An Efficient Framework for Winning Prediction in Real-Time Strategy Game Competitions

Abstract

In recent years, real-time strategy (RTS) games have gained attention in the AI research community for their multitude of challenging and relevant real-time decision problems that have to be solved in order to win against human experts or to collaborate effectively with other players in team games. However, a big challenge for creating human-level game AI is the different traits of races of opponents and their locations of enemy units are partially observable. To overcome this limitation, we explore evolutionary-based approach for estimating the location of enemy units that have been encountered. In this paper, we propose an efficient framework to predict the winning ratio between the different races used in the real-time strategy game. We represent state estimation as an optimization problem, and automatically learn parameters for the evolutionary-based model by learning a corpus of expert StarCraft replays. The evolutionary- based model tracks opponent units and provides conditions for activating tactical behaviors in our StarCraft bot. Our results show that incorporating a learned evolutionary-based model improves the performance of EISBot by 60% over baseline approaches.