LogicDeposit/Adamu_260331_RgbCube

AdamU 260331 RgbCube Dataset

Dual-arm manipulation dataset collected on the AdamU humanoid robot using 3 RGB cameras (head + dual wrist). Stored in LeRobot v3.0 format.

Dataset Details

Property	Value
Total Episodes	105
Valid Training Episodes	0–99 (100 episodes)
Total Frames	49,415
FPS	30
Data Format	Parquet + MP4 (AV1 codec)

Note: Episodes 100–104 are not valid training data. Episodes 103/104 contain the robot reaching and leaving its preparation pose. Use the filtering flag below when training.

Cameras

Camera	Resolution	Type
observation.images.head	640×360	RGB
observation.images.wrist_left	640×480	RGB
observation.images.wrist_right	640×480	RGB

State & Action Space

31-DOF joint positions for the full upper body (waist, neck, dual arms with 5-finger hands):

waistRoll, waistPitch, waistYaw,
neckYaw, neckPitch,
shoulderPitch_Left, shoulderRoll_Left, shoulderYaw_Left, elbow_Left,
wristYaw_Left, wristPitch_Left, wristRoll_Left,
shoulderPitch_Right, shoulderRoll_Right, shoulderYaw_Right, elbow_Right,
wristYaw_Right, wristPitch_Right, wristRoll_Right,
hand_pinky_Left, hand_ring_Left, hand_middle_Left, hand_index_Left,
hand_thumb_1_Left, hand_thumb_2_Left,
hand_pinky_Right, hand_ring_Right, hand_middle_Right, hand_index_Right,
hand_thumb_1_Right, hand_thumb_2_Right

Training Usage

When training with LeRobot / LogicMimic, filter to the valid episodes (0–99):

--dataset.episodes=$(python3 -c 'print("[" + ",".join(str(i) for i in range(100)) + "]")')

File Structure

data/chunk-000/          # Parquet data files
meta/
  info.json              # Dataset metadata
  stats.json             # Feature statistics
  tasks.parquet          # Task descriptions
  episodes/chunk-000/    # Episode metadata
videos/
  observation.images.head/chunk-000/         # Head camera videos
  observation.images.wrist_left/chunk-000/   # Left wrist camera videos
  observation.images.wrist_right/chunk-000/  # Right wrist camera videos