Combining deliberation and reactive behavior for AI players in the Mini-Tichu card-game

Tichu is a popular trick-taking climbing card game that involves strong strategic play along with elements of chance. Apart from the typical search-based methods that employ domain-dependent or domain-independent mechanisms for selecting the most promising action at each step of the game, the complex dynamics that arise from the team-play feature of Tichu suggests the use of agent-based techniques for specifying a higher-level tactical strategy for artificial intelligent players. In this work we focus on finite state machines, a technique that has been traditionally used for specifying the behavior of non-player characters in video-games, and investigate how they can be coupled with simple search-based methods as a means for developing strong AI players in a simplified version of the game that we call Mini-Tichu.