Case Learning and Indexing in Real Time Strategy Games

Development of real time strategy game AI is a challenging and difficult task. However, the current architecture of game applications doesn\´t support well the utilization of user contributed contents to get better game playability. The portability of the algorithms is quite poor due to the use of the problem specific heuristics. Real-time learning may be a possible solution, but it involves long training time. In this paper, we propose a case indexing method using neural-evolutionary learning approach in a "tower defense"-style real time strategy (RTS) game. Artificial neural network (ANN) is trained on the cannon placement combinations by the result of genetic algorithm (GA). This model provides an efficient indexing of past experience. Experimental results are provided to illustrate our idea.