huggingface.co web signal

Hugging Face publishes flagship SO-101 arm build guide

TL;DR

  • The SO-101 follower arm uses six STS3215 motors on a 1/345 gear ratio; the leader arm mixes three different gear ratios across its six joints.
  • Parts sourcing and 3D-print files live in TheRobotStudio's SO-ARM100 GitHub repo, not on Hugging Face's own site.
  • Hugging Face frames calibration as load-bearing because it 'allows a neural network trained on one robot to work on another.'

Hugging Face's SO-101 documentation calls the arm "our flagship robot" and walks through the physical build in unusual detail: every M2x6mm screw, every 3-pin cable, every EEPROM write.

The follower arm uses "6x STS3215 motors with 1/345 gearing." The leader is a different animal. It uses three differently geared motors "to make sure it can both sustain its own weight and it can be moved without requiring much force" — 1/191 at the base and elbow, 1/345 at the shoulder lift, and 1/147 through the wrist flex, wrist roll and gripper.

Parts and print files live in a separate repo. The bill of materials and 3D-print instructions are hosted at TheRobotStudio/SO-ARM100 on GitHub, and the LeRobot install pulls in the Feetech SDK to configure motor IDs and baudrates over USB.

The final step is calibration. The docs are blunt about why it matters: it "allows a neural network trained on one robot to work on another." That is the whole point of a documented reference arm — a build any policy researcher can reproduce, screw for screw.

Shared on Bluesky by 1 AI expert