Teaching robots how to maintain their balance takes work. We have already seen Optimus and other robots performing challenging moves. Some of them can even do forward and sideflips. HuB explores the idea of teaching humanoid robots extreme balance. It enables humanoids to perform quasi-static balance tasks with high stability.
With this approach, robots will be able to do deep squats, single leg stands, Bruce Lee’s kick, and high knees. As the researchers explain:
“HuB (Humanoid Balance) is a unified framework that integrates reference motion refinement, balance-aware policy learning, and sim-to-real robustness training, with each component targeting a specific challenge.”
[HT]