摘要
兵棋游戏的复杂性,增加了单一技术构建的智能体,在兼顾粗粒度策略调整和细粒度行动控制方面的难度,限制了智能体效能。为此提出了面向兵棋的“意图-任务-行动”多层级智能体架构,旨在为兵棋游戏提供一种能够整合多种技术优长的智能体建模技术。该架构通过自上而下的分解传递机制,将作战策略逐步分解、转化为可执行的任务和动作。使用有限状态机、联盟博弈和行为树等技术分别实现了智能体架构的各层级架构。当前智能体适用范围严格限定于陆战战斗环境。最后在兵棋平台上与多个基准AI进行多轮次对抗实验,验证了该架构的可行性和有效性。
The complexity of war games poses challenges for single-technique-built agents,as it requires balancing between coarse-grained strategic adjustments and fine-grained action control,thus limiting agent performance.To address this,this paper proposes a multi-level agent architecture for military strategy games based on the"Intent-Task-Action"framework,aiming to integrate various technical strengths for intelligent agent modeling in these games.Through a top-down decomposition and propagation mechanism,this architecture progressively breaks down combat strategies into executable tasks and actions.Techniques such as Finite State Machines,Coalition Games,and Behavior Tree are employed to implement the different levels of the agent architecture.The application scope of the realized agent is preliminarily defined as the tactical level scenario of the land combat unit.Finally,through multi-round adversarial experiments with various benchmark AI on a war gaming platform,the feasibility and effectiveness of the proposed architecture are validated.
作者
余晓晗
袁铎
姚昌华
YU Xiaohan;YUAN Duo;YAO Changhua(Army Engineering University,Nanjing 210001;Nanjing University of Information Science and Technology,Nanjing 210044,China)
出处
《指挥控制与仿真》
2025年第1期69-76,共8页
Command Control & Simulation
关键词
兵棋游戏
智能体
多层级
有限状态机
联盟博弈
行为树
war games
agent
multilevel
finite state machines
coalition games
behavior tree
