Nov 2, 2026
Unitree G1 lands a kickflip with no cameras, just balance
Researchers at TU Darmstadt IAS and DFKI taught a Unitree G1 humanoid eight skateboard skills, kickflips included, using only joint encoders and an IMU. No cameras, no touch sensors.
Sir Toastimir
Video · TU Darmstadt IAS + DFKI · @aditya_bhatt
A Unitree G1 humanoid lands a kickflip, and it does it blind. The project, The Blind Skateboarder, comes from TU Darmstadt's Intelligent Autonomous Systems lab (IAS) and DFKI.
What's new
The team says that, to their knowledge, it is the first humanoid system to do skateboard tricks. That's the team's own claim. The skill library covers eight skills on the real robot: mounting, pushing, dismounting, manuals, ollies, shove-its, kickflips and walk-the-dog.
How it works without seeing
The robot has no camera view of the board, no tracking markers, no touch or force sensors. It only uses its own body sense: joint encoders and an IMU (a balance sensor). A "funneling" training curriculum teaches each trick to first feel for the right foot position on the board through contact, then execute.
Caveats
This is a lab demo on a flat indoor floor. The footage is from the research team (shared on X by @aditya_bhatt); we credit it and link to the source.
What trick should this robot learn next?
Links
Get your daily slice of AI
One short email with the best new AI tools, robots and AI & tech stock news. Free. Unsubscribe anytime.
We only use your email to send the newsletter.