Continual Learning of Koopman dynamics for Linear Control of Legged Robots

Controlling high-dimensional legged robots like humanoids and quadrupeds has long been a challenge due to their complex nonlinear dynamics. Our Incremental Koopman algorithm bridges the gap by transforming nonlinear systems into scalable, accurate linear models, enabling efficient control across diverse terrains.
Project website: intelligent-co...
Paper link: arxiv.org/abs/...