The nascent field of artificial life promises to shed new light on issues in biology amenable to computer simulation. Researchers can benefit greatly from a toolkit designed to facilitate the development of evolutionary simulations by providing domain-specific, high-level components and a simulation architecture enforced through design affordances. This thesis describes the analysis, design, implementation, and evaluation of such a toolkit.
For the analysis, an ontological framework is developed by examining several representative simulation systems from the literature. Requirements for the toolkit are derived by abstracting common features from the surveyed systems. Based on these requirements, a prototype class library called SIMBIOSYS is described.
The library provides a basic time-stepped simulation architecture consisting of a virtual environment inhabited by any number of evolving populations of autonomous agents. Each agent's behavior is driven by a program which takes the agent's local perceptions as input and computes an intention which the environment resolves into an action.
Initial testing of the toolkit on example applications shows that it is useful for developing evolutionary simulations. The examples include a simple version of Conway's Life, a re-implementation of Genesys/Tracker, and an original variation of the Iterated Prisoner's Dilemma problem. Areas are suggested where the toolkit can be enhanced and extended.
Many people contributed to the evolution of this research, and it has been enriched because of them.
I wish to thank my supervisor, Brian Gaines, for encouraging me to pursue a topic on the fringes of traditional computer science. His suggestions have been invaluable.
I benefited greatly from the weekly meetings of the Biological Simulation Group. Jules Bloomenthal, Keith Ferguson, Mark Hammel, Ken Kittlitz, Mark James, and our token biologist, Duane Hewitt brought their unique perspectives to the discussions which ranged from the detailed technical to the esoteric philosophical. I hope our meetings continue and our ideas evolve for a long time to come.
Dave Astels, Mark Hammel, and Ken Kittlitz offered many excellent suggestions on how earlier versions of the thesis could be improved. I am very grateful to them for their help.
I wish to thank my friends at Merak Projects for their extraordinary support., especially James Henry, Adrian Zissos and Deane Stewart for their encouragement and for providing useful comments on drafts of the thesis.
Bill Leeb and cEvin Key deserve an acknowledgment for constant inspiration over the last three years.
Thanks also goes to my family for the moral support when it was needed most. And special thanks to Barbara Frizzell for everything; this thesis is dedicated to her.