In the course of this thesis, the feasibility of automatically creating players for the game 2-AntWars is studied. 2-AntWars is a generalization of AntWars which was introduced as part of a competition accompanying the Genetic and Evolutionary Computation Conference 2007. 2-AntWars is a two player game in which each player has control of two ants on a playing field. Food is randomly placed on the playing field and the task of the players is to collect more food than the opponent.
To solve this problem a model of the behaviour of a 2-AntWars player is developed and players are built according to this model by means of genetic programming. To show the feasibility of this approach, the players are evolved in an evolutionary setting against predefined strategies and in a coevolutionary setting where both players of 2-AntWars evolve and try to beat each other.
Another core part of this thesis is the analysis of the evolutionary and behavioural dynamic emerging during the development of 2-AntWars players. This entails specific characteristics of those players (e.g. which ant found how much food) and on a higher level their behaviour during games and the adaption to the behaviour of the opponent.
The results showed that it is indeed possible to create successful 2-AntWars players that are able to beat fixed playing strategies that oppose them. The attempt to create 2-AntWars players from scratch by letting the developed players battle each other was also successful.
This is a significant result as it shows how to automatically create artificial intelligence for games (and in principle for any problems that can be formulated as games) from scratch. The developed solutions to the 2-AntWars problem were surprisingly diverse. Ants were used as bait, were hidden or shamelessly exploited weaknesses of the opponent.