Collaborating With A Genetic Programming System To Generate Modular Robotic Code

While the choice of a set of primitives can strongly e ect the performance of genetic programming, the design of such a set remains more of an art than a science. We look at a joint approach that combines both hand-coding and genetic programming to dene and re ne primitives in the context of a creating distribute control code for modular robots. We give some rules of thumb for designing and re ning sets of primitives, illustrating the rules with lessons we learned in the course of solving several problems in control code for modular robots.