Summary of Results for Small Scale Combat in Wargus

This paper provides a status report on the progress of improving small-scale combat in Wargus. The main goal of the project was to automatically learn complex tactics through Monte Carlo strategies. Monte Carlo strategies can be used as a game tree search method and provide a mechanism for handling the uncertainty and search complexity in an RTS game. The idea to explore Monte Carlo search as a means of learning tactics is inspired by the Master’s thesis by Ronni and Nielsen (2005). The initial results show that Monte Carlo strategies outperform the default AI of Wargus in small battles. However, for battles with more than 10 units, Monte Cargo techniques failed to show complex behavior. The remainder of this paper is structured as follows. Section 2 provides an overview of the JNI-Wargus interface and additional modifications made to Wargus. Section 3 explores several fixed strategies for small-scale combat. Section 4 provides a description of the Monte Cargo implementation. Section 5 provides results and Section 6 provides conclusions.