A humanoid robot balancing on a single rolling sphere just mastered zero-shot transfer from a simulation to the real world.
April 29, 2026
Original Paper
asRoBallet: Closing the Sim2Real Gap via Friction-Aware Reinforcement Learning for Underactuated Spherical Dynamics
arXiv · 2604.24916
The Takeaway
Ballbots face extreme stability challenges because of unpredictable friction and underactuated dynamics. This reinforcement learning framework incorporates friction-aware training to bridge the gap between digital models and physical hardware. The robot successfully transitioned from its virtual training environment to real-world movement without any manual tuning on the physical device. Achieving this level of control on a spherical base opens the door for more agile and compact service robots. The success proves that complex physical dynamics are no longer a barrier for reinforcement learning when friction is modeled correctly.
From the abstract
We introduce asRoBallet, to the best of our knowledge, the first successful deployment of reinforcement learning (RL) on a humanoid ballbot hardware. Historically, ballbots have served as a canonical benchmark for underactuated and nonholonomic control, which are characterized by a reality gap in complex friction models for wheel-sphere-ground interactions. While current literature demonstrates successful handling of 3D balancing with LQR and MPC, transitioning to actual hardware for a humanoid