SO-101 robot arm
The SO-101 is a low-cost, open source robot arm designed for robot learning and the flagship robot in Hugging Face's LeRobot documentation. It is usually built as a pair: a leader arm that a person moves by hand and a follower arm that mirrors it. More than half of the community LeRobot datasets on the Hugging Face Hub were recorded on SO-100 and SO-101 arms.
In this article
Building visual or physical AI?
Let's talk.
What is the SO-101?
The SO-101 is a six-motor robot arm, five joints plus a gripper, made from 3D-printed parts and off-the-shelf servo motors. It is the successor to the SO-100, and its design files and bill of materials are open source, so anyone can source the parts and build one. That low cost is the point: an SO-101 puts real-world robot learning within reach of students, hobbyists, and small labs.
Most SO-101 setups use two arms. The follower arm performs the task. The leader arm uses differently geared motors so a person can move it easily by hand, and LeRobot copies its joint positions to the follower in real time. While the operator works, LeRobot records the follower's joint positions, the commanded actions, and camera frames as episodes in a LeRobot dataset.
Key takeaways
- The SO-101 is an open source, low-cost robot arm used throughout the LeRobot ecosystem.
- It is usually built as a leader and follower pair for recording teleoperated demonstrations.
- SO-100 and SO-101 arms recorded more than half of the community LeRobot datasets on the Hugging Face Hub.
How it works
Building an SO-101 involves printing the parts, setting each servo motor's ID, assembling the arm, and calibrating it so the leader and follower report the same joint positions in the same physical pose. Calibration matters for data quality as well as control, because a policy trained on one arm only transfers to another if both arms agree on what each joint position means. Once calibrated, the pair can record demonstrations, replay recorded episodes, and run trained policies on the follower.
Why it matters
The SO-101 matters because inexpensive, identical hardware is what made community robot data possible. Thousands of people recording on the same arm produce datasets that can be combined, and SO-100 community datasets were used to pretrain SmolVLA. The same popularity makes consistency the main challenge: camera placement, naming, and task descriptions vary from one contributor to the next.
Frequently asked questions
How is the SO-101 different from the SO-100?
The SO-101 is the successor to the SO-100. Its leader arm uses differently geared motors, so a person can move it easily by hand while it still holds its own weight.
What can you train with SO-101 data?
SO-101 datasets are commonly used to train imitation learning policies such as ACT and Diffusion Policy, and to fine-tune vision-language-action models such as SmolVLA, all through LeRobot.
Related terms
Jesse Mostipak
Director of Growth
Jesse Mostipak is the Director of Growth at Voxel51, where the work is helping humans find and trust what the brand knows, and teaching the Google knowledge graph and the LLMs answering on their behalf to do the same. That question, how knowledge gets built inside a system, is one Jesse has been chasing for years. Earlier versions of it ran through a New York City high school science classroom, data science and machine learning, and developer relations at Kaggle, Posit (formerly RStudio), and Baseten. The answer doesn't change much depending on whether the learner is a teenager, a software engineer, or a knowledge graph. Jesse holds a Master's in Education from CUNY Hunter College. LinkedIn
See all articles by Jesse Mostipak
Last updated October 8, 2026
Building visual or physical AI?
Let's talk.
© 2026 Voxel51 All Rights Reserved