What a way to end my last course at @ETH_en .
Our Robot Learning final project got picked to be presented live to @ylecun , Jitendra Malik, and many well known people in academia.
We built a robot arm you can just talk to. Put three faces on the table, say out loud "place the coke on" one of them, and it does it. We never trained it on those faces. Vision and language stay frozen, and we only trained the arm on a tiny bit of real data.
The part I am proudest of is that we also built a way to catch the model lying, to check if it actually listens to you or just fakes it.
None of this would have been possible without the open source work from Hugging Face. We built directly on LeRobot and SmolVLA, and it is genuinely incredible what @ClemDelangue and the team are putting in the hands of students and researchers. Open robotics is real and it matters.
Huge thanks to @oier_mees, Liam Achenbach, and Carl Brander. Best course to end the Masters degree.