ASPIRE Adaptive strategy prediction in a RTS environment

Abstract

When playing a Real Time Strategy(RTS) game against the non-human player(bot) it is important that the bot can do different strategies to create a challenging experience over time. In this thesis we aim to improve the way the bot can predict what strategies the player is doing by analyzing the replays of the given players games. This way the bot can change its strategy based upon the known knowledge of the game state and what strategies the player have used before. We constructed a Bayesian Network to handle the predictions of the opponent's strategy and inserted that into a preexisting bot. Based on the results from our experiments we can state that the Bayesian Network adapted to the strategies our bot was exposed to. In addition we can see that the Bayesian Network only predicted the possible strategies given the obtained information about the game state.