A biologically inspired adaptive control architecture based on neural networks for a four-legged walking machine

This paper presents a biologically inspired adaptive control architecture for a four-legged walking machine. In this architecture neural networks are used in two different aspects. First, simple recurrent neural networks are used as coupled neuro-oscillators to represent elementary periodic movements. Second, Radial Basis Functions are employed as state space representation for a Reinforcement Learning component, with which superimposing and coordination of elementary movements are learned. In the development of the presented architecture some results of research on mammalian locomotion are included. The architecture is used to model intralimb coordination of the four-legged walking machine BISAM