A simple hybrid algorithm for improving team sport AI

In the very popular genre of team sports games defeating the opposing AI is the main focus of the gameplay experience. However the overall quality of these games is significantly damaged because, in a lot of cases, the opposition is prone to mistakes or vulnerable to exploitation. This paper introduces an AI system which overcomes this failing through the addition of simple adaptive learning and prediction algorithms to a basic ice hockey defence. The paper shows that improvements can be made to the gameplay experience without overly increasing the implementation complexity of the system or negatively affecting its performance. The created defensive system detects patterns in the offensive tactics used against it and changes elements of its reaction accordingly; effectively adapting to attempted exploitation of repeated tactics. This is achieved using a fuzzy inference system that tracks player movement, which greatly improves variation of defender positioning, alongside an N-gram pattern recognition-based algorithm that predicts the next action of the attacking player. Analysis of implementation complexity and execution overhead shows that these techniques are not prohibitively expensive in either respect, and are therefore appropriate for use in games.