README.md

metadata

license: mit
task_categories:
  - robotics
tags:
  - LeRobot
  - robotics
  - imitation-learning
  - bimanual-manipulation
  - fold-clothes
configs:
  - config_name: foldcloths_yaleon
    data_files: lerobot_foldcloths_HF_Yaleon/data/*/*.parquet
  - config_name: foldcloths_ganrao_yaleon
    data_files: lerobot_foldcloths_ganrao_HF_Yaleon/data/*/*.parquet

This dataset was created using LeRobot.

Dataset Description

• License: MIT.
• Task Objective: Fold clothes using a dual-arm robot from visual observations.
• Operational Objects: cloth and clothing items.
• Recording Frequency: 30 Hz.
• Robot Type: tron2_v5.
• End Effector: left and right gripper channels in the robot state/action vectors.
• Dual-Arm Operation: Yes; left and right arm joints, grippers, and end-effector poses are recorded.
• Image Observations: included as embedded image bytes in the parquet rows.
• Camera Positions: cam_high, cam_left_wrist, cam_right_wrist.
• Total Episodes: 1,611.
• Total Frames: 1,466,001.
• Upload Scope: two LeRobot fold-cloth subsets with image observations, robot state, action, and end-effector pose fields preserved.
• Intervention Splits: foldcloths_yaleon contains episodes without human intervention; foldcloths_ganrao_yaleon contains episodes with human intervention.
• Number of rows: 1,466,001 frame rows across both subsets.
• foldcloths_ganrao_yaleon Size: 126.0 GB.

Task Preview

foldcloths_yaleon: no human intervention

High-camera preview sampled from the no-human-intervention fold-clothes split.

foldcloths_ganrao_yaleon: human intervention

High-camera preview sampled from the human-intervention fold-clothes split.

Subsets

Subset	Task	Episodes	Frames	Data files	Notes
lerobot_foldcloths_HF_Yaleon	fold clothes	970	695,939	lerobot_foldcloths_HF_Yaleon/data/chunk-000/episode_*.parquet	No human intervention.
lerobot_foldcloths_ganrao_HF_Yaleon	fold clothes	641	770,062	lerobot_foldcloths_ganrao_HF_Yaleon/data/chunk-000/episode_*.parquet	Human intervention.

Files

• assets/foldcloths_no_intervention_preview.mp4: compact preview from the no-human-intervention split.
• assets/foldcloths_human_intervention_preview.mp4: compact preview from the human-intervention split.
• lerobot_foldcloths_HF_Yaleon/data/chunk-000/episode_*.parquet: first fold-cloth subset with image observations and robot fields.
• lerobot_foldcloths_HF_Yaleon/meta/*.json*: task, episode, statistics, and dataset metadata for the first subset.
• lerobot_foldcloths_ganrao_HF_Yaleon/data/chunk-000/episode_*.parquet: ganrao fold-cloth subset with the same LeRobot schema.
• lerobot_foldcloths_ganrao_HF_Yaleon/meta/*.json*: task, episode, statistics, and dataset metadata for the ganrao subset.

Feature Summary

Field group	Details
Images	observation.images.cam_high, observation.images.cam_left_wrist, observation.images.cam_right_wrist; each image has shape 480 x 640 x 3.
State	observation.state, float32 vector with 16 channels.
Action	action, float32 vector with 16 channels.
End-effector poses	observation.ee_pose_left, observation.ee_pose_right, observation.ee_pose_cmd_left, observation.ee_pose_cmd_right; each is a 7D pose vector.
Indexing	frame_index, timestamp, episode_index, index, and task_index.

Metadata

Both subsets use LeRobot codebase version v2.1 and the tron2_v5 robot/camera schema. The first subset reports 695,939 frames over 970 episodes; the ganrao subset reports 770,062 frames over 641 episodes.