SO-100 Experiments (2025)

In October 2024, Hugging Face introduced the SO-100 robot arm (GitHub repository, LeRobot documentation), shortly after the internet was filled with demos of these little robot arms doing all kinds of things.

The SO-100 is cool because it is a fully open-source design, with a relatively cheap BOM and components you can 3D print at home. There is enough of a software framework that just about anyone can get going and “teach” the arm to do something. What I found most fascinating, though, was that the SO-100 allowed tele-operation out of the box—or more specifically, that the usual setup pairs a follower arm with a leader arm.

This configuration is close to what I consider a modern puppet. A person can operate the leader arm, and suddenly the follower arm appears like a conscious robot doing something smart. Of course the idea is to train a neural network to actually control the arm, but just as many saw tele-operation as a way to cheat into a cool demo, I thought tele-operation unlocked a huge opportunity to puppeteer a robot and create content that inspired people. Somehow this was a very basic concept that had not crossed my mind before, so I was eager to play with it.

I still have the SO-100 arms with me and I still play with them occasionally, so I may add more puppeteering experiments here over time. Overall the arms are a really cool platform which I am sure I will use in other projects. If you want to build your own set, be sure to check out Hugging Face’s newer design, the SO-101; you can even put wheels on these things with LeKiwi.