摘要
针对动态对抗的多agent系统(MAS)环境中agent行为前摄性较差的问题,提出了一种将意图跟踪和强化学习相结合的agent模型.该模型将对手信息和环境信息分开处理,在agent的BDI心智模型中引入了Q-学习机制应对环境变化;在强化学习的基础上注重对对手和对手团队的意图跟踪,改进Tambe的意图跟踪理论,针对特定对抗环境中的对手行为建立对手模型,跟踪对手和对手团队的意图,预测对手目标,以调整自身行为.实验证明,所提出的agent模型具有更强的自主性和适应性,在动态对抗系统中具有更强的生存能力.
A reinforcement learning agent model with intention tracking has been proposed to overcome the lagging in action in dynamic confrontation multi-agent systems (MAS) environment. The information of opponents and that of the environment have been treated differently. Based on reinforcement learning, the paper pays more attention on the intention tracking of the opponents. The intention tracking theory of Tambe have been improved, and opponent models and group-opponent models have been set up to track opponent intentions for forecasting the opponent's targets and revising agent-self's actions. Simulations have provided experimental results proving that agents with this model are more autonomic and adaptive.
出处
《北京理工大学学报》
EI
CAS
CSCD
北大核心
2004年第8期679-682,共4页
Transactions of Beijing Institute of Technology
基金
国家"八六三"计划项目(2002AA735051)