Online evolution and adaptation of central pattern generators for multi-robot organisms

Abstract

This thesis deals with possibility to provide a robot organism, consisting of an amount of single smaller robots, with the ability of locomotion. It is integrated into the SYMBRION project which is funded by the European Union. The used robots and the simulation environment are a product from this major project for swarm robotics. The presented locomotion approach uses artificial neural networks which are composed of third generation neurons called “Spiking Neurons”. For evaluating the generated motion patterns the artificial neural networks are evolutionary adapted which was realized by using “Evolutionary Acquisition of Neural Topologies”. In this thesis the evolutionary engine “EvoRoF”, launched by Florian Schlachter of the University of Stuttgart, was used. The findings of this scientific work were included directly in the adjustment process of this evolutionary engine. Specially the focus of this thesis is on distributed online evolution. Meaning that each robot of the whole organism has its own population of individuals and thus its own set of artificial neural networks. In the course of the evolutionary process the artificial neural networks start from scratch on directly on the robotic system. There are no networks which were precalculated on a desktop computer.