Towards a predictive model of an evolutionary swarm robotics algorithm

The Robotic Darwinian Particle Swarm Optimization (RDPSO) previously proposed is an evolutionary algorithm that benefits from a natural selection mechanism designed to solve complex tasks (e.g., search and rescue). Yet, the stochastic-ity inherent to this algorithm makes it hard to predict teams´ performance under specific situations and, therefore, almost impossible to synthesize the most rightful configuration (e.g., teamsizes) by means of a trial-and-error approach. This paper gives the first steps towards a predictive model that may be able to capture the RDPSO dynamics and, to some extent, estimate the collective performance of robots. The predictive model proposed is represented by a semi-Markov chain being compared to its microscopic counterpart by means of simulation experiments. The results show that the predictive model is able to predict the RDPSO performance with minor discrepancies, presenting itself as a reliable approach to synthesize robotic swarms.