Phase-dependent evaluation in RTS games

Modern video games present an environment in which game AI is expected to behave realistically (i.e., 'human-like'). One feature of human-like behaviour of game AI, is the ability to assess accurately the current situation. This requires an appropriate evaluation function. The high complexity of modern video games makes the task to generate such an evaluation function for game AI a difficult one. However, we assume that many difficulties can be solved automatically. Therefore, our aim is to generate fully automatically an evaluation function for game AI. This paper describes our approach, and discusses the experiments performed in the real-time strategy (RTS) game spring. Two evaluation terms are established, one to perform a unit-based evaluation, the other to evaluate positional safety. In addition, a mechanism is incorporated to perform the evaluation dependent on the phase of the game. From our results we may conclude that the generated evaluation function effectively predicts the outcome of a Spring game.