Hierarchical learning systems: robotic control using hierarchical genetic programming

In this thesis, we study the use of hierarchical genetic programming techniques to evolve robotic controllers for a simulated Khepera miniature robot. We study GP chromosome representation methods of linear-genome and tree-based and HGP techniques of Automatically Defined Functions, Module Acquisition, and Adaptive Representation. We train robotic controllers in the tasks of obstacle avoidance, wall following, and light avoidance. Our results enable us to compare and contrast the five studied representation methods and to provide suggestions for their improvement.