A Simulation-based Online Evolutionary Algorithm for Combat in StarCraft II

Real-time strategy (RTS) games have become one of the hotspots in the field of artificial intelligence research due to the large search space, long-term planning, and real-time constraint. In view of the fact that most of the current research on solving the RTS combat problem uses offline training methods, and these methods may need a lot of training time, this work uses an evolutionary algorithm (EA) to directly optimize the units’ action sequences online. In order to solve the problems of repetitive actions caused by the neutral area and the size correlation caused by cooperation between units, three search operators are proposed. It is shown that this method can effectively solve the combat problem in RTS games.