Competitive coevolution versus objective fitness for an autonomous motorcycle pilot

Evolution in the context of genetic algorithms is driven by the fitness function. For some applications, this factor is not easy to compute and coevolution represents an alternate solution. Thus, competition between individuals in the population can be used as a performance measure instead of an objective function, when the nature of the problem allows it. In this paper we explore the impact of such a choice on the overall performance of the solutions, as compared to the classic approach. We apply this model to a problem of configuring a multi-agent autonomous pilot for motorcycles.