Evolving co-operative homogeneous multi-robot teams

The application of artificial evolution to the design of co-operative homogeneous multi-robot teams encounters the basic yet important issue of how such teams are to be generated. One approach is to evaluate teams comprising identical copies of a single evolutionary individual. The alternative is to use a separate evolutionary individual to specify each member of a team. Intuitively the former seems better suited, and it has been widely applied to the evolution of many kinds of homogeneous system. However, so little consideration has been given to the latter approach that, despite its apparent unsuitability, there is insufficient empirical evidence on which to discount it. This paper reports on a comparison of the two approaches over multiple runs in the context of a non-trivial cooperative task carried out by simulated mobile robots controlled by arbitrarily recurrent neural networks. It was found that, contrary to expectations, the latter approach performed significantly better than the former.