Hybrid Control for Learning Motor Skills

We develop a hybrid control approach for robot learning based on combining learned predictive models with experience-based state-action policy mappings to improve the capabilities of robotic systems. Predictive provide an understanding task and physics (which improves sample-efficiency), while are treated as “muscle memory” that encode favorable actions experiences override planned actions. Hybrid tools used create algorithmic learning. is presented method efficiently motor skills by systematically improving performance policies. A deterministic variation derived extended into stochastic implementation relaxes some key assumptions in original derivation. Each tested methods (where interacts environment gain experience) well imitation experience provided through demonstrations environment). The results show our capable sample-efficiency variety experimental domains.