On the Development of an Intelligent Computer Player for CLUE: a Case Study on Preposterior Decision Analysis

The detective boardgame of CLUEreg can be viewed as an example of preposterior decision problem, where decisions on how to navigate the board are based on the expected utility of the observations, and the observations are aimed at improving an inference process. The same principles arise in modern surveillance systems, such as demining sensor networks, where the sensor platforms (e.g., autonomous ground or vehicles) move about the environment in order to collect measurements or evidence from unknown targets and improve inference of unknown features. The boardgame of CLUEreg serves as a well-known and intuitive example problem, that displays the same couplings between motion planning and inference, as modern surveillance systems. In this paper, a Bayesian network (BN) approach is used to develop an automated computer player for CLUEreg, that is tested through an interactive simulation of the game. The results show that the intelligent player plans its motions according to the evidence that needs to be collected, and is capable of winning the game of CLUEreg against experienced human players