Abstract
In this paper, a biomimetic control architecture suitable for multi-legged robots is presented, which is inspired by decision and control behaviors of reptiles. The architecture consists of four functional blocks: information sensing, communication, behavior generation and joint driving. The behavior models are classified into five types according to intelligence level, and in-between two successive layers, a learning and evolution mechanism is designed so that a deliberate decision for a reflex event that is generated in the higher layer can be memorized in the lower layer and be re-invoked by the lower layer directly without rethinking when the same event is triggered. The presented biomimetic control has not only the deliberateness of acquired intelligence, but also the innate flexibility. Furthermore, a physical quadruped robotic embodiment and its virtual prototype are constructed for experimental tests and simulations of the proposed control architecture. The effectiveness of the proposed control method is validated by simulations and experiments.