计算机技术与发展2025，Vol.35Issue(9)：1-8,8.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0082
智能博弈领域中的对手建模方法综述
Survey on Opponent Modeling Methods in Intelligent Gaming
摘要
Abstract
Opponent modeling,a collective term for methods that learn and construct models of opponents online or offline to predict their behavior,intentions,strategies,or other characteristics,has been applied across various fields such as cybersecurity,finance,and intelligent robotics.It has also become one of the hot topics in intelligent game research,especially in situations with partial observability,where in-complete information presents significant challenges to effective opponent modeling.Addressing the current issues of diverse opponent modeling methods,unclear application scenarios,and a lack of systematic comparison of strengths and weaknesses,a new classification system for opponent modeling methods is established based on the degree of observation of the game environment.The system categorizes opponent modeling into two main categories:fully observable opponent modeling and partially observable opponent modeling.Building on the discussion of fully observable opponent modeling methods,we focus on opponent modeling in partially observable settings,clarify the appropriate contexts for specific techniques and conduct a comparative analysis of the advantages,disadvantages,and limitations of each approach.Finally,we propose future research directions that consider these factors,thereby providing significant insights for ongoing investigations in this field.关键词
对手建模/智能博弈/部分可观察/完全可观察/知识重用/深度强化学习Key words
opponent modeling/intelligent gaming/partial observability/full observability/knowledge reuse/deep reinforcement learning分类
信息技术与安全科学引用本文复制引用
程恺,张金鹏,邵天浩,邹世辰,于本川..智能博弈领域中的对手建模方法综述[J].计算机技术与发展,2025,35(9):1-8,8.基金项目
国家自然科学基金(61806221) （61806221）
