【作者】 郭博；

【导师】 程家兴；

【作者基本信息】 安徽大学 ， 计算机应用技术， 2006， 硕士

【摘要】 国际人工智能协会组织的机器人世界杯RoboCup（The Robot World Cup Soccer Games and Conferences）仿真比赛系统，是人工智能和机器人研究的一种集中表现，它的仿真比赛融合了决策与对策、神经网络、人工生命与智能控制等学科内容，是研究多智能体的一个很好的平台。 这篇论文首先介绍了多智能体系统，同时回顾了机器人足球世界杯的以往赛事，随后又给出了其仿真系统及其主要模型。而本论文的研究主要都建立在UvA_Trilearn₂003的基础上，除了对其结果的诸多改善外，我们做的主要工作如下： 给出了Agent具体的组织结构； 针对Winglet2005的特点设计了Agent的详细流程； 借用BP算法的思想设计了一种新的传球策略； 给出了决策树在不同角色下的具体的实现； 提出了一种具体的Agent消减算法，并在其基础上实现非通讯多Agent之间的协作； 设计了一个完整的战术库，并给出了战术库的具体实现。 本文所完成的工作，都在Winglet2005仿真球队中进行了实现，并对其进行相应的实验，同时证明了相关方法的可行性和有效性。 最后还对当前的工作做了简单的总结，并展望了将来的工作。希望能够给类似的研究和工作提供一定的参考。更多还原

【Abstract】 RoboCup （ The Robot World Cup Soccer Games and Conferences ） simulation game system is a good sample of the research of artificial intelligence and robot. The simulation game includes such contents as decision-making 、 neural network and intelligent control. It is a good research platform for the research of MAS and has aroused the enthusiasm of many people.After presenting the concept of Multi-Agent System and the background of RoboCup. this paper introduces soccer server and some important models in detail. In addition to much improvement based on the UvA_Trilearn₂003. the main contents and related original achievement of this paper are as follow:This paper presents the framework of agent in detail;The flow of agent is developed in Winglet2005:A new strategy of pass is introduced based on BP algorithm;Different roles of decision-making tree are implemented;This paper has designed an agent-minify algorithm in detail, and implemented non-communicative multi-agent collaboration on it:This paper puts forward the concept of tactics system, and develops a complete tactics system.All the works in this paper have been implemented in Winglet 2005 simulation team. The experiment result shows that the approaches are valid and efficient.In the end, this paper makes a conclusion of the work that has been done and should have done. I hope this thesis can give help to others who are interested in this field.更多还原