Evolving non-trivial behaviors on real robots

Abstract

Concerns the development of autonomous robots. The automatic design process proposed is called evolutionary robotics because it involves a form of artificial evolution. However, it is still not clear if this approach is adequate for real life problems. In this paper we show how control systems that perform a nontrivial sequence of behaviors can be obtained with this methodology by “canalizing” the evolutionary process in the right direction. We also show how optimal results are obtained by reducing the amount of supervision to a minimum. (2 pages)