Conditions for Outperformance of Recombination in Online Evolution of Swarm Robots

Genetic recombination is commonly used in evolutionary al-gorithms and yet its benefits are an open question in evolu-tionary biology. We investigate when recombination is actu-ally beneficial in the evolutionary adaptation of swarm robot behaviour in dynamic environments. In this scenario, artifi-cial evolution has to deal with challenges that are similar to natural evolution: it must run online, distributed and evolve the genome structure. These requirements could diminish the benefit of recombination due to disruptive crossover. Using neural networks as robot controllers, we reduce this disrup-tiveness with an adaptive mate choice that evolves the prob-ability of recombination and the genetic similarity of mates. In two experiments with a multi-agent simulation, we com-pare the adaptive performance of this approach with random recombination and pure mutation. Whereas both recombina-tion treatments naturally outperform at low mutation rates, pure mutation achieves its best performance with high rates, where it also outperforms random recombination. The adap-tive mate choice, however, achieves the same performance as pure mutation at high rates and outperforms when the net-work size is increased. We also found that treatments with recombination evolved smaller neural networks with fewer links