| Computer game-playing is an important domain for artificial intelligence research. Games with imperfect information that can simulate decision making under the uncertainty and imperfection of the real world are more popular with artificial intelligence researchers. Siguo is a typical imperfect-information game,where completing players make decisions under conditions of uncertainty and leaguing players have to deal with cooperation. These characteristics make Siguo to be a new test-bed for computer game-playing research with imperfect information.This paper has made detail research about Siguo computer game-playing and the main works are as follows:1. This paper designs and develops the Siguo intelligent system, which is composed of server and client. Especially, client system is analyzed and implemented in this paper. The client system consists of knowledge representation, move generator, searching engine, evaluation function and inference module.2. Evaluation function is one of the basic elements in computer game-playing research. Based on two attitudes (i.e.optimism and pessimism) this paper designs two evaluation functions of Siguo, which consist of several features such as piece capture, position domain of piece and oriflamme guard. Comparing the optimal moving strategy and corresponding payoff between the two evaluation functions (just consider feature of piece capture) with different attitudes are discussed and analyzed by experiments under the same conditions.3. How to rapidly and accurately predict an opponent's types of pieces is an important problem in Siguo computer game-playing. This paper proposes to use LBP(Loopy Belief Propagation) for rapid inference and prediction the types of pieces of an opponent. Experimental results indicate that LBP can successfully infer the exact types of pieces of an opponent and the ratio of true inference is approximately 48.7%. This ratio of true inference is better than that of common human players of Siguo. |
PDF Full Text Request