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

动态环境下智能代理团队的突发智能研究。

• 批准号: 14350209
• 负责人: HIRATA Hironori
• 金额: $ 10.56万
• 依托单位: Chiba University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (B)
• 财政年份: 2002
• 资助国家: 日本
• 起止时间: 2002 至 2004
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-14350209/
• 关键词: emergent systems; multiagent; intelligent wheelchair; reinforcement learning; autonomous mobile robots; cooperative behavior; learning automata; 遺伝的プログラミング; 自律行動; セルオートマトン

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.

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。

项目成果

知的車椅子のための進化型シミュレータと行動統括機構の開発

智能轮椅进化模拟器及行为控制机制的开发

• 发表时间: 2005
• 期刊: 電気学会論文誌D(産業応用部門) 125
• 作者: 今井誠;川戸康司;濱上知樹;平田廣則
• 通讯作者: 平田廣則

Development of intelligent wheelchair acquiring autonomous, cooperative, and collaborative behavior

• DOI: 10.1109/icsmc.2004.1400888
• 发表时间: 2004-12
• 期刊: 2004 IEEE International Conference on Systems, Man and Cybernetics (IEEE Cat. No.04CH37583)
• 作者: T. Hamagami;H. Hirata

濱上知樹, 小圷成一, 平田廣則: "適応的な状態分割を行うQ-Learningにおける状態数の調整方法"電子情報通信学会論文誌. Vol.J86-D-I,No.7. 490-499 (2003)

Tomoki Hamagami、Seiichi Otono、Hironori Hirata：“一种通过自适应状态划分来调整 Q-Learning 中状态数量的方法”，电子、信息和通信工程师协会学报，第 J86-D-I 卷，第 7. 490 期。 -499（2003）

濱上知樹, 平田廣則: "セル上に配置されたマルチエージェントによる群集流のシミュレーション"電気学会論文誌C. Vol.123,No.11. 2028-2035 (2003)

Tomoki Hamagami、Hironori Hirata：“使用放置在细胞上的多智能体模拟人群流动”IEEJ Transactions C. Vol.123，No.11 2028-2035 (2003)。

松田泰明, 濱上知樹, 小圷成一, 平田廣則: "フローシステムにおける分岐構造形成の計算機シミュレーション"電気学会論文誌C. Vol.123-C No.3. 585-590 (2003)

Yasuaki Matsuda、Tomoki Hamagami、Seiichi Otoni、Hironori Hirata：“流动系统中分支结构形成的计算机模拟”日本电气工程师学会会刊 C. Vol.123-C No.3 (2003)。