Towards the Co-Evolution of Influence Map Tree Based Strategy Game Players

We investigate the use of genetic algorithms to play real-time computer strategy games. To overcome the knowledge acquisition bottleneck found in using traditional expert systems, scripts, or decision trees we use genetic algorithms to evolve game players. The spatial decision makers in our game players use influence maps as a basic building block from which they construct and evolve trees containing complex game playing strategies. Information from influence map trees is combined with that from an A* pathfinder, and used by another genetic algorithm to solve the allocation problems present within many game decisions. As a first step towards evolving strategic players we develop this system in the context of a tactical game. Results show the co-evolution of coordinated attacking and defending strategies superior to their hand-coded counterparts