Grabette: an open system to record robot-manipulation data
huggingface ·
The article introduces Grabette, an open-source, low-cost handheld system from Pollen Robotics for recording robot manipulation data without requiring a robot. Inspired by Stanford's UMI device, Grabette uses two cameras and sensors to capture demonstrations from a human hand, then processes them through a browser-based pipeline into standard LeRobot datasets on the Hugging Face Hub. The system includes a companion motorized gripper called Gripette for executing learned tasks on robot arms, and all hardware and software are open source. The authors aim to solve the data bottleneck in robot learning by making data collection easy enough for anyone to contribute to a large, collaborative dataset. As an end-to-end example, they trained a diffusion policy on 200 demonstrations and evaluated it on an OpenArm robot with Gripette. Future work includes a head-mounted device named Casquette, but the main goal is community participation in building the open dataset.
huggingface