Evolving predator control programs for an actual hexapod robot predator

In the development of autonomous robots, control program learning systems are important since they allow the robots to adapt to changes in their surroundings. Evolutionary Computation (EC) is a method that is used widely in learning systems. In previous research, we used a Cyclic Genetic Algorithm (CGA), a form of EC, to evolve a simulated predator robot to test the effectiveness of a learning system in the predator/prey problem. The learned control program performed search, chase, and capture behavior using 64 sensor states relative to the nearest obstacle and the target, a simulated prey robot. In this paper, we present the results of a new set of trials, which were tested on the actual robots. The actual robots successfully performed desired behaviors, showing the effectiveness of the CGA learning system.