Using Lanchester Attrition Laws for Combat Prediction in StarCraft.

摘要: Smart decision making at the tactical level is important for Artificial Intelligence (AI) agents to perform well in domain of real-time strategy (RTS) games. Winning battles crucial RTS games, and while ahumans can decide when how attack based on their experience, it challenging AI estimate combat outcomes accurately. A few existing models address this problem game StarCraft but present many restrictions, such as not modeling injured units, supporting only a small number unit types, or being able predict winner fight remaining army. Prediction using simulations popular method, generally slow requires extensive coding model engine This paper introduces Lanchester’s attrition laws which addresses mentioned limitations faster than running simulations. Unit strength values are learned maximum likelihood estimation from past recorded battles. We experiments that use simulator generating both training testing, show capable accurate predictions. Furthermore, we implemented our method bot uses either traditional retreat. tournament results (against top bots 2014 AIIDE competition) comparing performances two versions, increased winning percentages method.