Increasing evolvability of a locomotion controller using a passive-dynamic-walking embodiment

Abstract

Over the past decade it has been widely recognized that the emergence of intelligence is influenced significantly by not only the controllers implemented but also their embodiment, that is robots' body structure. In spite of its importance, still few studies have explicitly investigated this mutual interaction. In light of these facts, in this paper, the interaction between controllers and their embodiments is investigated. To do so, a synthetic approach is employed. More specifically, an embodiment performing a passive-dynamic-walking function is taken as a practical example, and its locomotion controller that can walk on flat terrain is evolved using a genetic algorithm. Simulations have been carried out and the results have showed that passive-dynamic-walking embodiments provide significantly high evolvability. This strongly suggests that "appropriate" embodiments allow us to increase the efficiency of developing their controllers.