The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #94108

Title:Using Co-Evolution Gas to Find Robust Dodgem Players
Date: 1994
Abstract:The GA and co-evolution GA, where the fitness values of the population members are interdependent, are used to explore the heuristics for the alpha-beta search which is employed by the computer players to play the two-player board game - Dodgem. Eight "features" associated with weights and activation situations (switching features off/on) are designed as the constituents of the heuristic evaluation function and three methods for switching features are proposed. The GA is used to explore the weights and activation situations of the features. In this project, a chromosome represents a single computer player, but contains information for two heuristics for playing first and second. The fitness values of the chromosomes are evaluated from the results of the "matches" which they played. The work of the co-evaluation GA can be examined by inspecting the number of groups in the population and this is presented in this project. The computer players' strategies can be examined by inspecting the heuristics. The ability of the computer players are examined by holding the competition between the computer and human players.

[Search These Pages] [DAI Home Page] [Comment]