Modular neural control for a reactive behavior of walking machines

Abstract

A small modular neural network is presented which is able to control the sensor-driven behavior of walking machines with many degrees of freedom. The controller is composed of a so called minimal recurrent controller (MRC) for sensory signal processing, a SO(2)-network as neural oscillator to generate the rhythmic leg movements, and a velocity regulating network (VRN) which expands the steering capabilities of the walking machine. This recurrent neurocontroller enables the machine to explore an in-door environment by avoiding obstacles. It was developed and tested using a physical simulation environment, and was then successfully transferred to the physical four-legged walking machine, called AMOS-WD02.