Human-Like Build-Order Management in StarCraft to Win against Specific Opponent's Strategies

摘要

StarCraft: Brood War is one of most-popular classic real-time strategy (RTS) games and has attracted much interest in recent researches. The Artificial Intelligence (AI) in StarCraft has not been that advanced, so professional players have consistently been stronger than any AI. We have considered the build order problem in RTS games for this paper and propose a simple human-like build-order management model to pursue human intelligence, where matrix B represents the build order in the model. Our experiments using state-of-art AI showed that the model is capable of strengthening the AI and effective at countering a specific opponent's strategies. Our experiments also showed that switching B in the model in accordance with the information on the opponent's strategies from scouts is effective.