Grabette: an open system to record robot-manipulation data

Grabette is an open-source system designed to simplify the process of recording robot manipulation data. By using a handheld gripper equipped with cameras and motion sensors, users can capture demonstration videos and convert them into robot-ready datasets. This eliminates the need for expensive robots, lengthy data collection sessions, and complex hardware setups.

The project was inspired by Stanford's Universal Manipulation Interface (UMI), which demonstrated the feasibility of this approach. However, Grabette goes a step further by making the system easy to use and integrate into existing robotic ecosystems. It is compatible with popular platforms like LeRobot for dataset management and the Hugging Face Hub for data sharing.

Grabette consists of a Raspberry Pi, a standard camera, an OAK-D depth camera, and magnetic encoders. It records a robot's 6-DoF pose and gripper state during demonstrations, resulting in a standardized dataset format that can be used with various robot learning methods. The system is designed to be modular and compatible with any robotic arm, making it accessible for anyone to build and contribute to the open dataset.

By enabling users to record their own manipulation tasks and share the resulting data, Grabette aims to foster a collaborative effort in robot learning. The open ecosystem encourages users to build upon the existing framework, expanding the dataset and reducing the barrier to entry for robot learning research.