Evolving ANN Controllers for Smart Mobile Robots

Abstract

In this work we present an overview of the application of evolution for obtaining autonomous robot controllers. It concentrates on controllers for behavior based robots implemented through Artificial Neural Networks. Specific approaches for taking into account temporal relationships are presented as well as a methodology for the progressive implementation of controllers comprising multiple behaviors. In addition we will consider the problem of transferring simulation results to real robots and the conditions that must be met for this process to be effective. Some examples of the application of these techniques to simple problems are included.