Design Strategies for Evolutionary Robotics

i Acknowledgments First and foremost, I would like to thank my thesis advisor, Sushil J. Louis, for guiding my research virtually every step of the way. It was he who got me interested in GAs and Evolutionary Robotics in the rst place, as well as helped me improve my structured programming skills. Without his support and guidance this thesis would not have been possible. In addition, his work with combinational circuits greatly innuenced the design of my logic circuits. Finally, I would like thank him for helping me with the document preparation system used to write this thesis. Without it, I would still be trying to gure out how to print greek symbols and include post script gures instead of submitting this nal work. I would also like to thank Dr. Carl Looney, who expanded the scope of my research by teaching me the fundamentals of neural networks. Although I did not use the data collected in this phase of experimentation for my thesis, it gave me a better understanding of others' work in the same eld of research. It also gave me an entirely new perspective on how to attack the problem of computational design. Finally, I would like to thank Dr. Dwight Egbert who's knowledge and instruction on PC graphics was instrumental in developing the software platform I used to conduct nearly all of my experiments. ii Abstract This thesis deals with designing the structure and control of SIMulated roBOTS through the use of genetic algorithms (GAs). The attributes of some input sensors are encoded as binary bit strings, which the GA manipulates to evolve desirable structures. The GA also designs combinational circuits made up of logic gates to control these simbots. The environment in which these simbots function consists of a known, bounded area which contains obstacles and food sources. The underlying goal of the experiments is to evolve simbots which can approach simulated food or other goal destinations, while avoiding the obstacles and boundary walls. In order to perform these simple behaviors, the simbots are given touch sensors, hearing sensors and locomotive tank tracks, which enable them to sense entities and move around in the environment. After these low level behaviors are developed, the GAs design switching circuits that lead to more complex and useful behaviors. Contents Acknowledgments ii Abstract iii 1 Introduction 1