The goal of this project is to develop effective finite-state machine (FSM) strategies for winning against an adversary in a Competition for Resources simulation. To achieve this goal, we use evolutionary algorithms for strategy improvement. A variety of evolutionary methods are compared experimentally in this context. Key empirical questions are addressed, such as how many FSM states are optimal, how effective is it to use an evolutionary algorithm that adapts the number of states, and how can one reduce the variance in fitness evaluation? Some of our experimental answers to these questions are quite intriguing. This research also explores and evaluates novel algorithms for detecting and repairing deleterious cycles in the evolved FSMs.
* This project was funded by ONR.
For more information, please contact
William M. Spears.
Last modified: 07/28/99