SDSU Department of Computer Science
  
Home Contact Us Undergraduate Student Info Graduate Student Info Faculty & Staff News & Events
News Calendar Problem of the Fortnight

October 2008 Calendar

CS Masters' Thesis Defense

Title: Starcat and Robocode Genetic Algorithm
Speaker: Lance Finfrock
Date: Monday, October 20, 2008
Time: 2:00 p.m.
Location: GMCS 405
Thesis advisor: Dr Joseph Lewis

Abstract:
Starcat is a framework which provides the common functionality of Copycat, an autonomous analogy-making system. This functionality is capable of integration into many domains. This paper describes an experiment which integrated the Starcat framework into Robocode, a program that allows virtual user-modified robots to compete against each other in battles. The integration of these two frameworks resulted in an autonomous robot known as Botcat, which was used to prime a population of robots.

The Botcat robots were run through fitness tests in the Robocode environment, and then evolved using a Genetic Algorithm with roulette wheel selection. Many generations of Botcat robots were created using this technique in order to obtain optimal individuals which had autonomously learned how to excel in the Robocode environment. Emergent behavior was displayed from the robot populations, and some unexpected results were discovered. It was concluded that the as the generations increased, the populations tended to perform better in the tests; however at the exceptionally higher generations the diversity in the population tapered and the populations’ ability to evolve beneficial traits decreased. Various extensions of the study included comparisons involving fuzzy and crisp codelets, as well as alterations of the fitness tests to produce a more advanced Botcat individual. The results of the study are provided as well as suggestions for future experimentation with the Botcat system.
webmaster@cs.sdsu.edu page counter College of Sciences