Evolving a behavior-based control architecture - from simulations to the real world

Abstract

Genetic programming makes it possible to automatically search the space of possible programs. First we evolved a behavior-based control architecture using computer simula-tions. Then we replicated one of the exper-iments with a service robot, showing that Koza's classic experiment of evolving a con-trol architecture can be transfered to the real world with a change of representation. The use of a service robot necessitates safety mea-sures which are also explained. Results are reported for the experiments using computer simulations and with the real robot. 1