Embodied AI Hackathon Winners Announced
Two LeRobot SO-ARM100 arms work together for hackathon pick-and-place tasks.
This past weekend (March 22nd and 23rd), Circuit Launch hosted the Embodied AI Hackathon in Mountain View, California. Participants were invited to explore Seeed Studio's SO-ARM100 robotic arm kit with Hugging Face's LeRobot framework and NVIDIA's Jetson hardware and Issac Lab for teleoperation and simulation. Prizes included SO-ARM100 arms, NVIDIA hardware.
The hackathon winner was Team Firebreathing Rubber Duckies, which leveraged two LeRobot SO-ARM100 robots and trained them for collaborative pick-and-place tasks. This meant first setting up the two robot arms, then manually demonstrating how pick-and-place tasks should be performed. Each robot was equipped with an Arducam for visual processing, while an external Logitech HD Pro webcam was used to record the overall process.
Multiple vision language action (VLA) models were used to teach the robots to perform collaborative pick-and-place actions (e.g. play nicely together), moving objects from one bowl to another. Models implemented include NVIDIA’s Groot N1, ACT, and Pi0.
Team Firebreathing Robot Duckies notes that:
This initiative aims to inspire researchers, engineers, and enthusiasts to explore and expand upon the project, unlocking new possibilities in robotics and artificial intelligence.
By bridging human intuition with robotic precision, this project paves the way for the future of teleoperated and autonomous robotic manipulation, transforming industries from manufacturing to service automation.
To encourage further development, the demonstration dataset was uploaded to the Hugging Face AI collaboration platform, while code and implementation details are available on GitHub. It’s neat to see what a dedicated team, with the proper hardware and dedication, can do in just two days of work. Check some of the experimentation in the video below!
