results/omni_finetune/dataset_manifest.json

{
  "run_id": "xperience10m_qwen3_omni_32ep_dataset",
  "dataset_path": "/path/to/ropedia_workspace/ropedia-episode-task-suite/results/omni_finetune/xperience10m_qwen3_omni_32ep_dataset/dataset.jsonl",
  "num_samples": 128,
  "num_episodes": 1,
  "split_counts": {
    "train": 128
  },
  "label_counts": {
    "Close bottle cap": 9,
    "Pick up kettle": 8,
    "Position kettle to pour": 8,
    "Move kettle": 8,
    "Hold coffee carafe": 8,
    "Grasp coffee scoop": 8,
    "Transfer coffee to dripper": 8,
    "Hold gooseneck kettle": 8,
    "Lift gooseneck kettle": 8,
    "Move kettle away": 8,
    "Wait/Prepare for pouring": 8,
    "Pour coffee": 8,
    "Grasp gooseneck kettle": 6,
    "Place kettle on table": 6,
    "Pick up white bottle": 6,
    "Pour liquid from white bottle": 6,
    "Place item on table": 6,
    "Pour milk into coffee": 1
  },
  "action_options": [
    "Close bottle cap",
    "Grasp coffee scoop",
    "Grasp gooseneck kettle",
    "Hold coffee carafe",
    "Hold gooseneck kettle",
    "Lift gooseneck kettle",
    "Move kettle",
    "Move kettle away",
    "Pick up kettle",
    "Pick up white bottle",
    "Place item on table",
    "Place kettle on table",
    "Position kettle to pour",
    "Pour coffee",
    "Pour liquid from white bottle",
    "Pour milk into coffee",
    "Transfer coffee to dripper",
    "Wait/Prepare for pouring"
  ],
  "subtask_options": [
    "Handle gooseneck kettle",
    "Lift gooseneck kettle",
    "Move kettle",
    "Pick up and position kettle",
    "Pour and close white bottle",
    "Pour coffee",
    "Pour milk into coffee",
    "Prepare coffee equipment and scoop grounds",
    "Prepare for pouring",
    "Set down kettle and retrieve white bottle",
    "Transfer coffee grounds to dripper"
  ],
  "clip_policy": {
    "label_window_frames": 20,
    "qwen_context_frames": 120,
    "max_video_frames": 16,
    "audio_span": "same_as_video_context",
    "mosaic": "2x3 multi-camera grid"
  },
  "feature_manifest": [
    {
      "name": "hand_left_joints",
      "start": 0,
      "end": 441,
      "dim": 441
    },
    {
      "name": "hand_right_joints",
      "start": 441,
      "end": 882,
      "dim": 441
    },
    {
      "name": "body_joints",
      "start": 882,
      "end": 1974,
      "dim": 1092
    },
    {
      "name": "body_contacts",
      "start": 1974,
      "end": 2121,
      "dim": 147
    },
    {
      "name": "camera_translation",
      "start": 2121,
      "end": 2142,
      "dim": 21
    },
    {
      "name": "camera_rotation_matrix",
      "start": 2142,
      "end": 2205,
      "dim": 63
    },
    {
      "name": "imu_accel_gyro",
      "start": 2205,
      "end": 2247,
      "dim": 42
    },
    {
      "name": "depth_confidence",
      "start": 2247,
      "end": 3227,
      "dim": 980
    },
    {
      "name": "caption_objects_interaction_text",
      "start": 3227,
      "end": 4123,
      "dim": 896
    },
    {
      "name": "slam_point_cloud",
      "start": 4123,
      "end": 4145,
      "dim": 22
    },
    {
      "name": "calibration",
      "start": 4145,
      "end": 4262,
      "dim": 117
    }
  ],
  "available_modalities": [
    {
      "episode_id": "xperience-10m-sample",
      "modalities": [
        {
          "modality": "depth_confidence",
          "shape": [
            5821,
            140
          ]
        },
        {
          "modality": "caption_text",
          "shape": [
            5821,
            128
          ],
          "fields": "objects,interaction"
        },
        {
          "modality": "slam_point_cloud_static",
          "shape": [
            22
          ]
        },
        {
          "modality": "calibration_static",
          "shape": [
            117
          ]
        }
      ]
    }
  ],
  "notes": [
    "Assistant answers are strict JSON for episode understanding, not robot-control policies.",
    "Sensor features are stored as NPZ pointers; raw annotation.hdf5 is not copied into the dataset records."
  ]
}