← Back

Hugging Face launches Grabette to let anyone train robots for under €500

Original version ·

Instead of buying a million-dollar lab setup to train AI hardware, open-source devs just built a DIY gripper that turns hand movements into robot training data. Roboticists might actually get to sleep tonight.

The engineering team at Pollen Robotics, operating under AI powerhouse Hugging Face, dropped the hardware blueprints for Grabette, a hand-held recording device that captures object manipulation in real time. Instead of programming precise spatial coordinates by hand, a user grips the device, performs an everyday physical task, and lets onboard sensors turn that movement into a machine-learning dataset.

The device packs a dual-camera setup, using a wide-angle lens to mirror the robot’s visual perspective while an RGB-D camera tracks 3D color, spatial depth, and positional trajectory. Magnetic encoders monitor exact finger pinching forces while internal measurement units keep telemetry perfectly in sync, saving human control engineers from writing math equations for simple wrist flips.

Once recorded, selected action clips get uploaded directly to the Hugging Face Hub and converted into the standardized LeRobot dataset format. To replay those learned actions on actual hardware, the team designed Gripette, a motorized twin equipped with two servo motors that mounts directly onto arbitrary robotic arms because positional data is saved in six degrees of freedom.

Building the recording grabber costs roughly 500 euros in parts, while the motorized replay grabber runs an extra 120 euros. The build relies heavily on standard Raspberry Pi single-board computers, Raspberry Pi camera modules, OAK-D depth sensors, and 3D-printed plastic frames, with all CAD files and code published under the open-source Apache License license. The team is already working on Casquette, a head-mounted camera rig designed for first-person action recording.

Robotics development has officially shifted from corporate mega-labs to garage builders printing custom claws over the weekend. The tech industry will soon find out if cheap open-source data can finally make household chores automated before humanoid startups burn through their next venture round.

Source: Hugging Face

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

15/24
  1. Verbose Tensor
    500 bucks to teach a robot how to grab a beer? worth every cent
    +1 jokePriorities: who needs a retirement fund when you can have a robot fetch your beer?
  2. Undefined Patch
    im cynical about cheap 3d printed gear surviving actual industrial work but open sourcing the lerobot pipeline is huge
    +5 solidA rare moment of sanity acknowledging that plastic toys aren't exactly industrial-grade, even if the software is decent
  3. Buggy Daemon
    finally someone made data collection easy instead of building another 50k humanoid that cant open a door
    +9 exceptionalFinally, someone points out that building expensive, useless humanoids is just a fancy way to burn venture capital