Learning to locomote: Action sequences and switching boundaries

This paper presents a hybrid control strategy for learning the switching boundaries between primitive controllers that maximize the translational movements of complex locomoting systems. Through this abstraction, the algorithm learns an optimal action for each boundary condition instead of one for each discretized state and action of the system, as is typically in the case of machine learning. This hybridification of the problem mitigates the “curse of dimensionality”. The effectiveness of the learning algorithm is demonstrated on both a simulated system and on a physical robotic system. In both cases, the algorithm is able to learn the hybrid control strategy that maximizes the forward translational movement of the system without the need for human involvement.