exidekat/chessbot-lerobot

ChessBot LeRobot Dataset

Vision-language-action (VLA) training dataset for chess piece manipulation with SO-100 robot arm.

Dataset Description

This dataset contains teleoperated demonstrations of a robot arm manipulating chess pieces on a physical chessboard. Each episode captures:

• Global camera: Overhead view of the chessboard (1280x720)
• Gripper camera: Wrist-mounted camera view (224x224)
• Robot state: 6-DOF joint positions
• Actions: Target joint positions

Intended Use

Fine-tuning VLA models (PI0, SmolVLA) for chess manipulation tasks with natural language instructions.

Dataset Structure

lerobot_episodes/
  data/           # Parquet files with state/action data
  videos/         # MP4 videos per camera
  meta/           # Episode metadata and statistics
  images/         # Decoded frames (optional)

Citation

If you use this dataset, please cite the ChessBot project.