Opponent Modeling and Strategic Reasoning in the Real-time Strategy Game Starcraft

Abstract

Since the release of BWAPI in 2009, StarCraft has taken the position as the leading platform for research in artificial intelligence in real-time strategy games. With competitions being held annually at AIIDE and CIG, there is much prestige in having an agent compete and do well. This thesis is aimed at presenting a model for doing opponent modeling and strategic reasoning in StarCraft.We present a method for constructing a model based on strategies, on the form of build orders, learned from expert demonstrations. This model is aimed at recognizing the strategy of the opponent and selecting a strategy that is capable of countering the recognized strategy. The method puts weight on the ordering and timing of buildings in order to do advanced recognition.