provenance/source.json

{
  "benchmark": "DynamicVLA",
  "repo_id": "hzxie/DOM",
  "revision": "bbf6baa15f7253746d271a6e958e037afa537b1e",
  "code_repository": "hzxie/DynamicVLA",
  "code_revision": "434dc7105cf6d559018d15af5fecbcdad44f036e",
  "note": "DOM has 3 task families, not 3 task_index values: 140,932 structured instruction IDs and 207,306 episodes. This subset selects 6 demonstrations per family (18 total), preserving actual task_index/episode_index and structured instructions. All 3 published cameras are single-arm cameras: opposite, wrist, side, not top/left/right arms.",
  "date": "2026-10-03",
  "video_policy": "Complete selected episodes, converted to browser-friendly H.264 MP4 at original 25 FPS. No generated simulation rollouts.",
  "actual_episode_score": "Not published in released LeRobot features; null, not assumed successful.",
  "episode_selection": "Two episodes from three spaced metadata windows for pick/place, six spaced long-horizon scenarios.",
  "camera_order": [
    "opst_cam",
    "wrist_cam",
    "side_cam"
  ]
}