Evolved Neural Reflex-Oscillators for Walking Machines

Abstract

Legged locomotion has not been understood well enough to build walking machines that autonomously navigate through rough terrain. The current biological understanding of legged locomotion implies a highly decentralised and modular control structure. Neurocontrollers were developed for single, morphological distinct legs of a hexapod walking machine through artificial evolution and physical simulation. The results showed extremely small reflex-oscillators which inherently relied on the sensori-motor loop and a hysteresis effect. Relationships with biological findings are shortly discussed.