This paper is concerned with the analysis of coevolutionary dynamics among 3D artificial creatures, similar to those introduced by Sims [1]. Coevolution is subject to complex dynamics which are notoriously difficult to analyse. We introduce an improved analysis method based on Master Tournament matrices [2], which we argue is both less costly to compute and more informative than the original method. Based on visible features of the resulting graphs, we can identify particular trends and incidents in the dynamics of coevolution and look for their causes. Finally, considering that coevolutionary progress is not necessarily identical to global overall progress, we extend this analysis by cross-validating individuals from different evolutionary runs, which we argue is more appropriate than single-record analysis method for evaluating the global performance of individuals.
展开▼