RoboCup中基于效果操作的动态行为规划模型

如何提高agent的学习能力、对手建模能力以及多agent团队运作能力是目前RoboCup研究所面临的3项挑战，在上述的挑战中，行为规划起了非常重要的作用。agent如何能够在动态实时的复杂环境中根据场景变化来动态规划自己的行为是RoboCup目前急需解决的问题。提出一种面向效果操作方法的动态行为规划模型，使队员能够在场景分析的基础上，根据经验动态选择和执行行为策略，且具有持续学习的能力，采用贝叶斯信念网络和基于示例推理相结合的方法来实现。实验结果表明，该方法有效提高了队员适应环境的能力。

Dynamic Planning Model on the Basis of Effect-based Operations for RoboCup

The simulated RoboCup is a complex, dynamic multi-agent system full of uncertainty. In such an environment that the state-space complexity restricts the designer's ability, the designer can not supply agents with complete and exact information about the correct response at any state. So learning, teamwork and opponent modeling become the three challenging problems in RoboCup, and planning is playing an important role in these challenges. How to plan activities to influence adversary's actions based on situation assessment in the dynamic, real-time environment must be solved now. In this paper, we propose a dynamic planning model on the basis of effect-based operations. Agents can select and execute actions from their own experiences based on the analysis of the situation. Furthermore, the system has life-long learning ability. We integrate bayesian belief network and case-based reasoning to implement it. Experimental results show that the agents' adaptation has been improved.