UCT-RAVE algorithm applied to multi-player games with imperfect information

Author

Rui Xiongli ; Rui Xiongxing ; He Yinglai

Author_Institution

Nanjing Inst. of Technol., Col. Commun. Eng., Nanjing, China

Volume

1

fYear

2011

fDate

20-22 Aug. 2011

Firstpage

312

Lastpage

315

Abstract

Aiming at the problems that traditional gaming search algorithms do not suit to multi-player games with imperfect information, a method of combining UCT-RAVE(Upper Confidence bound applied to Tree - Rapid Action Value Estimation) and Monte-Carlo sampling is proposed after analyzing the principle and characteristic of UCT-RAVE algorithm. First, the imperfect information is replaced by simulated perfect information through Monte-Carlo sampling then UCT-RAVE is used for searching based on that perfect information, at last most suitable action is selected after considering the best profits of many Monte-Carlo samples. Simulation demonstrated the feasibility and the effectiveness of the method.

Keywords

Monte Carlo methods; computer games; game theory; learning (artificial intelligence); sampling methods; search problems; Monte-Carlo sampling; UCT-RAVE algorithm; gaming search algorithm; multiplayer game; reinforcement learning; simulated perfect information gaming; upper confidence bound applied to tree-rapid action value estimation; Analytical models; Computational modeling; Estimation; Games; Heuristic algorithms; Learning; Monte Carlo methods; Monte-Carlo sampling; UCT-RAVE algorithm; card gaming; gaming search; multi-player games with imperfect information;

fLanguage

English

Publisher

ieee

Conference_Titel

Information Technology and Artificial Intelligence Conference (ITAIC), 2011 6th IEEE Joint International

Conference_Location

Chongqing

Print_ISBN

978-1-4244-8622-9

Type

conf

DOI

10.1109/ITAIC.2011.6030213

Filename

6030213