Effects of encodings and quality-diversity on evolving 2D virtual creatures

Abstract

How to jointly optimize the morphology and controller is a challenging problem in evolutionary robotics. Due to the large search space, both quality diversity algorithms and types of encodings have been employed to search the solution space more effectively. Here we compare Multi-dimensional Archive of Phenotypic Elites (MAP-Elites) and a standard evolutionary algorithm as well as the effect of using a direct versus an indirect encoding. The results showed that the MAP-Elites algorithm found diverse solutions, yet the encodings accounted for a larger performance discrepancy. This indicates that the representation is at least as important as the optimization method for effectively creating robots.