Modular Reactive Neurocontrol for Biologically Inspired Walking Machines

A neurocontroller is described which generates the basic locomotion and controls the sensor-driven behavior ofa four-legged and a sixlegged walking machine. The controller utilizes discrete-time neurodynamics, and is ofmodular structure. One module isfor processing sensor signals, one is a neural oscillator network serving as a central pattern generator, and the third one is a so-called velocity regulating network. These modules are smnll and their structures and theirfunctionalities are analyzable. In combination, they enable the machines to autonomously explore an unknown environment, to avoid obstacles, and to escapefrom corners or deadlock situations. The neurocontroller was developed and testedfirst using a physical simulation environment, and then it was successfully transferred to the physical walking machines. Locomotion is based on a gait where the diagonal legs are paired and move together, e.g. trot gaitfor the four-legged walking machine and tripod gaitfor the six-legged walking machine. The controller developed is universal in the sense that it can easily be adapted to different types of even-legged walking machines without changing the internal structure and its parameters. KEY WORDS-autonomous robots, walking machines, recurrent neural networks, neural control, reactive behavior