Multiobjective exploration of the StarCraft map space

• Published in: Proceedings of the 2010 IEEE Conference on Computational Intelligence and Games pp. 265 - 272
• Main Authors: Togelius, J, Preuss, M, Beume, N, Wessing, S, Hagelback, J, Yannakakis, G N
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.08.2010
• Subjects: Evolutionary computation; evolutionary multiobjective optimization; Fractals; Games; Humans; Minerals; procedural content generation; Real time systems; Real-time strategy games; RTS; Visualization
• ISBN: 9781424462957; 1424462959
• ISSN: 2325-4270
• DOI: 10.1109/ITW.2010.5593346

This paper presents a search-based method for generating maps for the popular real-time strategy (RTS) game StarCraft. We devise a representation of StarCraft maps suitable for evolutionary search, along with a set of fitness functions based on predicted entertainment value of those maps, as derived from theories of player experience. A multiobjective evolutionary algorithm is then used to evolve complete StarCraft maps based on the representation and selected fitness functions. The output of this algorithm is a Pareto front approximation visualizing the tradeoff between the several fitness functions used, and where each point on the front represents a viable map. We argue that this method is useful for both automatic and machine-assisted map generation, and in particular that the Pareto fronts are excellent design support tools for human map designers.