How to do Bicycle Acrobatics with RL?

Bicycle robots are incredibly fast but hard to control. In our recent work, we show that Reinforcement Learning (RL) can solve this issue and unlock a diverse repertoire of dynamic acrobatic behaviors.

Our Ultra Mobility Vehicle (UMV) autonomously learns front and lateral table jumps, front flips, kip-ups and kip-downs, wheelies, bunny hops, three-point turns, and long-horizon compositions of these skills through a policy orchestrator; all with RL.

Looking forward to hearing your thoughts! For more information, please visit this page.