Enhancing embodied evolution with punctuated anytime learning

Abstract

This paper discusses a new implementation of embodied evolution that uses the concept of punctuated anytime learning to increase the complexity of tasks that the learning system can solve. The basic idea is that there is one population of chromosomes per robot rather than one chromosome per robot and reproduction between robots involves a combination of two entire populations of chromosomes instead of the recombination of two single chromosomes. The embodied evolution with punctuated anytime learning system is compared with embodied evolution alone and evolutionary computation alone, as the three methods are used to solve a common problem. The results show that this new learning system is superior to the other methods for evolving colony robot control.