Grabette: Open System for Recording Robot Manipulation Data

Pollen Robotics has released Grabette, an open-source, low-cost system for recording robotic manipulation data. The device is inspired by Stanford University's Universal Manipulation Interface (UMI). Grabette is a handheld gripper with two cameras: a wide-angle fisheye camera for contextual views and an RGBD camera for spatial position tracking (6 degrees of freedom). Data is recorded on a Raspberry Pi, synchronized with gripper encoder readings, and processed in a browser via a web interface without software installation. The output is a standard LeRobot dataset, ready for model training. Alongside this, Gripette has been introduced—a robotic end effector compatible with various manipulators. The entire system is fully open-source, and components are available for order. Grabette already includes an example training pipeline with 200 demonstrations. The developers encourage the community to collect and upload datasets to Hugging Face Hub to collectively address the shortage of high-quality data for robot training.