Multilayered learning architecture applied to humanoid locomotion

This paper presents a new learning architecture dealing with humanoid stability control based on a pluridisciplinary bibliographical study. It brings a new insight inspired from recent studies on the central nervous system (CNS) behaviour in the locomotion process. A general architecture of the reproduction of the CNS main functions is proposed with the corresponding areas involved and the associated learning process techniques. The coordination is then developed using internal models concepts and introducing a multi-layered/multi-level learning structure able to deal smoothly with the change of gaits.