Facilitating Controller Evolution in Morpho-functional Machines — A Bipedal Case Study

Abstract

Simulations of bipedal walking showed that the evolution of stable gaits can be greatly facilitated by harnessing design solutions inspired by those of biological organisms. Three such areas are addressed in this paper: a) the use of passive dynamics such as free-swinging lower legs, knee caps and springy ankles, b) proportional derivative controllers (or their variants) as actuators, and c) a modular, specifically coupled neural controller architecture. It is shown that an appropriate implementation of these components greatly improves the speed of evolution – stable straight line walking is typically achieved in substantially less than 100 generations. Moreover, compared to earlier experiments, the quality of the solutions arrived at is more satisfying with respect to their resemblance to human body dynamics during walking.