Evolving a robot controller for a robotic swarm by TWEANN

Abstract

This paper focuses on Topology and Weight Evolving Artificial Neural Network (TWEANN) approaches for designing controllers of a robotic swarm. TWEANN approaches are expected to design not only synaptic weights but also the appropriate network topology without the intervention of the designers. Mutation-Based Evolving Artificial Neural Network(MBEANN) is a TWEANN algorithm that only use mutations to evolve neural networks. In this paper, we applied MBEANN to design the controller for a robotic swarm. For comparison with the MBEANN approach, we used the NeuroEvolution of Augmenting Topologies (NEAT), which is a widely used TWEANN algorithm.The performance and the topology of robot controllers are compared in collective foraging tasks. The results show that MBEANN could perform as well as NEAT with smaller network topologies.