Evolving composite robot behaviour - a modular architecture

摘要

We have developed a composite control system for solving complex tasks with autonomous robots. The control system is evolved using artificial evolution and can be regarded as a modular decision tree where every node is a neural network. We show that the control system is robust to noise, is reactive, is extendable and can be set up to be configured automatically. Furthermore we show that robots in the real world work using this control system.