Study on Emergent Intelligence from Intelligent Agent Team under Dynamic Environments.

本研究从理论构建到原型开发，对“动态环境下智能主体团队的涌现智能”进行了全面的研究和实践总结。(1)该项目的主题是“智能轮椅”，已将商用电动轮椅转化为原型。在原型轮椅的规范下，我们成功地从代理团队的交互中出现了专注于自主，合作和协作行动的智能行为。（2）以“群体行为”为核心，利用智能主体进行宏观行为的基础研究。人群行为和交通流的应急模拟实验结果显示了导致严重交通事故的自发组织现象和极其复杂的行为。(3)In在模拟或构造性方法的每一种情况下，为了在动态环境中获得有效的行为，Agent需要附加自适应学习和进化方法。发展强化学习和进化算法的先进技术，我们提出了实用的技术，使代理人获得智能行为的智能轮椅控制和多智能体仿真。(4)A近年来发展的一个显著趋势是采用“进化硬件”的动态配置系统，它能使Agent在故障情况下具有自主恢复能力。仿真实验的结果显示了自主机器人在迷宫问题上的优势。作为这些研究的结果，我们证实了该项目对完成基础技术的开发和显示多智能体应用程序的实际应用都有重要贡献。

With comprehensive studies and practical conclusions for "emergent intelligence from intelligent agent team under dynamic environments," this study has been comprehensively examined from a theory construction to a prototype development.(1)"Intelligent wheelchair," the major theme of this project, has been prototyped with converting commercial electric wheelchair. Under specifications of the prototyped wheelchair, we have succeeded in emergence of intelligent behavior focused on autonomous, cooperative, and collaborative actions from interaction in agent team.(2)Focusing on "crowd behavior", we conducted fundamental studies of macro-level behavior with intelligent agents. The results of emergence simulation experiments of crowd behavior and traffic flow have shown the phenomenon of autonomous organization and extremely complex behavior which leads to serious accidents.(3)In each case of simulation or constructive approaches, in order to acquire efficient behavior in dynamic environment, agents need to attach the adaptive learning and evolutionary approach. Developing advanced technique of reinforcement learning and evolutionary algorithm, we have proposed practical techniques which enable agents to acquire intelligent behavior in the intelligent wheelchair control and the niultiagent simulation.(4)A dynamic configuration system with "Evolutionary hardware," which is the notable trend of recent years, has been developed for enabling agent to have autonomous restoration in the case of fail. The result of the simulation experiment has shown the advantage at a maze problem of an autonomous robot.As the results of these studies, we confirmed that this project contributed importantly to the accomplishment of both development of fundamental techniques and showing practical applications for establishment of multiagent applications.