Open Robot Data Collection System Grabette Unveiled

To address the data scarcity problem that bottlenecks robot learning, the Grabette system has been unveiled, which collects data using human hands and a gripper instead of expensive robot teleoperation.

Key Features and Components:

• Grabette (Handheld Device): A handheld device equipped with a camera, IMU, and gripper that converts the user's movements into 6-DoF trajectories, automatically generating datasets for robot learning. (BOM cost approximately €490)
• Gripette (Robot End-effector): A motor-driven gripper designed to let motions learned via Grabette be executed on an actual robot arm. (BOM cost approximately €120)
• Open Ecosystem: Integrates with Hugging Face's LeRobot datasets, and the data processing pipeline can be run directly in the browser without any separate installation.

The system was inspired by Stanford's UMI (Universal Manipulation Interface) and aims for collaborative dataset building, where anyone can generate and share high-quality manipulation data at low cost.