.proprio_dim),
         history_steps=int(cfg.data.get("history_steps", 2)),
+        supervise_action_steps=cfg.data.get("supervise_action_steps"),
         max_samples=cfg.data.get("max_val_samples"),
     )
     train_loader = _make_loader(

environment/README.md

@@ -28,6 +28,11 @@ This directory contains the machine snapshot and setup helpers for the runpod no
   - bundle-aware bootstrap script for a matching `/workspace` machine
 - `validate_same_machine.sh`
   - validation helper that runs `glxinfo`, RLBench import smoke, and `open_drawer` launch smoke
+- `hf_cli_version.txt`
+  - raw `hf version` output from the upload session
+- Hugging Face CLI used for upload in this session
+  - install command: `curl -LsSf https://hf.co/cli/install.sh | bash`
+  - install path on this node: `/workspace/.local/bin/hf`
 
 ## Default Layout
 
@@ -35,6 +40,7 @@ This directory contains the machine snapshot and setup helpers for the runpod no
 - bundled project code: `/workspace/VLAarchtests/code/reveal_vla_bimanual`
 - runtime symlink target used by the setup scripts: `/workspace/reveal_vla_bimanual`
 - third-party checkout root: `/workspace/third_party`
+- handoff instructions copied into the bundle: `/workspace/VLAarchtests/handoff/instructions.md`
 
 ## Validation Commands
 
@@ -42,4 +48,5 @@ This directory contains the machine snapshot and setup helpers for the runpod no
 DISPLAY=:99 glxinfo -B
 /workspace/.tools/micromamba/bin/micromamba run -r /workspace/.micromamba -p /workspace/envs/rlbench python -m sim_rlbench.launch_smoke --headless
 /workspace/VLAarchtests/environment/validate_same_machine.sh
+env PATH="/workspace/.local/bin:$PATH" hf version
 ```