Research on Board Game Strategy Methods Based on Reinforcement Learning
Runzhou Luo, Boning Yao, Qingwen Zhang · Applied and Computational Engineering · 2025
As a typical sequential decision-making and gaming problem, board games have the complexity of large state space and strong dynamic confrontation, however, traditional methods have many limitations in dealing with them, so they need to be based on reinforcement learning to achieve strategy optimization by virtue of data-driven. Reinforcement learning can promote the realization of AI decision-making ability from rule-dependent to data-driven leap, and show significant advantages in game AI. This paper systematically sorts out the core algorithms of reinforcement learning in board games, comparatively analyzes their technical characteristics, applicable scenarios, advantages and disadvantages, discusses the current technical bottlenecks and ethical challenges, and look forward to the future development direction. This paper concludes that reinforcement learning is effective in board games, which not only helps AIs such as AlphaGo and Libratus to surpass the human level in Go, Texas Hold'em and other scenarios, but also forms the transition from “model-dependent” to “data-driven”, From “model-dependent” to “data-driven”, and from “single-intelligence” to “multi-intelligence”, it has also formed a technological evolution vein. At the same time, reinforcement learning has been breaking through in processing high-dimensional states, complex reward functions, etc., and has shown the potential of generalization in the fields of education, healthcare, etc. [1]. This paper can provide theoretical references and practical guidance for subsequent AI research on board games, as well as a universal methodology for complex decision-making problems.