ABSTRACT
This chapter discusses the AI system designed for Prismata, the online strategy game developed by Lunarch Studios. For the Prismata AI, a new algorithm called Hierarchical Portfolio Search (HPS) was created which reduces the action space for complex strategy games, which helps deal with some of challenges. Prismata is a two-player online strategy game, best described as a hybrid between a real-time strategy (RTS) game and a collectible card game. HPS was designed to make decisions in games with extremely large state and action spaces, such as strategy games. It is an extension of portfolio greedy search algorithm, which is a hill climbing algorithm that has been used to guide combat in RTS games. HPS is a bottom-up, two-level hierarchical search system which was originally inspired by historical military command structures. An important factor in success of HPS is the creation of Portfolio itself, since only actions generated by partial players within the portfolio will be considered by top-level search.
