Cluster-based Alpha-Beta Search for Real-Time Strategy Games

2018 2nd International Conference on Artificial Intelligence Applications and Technologies

计算机科学

Han, X.^1 ; Yan, H.P.^1 ; Zhang, J.G.^2 ; Wang, L.F.^2

Department of Information Engineering, China University of Geosciences, No. 29 College Road, Haidian District, Beijing, China^1

CRIPAC, NLPR, Institute of Automation, Chinese Academy of Sciences, No. 95 Zhongguancun East Road, Haidian District, Beijing, China^2

关键词: Alpha-beta;    Alpha-Beta searches;    Branching factors;    Cluster-based;    Optimal actions;    Partial observability;    Real-time properties;    Real-time strategy games;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/435/1/012012/pdf
DOI  :  10.1088/1757-899X/435/1/012012

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

From an artificial intelligence point of view, Real-Time Strategy (RTS) game has been proven to be one of the most challenging areas. Due to the huge action state space, partial observability and real time property, the previous AI solutions are sill low in terms of speed. In this paper, we propose a Cluster-based Alpha-Beta Considering Durations (CABCD) algorithm that searches the optimal action for each unit based on the cluster. When the improved algorithm is applied to large RTS games, the enormous branching factors are significantly reduced. The approach is evaluated in StarCraft I, and outperforms competing methods promisingly.