README.md

metadata

license: cc-by-4.0
tags:
  - lerobot
  - robotics
  - so101
  - manipulation
  - pick-and-place
  - imitation-learning
  - robot-learning
  - h264
  - corl2026
  - ma-pipeline
pretty_name: SO101 Pick-and-Place MA 100 Episodes
size_categories:
  - n<1K
task_categories:
  - robotics

SO101-pnp_ma_100epi

This dataset contains 100 successful SO-101 robot pick-and-place episodes collected with the MA pipeline.

Task

Instruction: Pick up the red block and place it on the blue dish.

Dataset Summary

• Episodes: 100
• Frames: 98,187
• FPS: 30
• Robot: SO-101 follower
• Cameras:
  • observation.images.top at 640x480
  • observation.images.left_wrist at 640x480
• Video codec: H.264 (yuv420p)
• Dataset format: LeRobot v3.0-style local dataset structure

Features

The dataset includes robot state/action, top and wrist RGB observations, end-effector pose, gripper state, skill annotations, subtask annotations, timestamps, frame indices, and task indices.

Collection Notes

Episode 58 was re-collected after manual inspection. The upload version was cleaned so the parquet data follows the metadata episode index ranges and all videos are re-encoded to H.264.

Intended Use

This dataset is intended for robot imitation learning, policy learning, and evaluation of pick-and-place behavior with SO-101.