Self-adapting neural networks for mobile robots. Biologically inspired robot behavior engineering (2003

Abstract. In the context of research on intelligence, autonomous agents and in particular mobile robots are to behave on their own without any human con-trol. Unfortunately, the real world exhibits plenty of noise, uncertainties, sudden changes, etc, which all imposes significant challenges on the design of appro-priate control architectures. This chapter starts off with an existing controller, known as the distributed adaptive control architecture and shows how significant improvements can be achieved by incorporating biological mechanisms, such as proprioception. The resulting controller requires much less preprogrammed de-sign knowledge, exhibits more flexible adaptation capabilities, and is more fault tolerant with respect to environmental changes and sensor failures as its prede-cessors.