Hugging Face's logo

Datasets:
THULab
/
koch_bimanual_grasp_3

Tasks:
Robotics
Modalities:
Time-series
Size:
10K<n<100K
Tags:
LeRobot
koch_bimanual_3
robotics
koch
bimanual
imitation-learning
License:
Dataset card Files
xet
 Community
Episodes Preview Koch Visualizer
50 episodes · 30 fps

Koch Bimanual Grasp 3 (TsFile)

Apache TsFile version of Yuanzhu/koch_bimanual_grasp_3.

Overview

Teleoperation demonstrations of a grasping task ("Grasp3.") recorded on a Koch bimanual robot — two low-cost 6-DOF arms operated together. Each episode is one demonstration with per-frame joint positions (observed state) and the commanded joint actions for both arms, captured at 30 fps.

  • Robot: Koch bimanual (left + right 6-DOF arms, robot_type = koch_bimanual).
  • Episodes: 50 demonstrations of one task ("Grasp3.").
  • Frames: 16,763 timesteps total, recorded at 30 fps.
  • Source format: LeRobot v2.0.

Schema (TsFile structure)

All 50 episodes are stored in a single .tsfile using the TsFile table model.

  • Time (INT64, milliseconds) — per-frame timestamp, restarts at 0 for each episode (round(timestamp * 1000); ~33 ms apart at 30 fps).
  • episode_index (TAG) — demonstration id (0–49). Query a single episode with WHERE episode_index=0.
  • task_index (TAG) — task id (always 0; the dataset has one task).
  • frame_index (FIELD, INT64) — frame number within the episode.
  • sample_index (FIELD, INT64) — the source global index column.
  • observation_state_0 .. observation_state_11 (FIELD, FLOAT) — 12-DOF observed joint positions: left arm (0–5) then right arm (6–11), each as shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper.
  • action_0 .. action_11 (FIELD, FLOAT) — 12-DOF commanded joint positions, same left-then-right ordering as the state.

The source timestamp column is dropped because it equals Time / 1000. Camera videos are not included in this repository (see below).

Usage

Read the .tsfile file with the Apache TsFile Java or Python SDK.

# pip install tsfile
from tsfile import TsFileReader
reader = TsFileReader("data/koch_bimanual_grasp_3.tsfile")
# query one episode: WHERE episode_index=0

Videos

The original dataset records three camera streams per episode (observation.images.top_camera, observation.images.side_camera, observation.images.hand_camera). These MP4 videos are not included here — they are not time-series data. Retrieve them from the original dataset under Yuanzhu/koch_bimanual_grasp_3/videos.

Source & license

Downloads last month
63
Total file size:
484 kB