Add NewModel v2 raw action sidecars

NewModel/v2/sidecars/raw/action_manifest.json

@@ -0,0 +1,492 @@
+{
+  "config": {
+    "data": {
+      "dataset_path": "../Data/stacking_cups_action_30hz",
+      "prepared_path": "outputs/FULL_VISION_512/cache/raw",
+      "action_representation": "raw",
+      "vision_cache_path": "outputs/FULL_VISION_512/cache/vision",
+      "camera_keys": [
+        "observation.images.global",
+        "observation.images.hand"
+      ],
+      "state_key": "observation.state",
+      "action_key": "action",
+      "observation_horizon": 1,
+      "action_horizon": 30,
+      "frequency_hz": 30.0,
+      "split_seed": 20260915,
+      "train_episodes": 42,
+      "val_episodes": 5,
+      "test_episodes": 5
+    },
+    "spline": {
+      "degree": 3,
+      "span_length_steps": 2,
+      "num_basis": 18,
+      "vocab_size": 256,
+      "regularization": 0.0001,
+      "implementation_version": "uniform_left_direct_fit_v1"
+    },
+    "vision": {
+      "image_size": 224,
+      "pooled_grid": 16,
+      "dino_model": "vit_large_patch14_reg4_dinov2.lvd142m",
+      "siglip_model": "vit_so400m_patch14_siglip_224",
+      "extraction_layer_offset": -2,
+      "pretrained": true,
+      "cache_dtype": "float16"
+    },
+    "policy": {
+      "architecture": "fm",
+      "model_size": "DiT-S",
+      "hidden_dim": 384,
+      "depth": 6,
+      "heads": 4,
+      "mlp_ratio": 4.0,
+      "dropout": 0.0,
+      "discrete_rounds": 8,
+      "block_size": 21,
+      "fm_steps": 12
+    },
+    "train": {
+      "seed": 7,
+      "updates": 50000,
+      "rtc_updates": 5000,
+      "batch_size": 32,
+      "effective_batch_size": 32,
+      "learning_rate": 0.0003,
+      "weight_decay": 0.0001,
+      "warmup_updates": 100,
+      "min_lr_ratio": 0.1,
+      "grad_clip": 1.0,
+      "eval_every": 100,
+      "validation_max_batches": 4,
+      "save_every": 1000,
+      "num_workers": 4,
+      "precision": "bf16",
+      "lambda_l1": 1.0,
+      "deterministic": true
+    },
+    "rtc": {
+      "spline_delay_min": 1,
+      "spline_delay_max": 5,
+      "raw_delay_min": 1,
+      "raw_delay_max": 10,
+      "delay_distribution": "truncated_decreasing_exponential_reference",
+      "zero_delay_eval": true
+    },
+    "wandb": {
+      "enabled": true,
+      "mode": "online",
+      "entity": "401910710-university-of-california-berkeley",
+      "base_project": "robot-policy-full-vision-512-basic",
+      "rtc_project": "robot-policy-full-vision-512-ttRTC",
+      "run_suffix": "dit-s-raw-512vision-generation-val",
+      "output_dir": "outputs/FULL_VISION_512/dit_s/raw/wandb"
+    }
+  },
+  "dataset_audit": {
+    "dataset": "/scratch/wangpc/B-Spline-Inpainting/Data/stacking_cups_action_30hz",
+    "codebase_version": "v2.1",
+    "robot_type": "piper",
+    "episodes": 52,
+    "frames": 31706,
+    "episode_length_min": 500,
+    "episode_length_max": 786,
+    "fps": 30.0,
+    "dt_min": 0.03333282470703125,
+    "dt_median": 0.03333330154418945,
+    "dt_max": 0.03333473205566406,
+    "dt_anomalies_over_1ms": 0,
+    "nonconsecutive_frame_episodes": [],
+    "camera_keys": [
+      "observation.images.global",
+      "observation.images.hand"
+    ],
+    "state_key": "observation.state",
+    "action_key": "action",
+    "state_modality": {
+      "joints": {
+        "original_key": "observation.state",
+        "start": 0,
+        "end": 6,
+        "absolute": true,
+        "dtype": "float32"
+      },
+      "gripper": {
+        "original_key": "observation.state",
+        "start": 6,
+        "end": 7,
+        "absolute": true,
+        "dtype": "float32",
+        "range": [
+          0.0,
+          1.0
+        ]
+      }
+    },
+    "action_modality": {
+      "joints": {
+        "original_key": "action",
+        "start": 0,
+        "end": 6,
+        "absolute": true,
+        "dtype": "float32"
+      },
+      "gripper": {
+        "original_key": "action",
+        "start": 6,
+        "end": 7,
+        "absolute": true,
+        "dtype": "float32",
+        "range": [
+          0.0,
+          1.0
+        ]
+      }
+    },
+    "action_alignment": "action[t] paired with observation[t] per LeRobot row; collection semantics not independently encoded in metadata"
+  },
+  "splits": {
+    "train": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15,
+      20,
+      23,
+      24,
+      25,
+      26,
+      27,
+      28,
+      29,
+      30,
+      31,
+      32,
+      33,
+      35,
+      36,
+      37,
+      38,
+      39,
+      40,
+      41,
+      42,
+      44,
+      45,
+      46,
+      47,
+      48,
+      49,
+      50,
+      51
+    ],
+    "val": [
+      7,
+      16,
+      17,
+      21,
+      43
+    ],
+    "test": [
+      8,
+      18,
+      19,
+      22,
+      34
+    ]
+  },
+  "episodes": [
+    {
+      "episode_id": 0,
+      "frames": 718,
+      "split": "train"
+    },
+    {
+      "episode_id": 1,
+      "frames": 576,
+      "split": "train"
+    },
+    {
+      "episode_id": 2,
+      "frames": 618,
+      "split": "train"
+    },
+    {
+      "episode_id": 3,
+      "frames": 616,
+      "split": "train"
+    },
+    {
+      "episode_id": 4,
+      "frames": 586,
+      "split": "train"
+    },
+    {
+      "episode_id": 5,
+      "frames": 653,
+      "split": "train"
+    },
+    {
+      "episode_id": 6,
+      "frames": 654,
+      "split": "train"
+    },
+    {
+      "episode_id": 7,
+      "frames": 666,
+      "split": "val"
+    },
+    {
+      "episode_id": 8,
+      "frames": 616,
+      "split": "test"
+    },
+    {
+      "episode_id": 9,
+      "frames": 563,
+      "split": "train"
+    },
+    {
+      "episode_id": 10,
+      "frames": 654,
+      "split": "train"
+    },
+    {
+      "episode_id": 11,
+      "frames": 585,
+      "split": "train"
+    },
+    {
+      "episode_id": 12,
+      "frames": 613,
+      "split": "train"
+    },
+    {
+      "episode_id": 13,
+      "frames": 593,
+      "split": "train"
+    },
+    {
+      "episode_id": 14,
+      "frames": 706,
+      "split": "train"
+    },
+    {
+      "episode_id": 15,
+      "frames": 661,
+      "split": "train"
+    },
+    {
+      "episode_id": 16,
+      "frames": 548,
+      "split": "val"
+    },
+    {
+      "episode_id": 17,
+      "frames": 631,
+      "split": "val"
+    },
+    {
+      "episode_id": 18,
+      "frames": 786,
+      "split": "test"
+    },
+    {
+      "episode_id": 19,
+      "frames": 623,
+      "split": "test"
+    },
+    {
+      "episode_id": 20,
+      "frames": 625,
+      "split": "train"
+    },
+    {
+      "episode_id": 21,
+      "frames": 601,
+      "split": "val"
+    },
+    {
+      "episode_id": 22,
+      "frames": 548,
+      "split": "test"
+    },
+    {
+      "episode_id": 23,
+      "frames": 578,
+      "split": "train"
+    },
+    {
+      "episode_id": 24,
+      "frames": 500,
+      "split": "train"
+    },
+    {
+      "episode_id": 25,
+      "frames": 608,
+      "split": "train"
+    },
+    {
+      "episode_id": 26,
+      "frames": 571,
+      "split": "train"
+    },
+    {
+      "episode_id": 27,
+      "frames": 703,
+      "split": "train"
+    },
+    {
+      "episode_id": 28,
+      "frames": 593,
+      "split": "train"
+    },
+    {
+      "episode_id": 29,
+      "frames": 538,
+      "split": "train"
+    },
+    {
+      "episode_id": 30,
+      "frames": 600,
+      "split": "train"
+    },
+    {
+      "episode_id": 31,
+      "frames": 661,
+      "split": "train"
+    },
+    {
+      "episode_id": 32,
+      "frames": 603,
+      "split": "train"
+    },
+    {
+      "episode_id": 33,
+      "frames": 602,
+      "split": "train"
+    },
+    {
+      "episode_id": 34,
+      "frames": 576,
+      "split": "test"
+    },
+    {
+      "episode_id": 35,
+      "frames": 631,
+      "split": "train"
+    },
+    {
+      "episode_id": 36,
+      "frames": 541,
+      "split": "train"
+    },
+    {
+      "episode_id": 37,
+      "frames": 575,
+      "split": "train"
+    },
+    {
+      "episode_id": 38,
+      "frames": 651,
+      "split": "train"
+    },
+    {
+      "episode_id": 39,
+      "frames": 644,
+      "split": "train"
+    },
+    {
+      "episode_id": 40,
+      "frames": 586,
+      "split": "train"
+    },
+    {
+      "episode_id": 41,
+      "frames": 599,
+      "split": "train"
+    },
+    {
+      "episode_id": 42,
+      "frames": 613,
+      "split": "train"
+    },
+    {
+      "episode_id": 43,
+      "frames": 631,
+      "split": "val"
+    },
+    {
+      "episode_id": 44,
+      "frames": 549,
+      "split": "train"
+    },
+    {
+      "episode_id": 45,
+      "frames": 616,
+      "split": "train"
+    },
+    {
+      "episode_id": 46,
+      "frames": 617,
+      "split": "train"
+    },
+    {
+      "episode_id": 47,
+      "frames": 611,
+      "split": "train"
+    },
+    {
+      "episode_id": 48,
+      "frames": 572,
+      "split": "train"
+    },
+    {
+      "episode_id": 49,
+      "frames": 549,
+      "split": "train"
+    },
+    {
+      "episode_id": 50,
+      "frames": 541,
+      "split": "train"
+    },
+    {
+      "episode_id": 51,
+      "frames": 607,
+      "split": "train"
+    }
+  ],
+  "metrics": {
+    "fit_mae": 0.0,
+    "fit_rmse": 0.0,
+    "fit_max": 0.0,
+    "fit_quantiles": {
+      "0.5": 0.0,
+      "0.9": 0.0,
+      "0.95": 0.0,
+      "0.99": 0.0
+    },
+    "additional_quantization_mae": 0.0016609944868832827,
+    "additional_quantization_rmse": 0.0020821390207856894,
+    "additional_quantization_max": 0.003921657335013151,
+    "additional_quantization_quantiles": {
+      "0.5": 0.0016143321990966797,
+      "0.9": 0.0034426748752593994,
+      "0.95": 0.003694087266921997,
+      "0.99": 0.003884613513946533
+    }
+  }
+}

NewModel/v2/sidecars/raw/encoder.json

@@ -0,0 +1 @@
+{"calibration": {"high": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "low": [-1.0, -1.0, -1.0, -1.0, -1.0, -1.0, -1.0]}, "config": {"action_dim": 7, "action_horizon": 30, "vocab_size": 256}, "fixed_fields": "action horizon, sample period, normalization quantiles", "predicted_fields": "30x7 normalized raw actions or ordered bins", "rtc_history_steps": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10], "tokenizer_id": "raw_action_q01_q99_uniform_256_v1", "type": "RawActionSequenceConfig"}

NewModel/v2/sidecars/raw/normalization.json

@@ -0,0 +1,82 @@
+{
+  "state_mean": [
+    -0.06821590662002563,
+    1.3402678966522217,
+    -1.029311180114746,
+    -0.026913858950138092,
+    0.482304185628891,
+    0.12792791426181793,
+    0.19733327627182007
+  ],
+  "state_std": [
+    0.24829290807247162,
+    0.6473751068115234,
+    0.5541877150535583,
+    0.14618821442127228,
+    0.38715362548828125,
+    0.1823183298110962,
+    0.2029987871646881
+  ],
+  "action_q01": [
+    -0.5571745413541794,
+    -0.015232279896736145,
+    -2.0112431049346924,
+    -0.39771774411201477,
+    -0.18397162854671478,
+    -0.3426015079021454,
+    0.0
+  ],
+  "action_q99": [
+    0.41843727231025696,
+    2.1062631607055664,
+    0.0013397021684795618,
+    0.36006876826286316,
+    1.3638149499893188,
+    0.4811461567878723,
+    1.0
+  ],
+  "encoder": {
+    "type": "RawActionSequenceConfig",
+    "config": {
+      "action_horizon": 30,
+      "action_dim": 7,
+      "vocab_size": 256
+    },
+    "tokenizer_id": "raw_action_q01_q99_uniform_256_v1",
+    "calibration": {
+      "low": [
+        -1.0,
+        -1.0,
+        -1.0,
+        -1.0,
+        -1.0,
+        -1.0,
+        -1.0
+      ],
+      "high": [
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0,
+        1.0
+      ]
+    },
+    "predicted_fields": "30x7 normalized raw actions or ordered bins",
+    "fixed_fields": "action horizon, sample period, normalization quantiles",
+    "rtc_history_steps": [
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10
+    ]
+  },
+  "encoder_sha256": "001303ea9a02dbe0b36d787f2d1dbce87bb275ba4c1cff51eb06cdecf62b200e"
+}

NewModel/v2/sidecars/raw/splits.json

@@ -0,0 +1,60 @@
+{
+  "train": [
+    0,
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    9,
+    10,
+    11,
+    12,
+    13,
+    14,
+    15,
+    20,
+    23,
+    24,
+    25,
+    26,
+    27,
+    28,
+    29,
+    30,
+    31,
+    32,
+    33,
+    35,
+    36,
+    37,
+    38,
+    39,
+    40,
+    41,
+    42,
+    44,
+    45,
+    46,
+    47,
+    48,
+    49,
+    50,
+    51
+  ],
+  "val": [
+    7,
+    16,
+    17,
+    21,
+    43
+  ],
+  "test": [
+    8,
+    18,
+    19,
+    22,
+    34
+  ]
+}