Grabette: Open Low-Cost System for Recording Robot Manipulation Data
Pollen Robotics
Hugging Face
Hugging Face and Pollen Robotics released Grabette, an open, low-cost handheld gripper with cameras for recording robot manipulation demonstrations. Data is processed via LeRobot and shared on the Hugging Face Hub; the system is robot-agnostic and aims to create a large collaborative dataset.
Grabette is an open, low-cost system for recording manipulation data, announced by the Hugging Face team and Pollen Robotics. Inspired by Stanford's Universal Manipulation Interface (UMI), it consists of a handheld gripper with two cameras: a wide fisheye for context and an RGBD camera for 6-DoF tracking. Recordings are processed via SLAM and stored as LeRobot datasets on the Hugging Face Hub. A robotic counterpart called Gripette executes learned movements. The system is robot-agnostic, uses off-the-shelf components (Raspberry Pi, Pi camera, OAK-D depth camera, magnetic encoders), and the entire pipeline runs from a browser with no installation. A complete example with 200 demonstrations is provided. The goal is to enable anyone to contribute to a large, open manipulation dataset. Future development includes Casquette, a head-mounted POV device.
- Сокращения
- SLAM = Simultaneous Localization and Mapping — одновременная локализация и построение карты
- 6-DoF = Six Degrees of Freedom — шесть степеней свободы
- RGBD = Red Green Blue Depth — цветное изображение с глубиной
- IMU = Inertial Measurement Unit — инерциальный измерительный блок
- POV = Point of View — вид от первого лица
Source: Hugging Face blog —
original