Abstract

Body-brain co-evolution concerns the simultaneous optimization of body and brain to promote the discovery of optimized solutions. A key benefit of this approach lies in its adaptability, since it alleviates the burden of identifying suitable morphologies for a given problem. A widespread application of body-brain co-evolution regards robot locomotion, in which enabling morphological evolution provides notable advantages over using fixed (and potentially sub-optimal) morphologies. However, assessing the impact of morphological rate on the final performance is often overlooked. In this work, we delved into the analysis of how big the morphological rate should be in order to foster a significant performance enhancement. To this end, we performed an investigation on three 2D robot locomotion problems, BipedalWalker, Embryo and Halfcheetah2D, by considering both simple and challenging environmental conditions. To co-evolve body and brain, we employed the OpenAI Evolutionary Strategy (OpenAI-ES) and the Generational Genetic Algorithm (GGA). Our analysis indicates that, regardless of the considered algorithm, body-brain co-evolution is remarkably more effective than using fixed morphologies, and the advantage increases as the morphological rate becomes bigger.

Statistics