多智能体分布式实时仿真实验系统开发与规划算法研究
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
随着智能技术的发展,多智能体系统与控制理论和计算机技术等多学科的融合已成为科学研究的新热点。多智能体系统具有在空间上是分布式的、并行的,且系统的容错能力较强等特点。分布式多智能体系统是分布智能研究的一个重要分支。它的理论研究价值在于它将过去的封闭的、孤立的知识系统发展为分布式的智能知识系统,将智能集中型发展为非独立的分散智慧型。
    目前,对多智能体系统的研究可分成两个层次:一是关于多智能体系统理论的基础研究;二是关于多智能体系统特定的体系结构、软件实现等的研究,人们已把面向对象的思想引入智能体对象体系中,并形成了面向Agent的编程风范。
    分布式仿真系统主要研究的课题有两类:一是仿真模型和试验任务的并行化与分配;二是建立高效的分布仿真环境。本文通过对多智能体系统国内外研究现状的总结,在参考现有的最先进的方法和理论基础上,综合分布仿真的有关概念和技术,设计了一个以机器人在码头搬运货物的任务为实验背景,利用局域网进行分布式规划的仿真实验系统,并通过各项关键技术对系统进行了实现,该系统能够模拟多种任务环境和机器人模型,各机器人间通过通信来传递信息,用户可以添加自己设计的控制算法和协调策略库,实现任务完成过程的动态显示。
    多智能体协调合作问题是分布式人工智能的核心问题,多智能体群体协作机制用以解决各智能体之间的管理、调度、优化,是实现多智能体分布自主协作系统的关键。本文主要研究多智能体协作系统中的资源分配问
    
    
    题,即如何进行任务分配,以及任务分配之后如何对完成任务进行路径规划的问题。
    对于一个给定了总体任务的多智能体系统,首先面临的问题是如何将给定的任务合理有效的分配给各机器人。任务分配的不合理将直接影响机器人完成任务的效率,而分配算法的执行速度和所用的时间也将影响到机器人完成任务的情况。所以,在设计任务分配算法的时候,应该遵循一定的性能指标,以达到系统的需求。任务分配是在众多的匹配方法中寻找一个最合理的子任务分配方案。在数学形式上表现为在离散的、有限的数据结构上,寻找一个满足给定约束条件并使目标函数值达到最大或者最小的解。
    本文提出的OD任务分配算法正是考虑到了任务分配的最优性,以二维任务分配问题为模型,使每个机器人能够最优地得到任务。通过性能测试我们看到,该算法的运算量大,使得运行时间较长,尤其在机器人和任务较多时,更是无法得出结果。
    为了解决以上问题,我们用匈牙利算法(Hungary)实现了多智能体系统的任务分配。匈牙利算法是图论中完成二分图匹配的经典算法之一,它的应用背景是解决二维任务的分配问题。该算法的特点是能够最优地对任务进行分配,同时又能保证消耗较短的时间。在算法性能测试的过程中我们看到,算法的运行时间明显小于本文提出的OD算法,能够解决OD算法运行时间长、对机器人和任务仿真数量较少的问题,并使任务达到了最优地分配。最后我们得出,OD算法较适合解决小规模的任务分配问题,匈牙利算法较适合于解决中等规模的任务分配问题。
    任务分配完之后,需要对各自的任务进行路径规划。为了使机器人在某一规定的时间内完成任务,规划出的路径既要保证能够到达目标位置,又要保证其为最优的路径。本文根据深度优先搜索的思想提出了一种路径规划算法——SP算法,它主要适用于用拓扑图形构造的环境,并且以路径长度作为性能指标,寻找一条最短的路径。
    算法设计完成后,我们将其在本文设计的仿真系统上进行了测试,并给出了算法相应的性能指标。
    尽管我们对多智能体系统领域的一些问题进行了探讨,但是无论深度
    
    
    还是广度都是远远不够的,尚存许多问题需要深入的研究和解决。虽然如此,由于多智能体系统所具有的广泛实际意义,我们相信,本文所做的理论和实践研究对以后的工作是有所贡献的。
With the develop of agent technology, the fuse of multi agent system and control theory and computer technique has been the new focus for science study. The characteristics of the multi agent system are distributed and collateral and strong fault tolerant ability. The distributed system is an important embranchment of the distributed agent study. The value of the theory for study is that it makes the foregone close and isolate knowledge system distributed intelligent knowledge system and develops the intelligent centralization mode to non-absolute distributed wisdom mode.
    Now, the study on multi agent system can be carved up two levels: one is the study on the basic theory and the other is the study on the specific system frame and software realization.
    The main study of the distributed simulational system has two classes: one is the distribution and collating of the simulational mode and experimental tasks and the other is the building of the distributed simulational environment. This paper designs an emulational test bed based on the multi-robot material flow system of the storages and docks and using the LAN to communicate through
    
    
    the reference of the existing advanced method and the integration of correlative concept and technology. The system can simulate environments for many tasks and the user can append the control algorithms designing by themselves and can realize the dynamic shows of the process.
    The cooperating problem of the multi agent system is the nuclear issue of the distributed artificial intelligence and the cooperation mechanism of the multi agent system is used to solve the management and optimization of the agents and is the key realization of the distributed multi agent cooperation system. This paper aims to study the resource distributing problem of the multi agent cooperation system and the path planning after the completing of the task allocation.
    For the multi agent system with the task given the problem faced firstly is how to distribute effectively and reasonably the task to other agents. If the distribution for task is unreasonable, it affect the efficiency of the completion for tasks. So, we should follow some capability guide at designing the task allocation algorithm to satisfy the need for the system. Task allocation is to find a most reasonable allocation project in many method. It behaves to find a result satisfied the given restricted conditions and to make the result of object function maximal or minimal.
    The OD algorithm this paper advanced is in view of the optimization and uses the two-dimensional assignment mode so as to make every robot get the task. optimally. We can see through the test that the operation of the OD algorithm is complex so that the runtime of it is long and especially when the number of robot and task is big it can not get the result.
    To solve the problem above, we use the Hungary algorithm to realize the task allocation problem. Hungary algorithm is one of
    
    
    the classical algorithms and its application background is to solve the two-dimensional assignment problem. The characteristic of the algorithm is that it can both distribute the task optimally and ensure the runtime consumed by the algorithm is short. We can see that the runtime of the Hungary algorithm is shorter than the OD algorithm and it can solve the problem of the long runtime and distributes the task optimally.
    After task allocation, we need to make path planning for task distributed. To make the robots complete the tasks in the set time, the path planned ensures both to get to the aim and to be the shortest one. This paper advanced a path planning algorithm—SP algorithm based on the ideal of the depth-first search. The algorithm is used to find a shortest path viewing the length of path as the performance in the environment constructed using topological figure.
    After the algorithm was designed, we tested the algorithms in the simulational system designed and showed the performance of these algorithms.
    Though we discussed some problems in the multi agent system, bu
引文
1. YUCao,ASFukunaga,ABKahngetal,CooperativeMobileRobotics:
    AntecedentsandDirections,IEEEIROS95,1995:226-234
    2. 谈大龙,黄 闪,分布自主协作式的多机器人系统研究,机器人,1996,
     18(6):338-342
    3. J.Liu and J.Wu,On emergence of group behavior in a genetically
    controlled autonomous agent system,In Proceedings of the 1998 IEEE
    International Conference on Evolutionary Computation,Anchorage,AK,
    May4-9,1998:470-475
    4. J.Liu,J.Wu and X.Lai,Analytical and experimental results on multiagent
    cooperative behavior evolution,In Proceedings of the IEEE/IEE
    Congress on Evolutionary Computation,Washington,DC,July6-9,
    1999:1732-1739
    5. J.Han,J.Liu and Q.Cai,From ALIFE agents to a kingdom of n queens,
    In J.Liu and N.Zhong,editors,Intelligent Agent Technology:Systems,
    Methodologies,and Tools,World Scientific Publishing Co.,Singapore,
    November,1999:110-120
    6. 周绥平,陈宗基,分布交互仿真环境研究,系统仿真学报,1995:54-56
    7. 张 芳,林良明,多移动机器人协调系统体系结构与相关问题,机器
    人,Nov.,2001,Vol.23,No.6:554-558
    8. 黄 重,近年来仿真技术的新发展,华北电力技术,No.9,1996:52-54
    9. Cohen P,Levesque H.Confirm actions and Joint action[A],IJCAI-91,
     1991 [C]. 951-957
    10.Nicholas R Jennings,Commitments and Convertions:The Foundation of
    Coordination in Multi-Agent Systems[J],The Knowledge Engineering
     Review,1993,8(3):223-250
    11. Barry Brian Werger,Cooperation without Deliberations:A Minimal
    Behavior-based Approach to Multi-robot Teams[EB/OL],
    
    http://www-robotics.usc.edu/~barry/ullanta
    12.薛宏涛,叶媛媛,沈林成,常文森,多智能体系统体系结构及协调机
    制研究综述[J],机器人,2001,1
    13.薛宏涛,沈林成,朱华勇,常文森,基于协进化的多智能体系统仿真
    框架及其面向对象设计与实现[J],机器人,2001,5
    14.郭戈,柴天佑,基于需求的多机器人决策算法,机器人, Vol.23,No.6,
    5,2001:20-524
    15. Huosheng Hu et al,A Parallel Processing Architecture for Sensor- based
    Control of Intelligent Mobile Robots,Journal of Robotics and Autonomous Systems,1996:235-257
    16. 陈仁际,吴镇炜,王 韬,谈大龙,分布式多机器人装配系统任务合
     作规划算法研究,中国机械工程,2000.4,11(4):418-421
    17. 高志军,颜国正,丁国清,颜德田,陈忠泽,多机器人协调与合作系
    统的研究现状和发展,光学精密工程, Vol.9,No.2,2001:99-103
    18. 王经卓,秦培军,胡小兵,殷国富,用遗传算法实现MultiAgent协同
    设计中的子任务调度,淮海工学院学报,Vol.9,No.1,Mar.2000:18-23
    19. 陈华平,黄刘生,启发式任务调度中的处理器选择策略,软件学报,
    1999.11,Vol.10,No.11:1194-1198
    20. 张 颖,吴成东,原宝龙,机器人路径规划方法综述,控制工程,Vol.10,
    May, 2003:152-155
    21. RAVBrooks,Solving the Find-Path Problem by Good Representation of Free Space,IEEETransonSysManandCybern,1983.13(3):190~197
    22. C.Womgngamnit,D.Angluin,The robot the grid and the algorithm technical report,YALE DCS TR-1188,Yale university Computer Science Department,NewHaven,CT,1999
    23. Okhatib,Real-time Obstacle Avoidance for Manipulator sand Mobile
    Roborts,Int.J Robotics Research,1986.5(1):90-98
    24. 顾国昌,张巧荣,移动机器人分层路径规划方法研究,哈尔滨工程大
    学学报,Vol.22,No.5,2001:31-35
    25. T Ueyama,T Fukuda,A Sakaietal. Hierarchica lControl Architecture with
    
    
    Learning and Adaptation Ability for Cellular Robotic System,Distributed Autonomous Robotic Systems,1994: 17-28
    26. T Ueyama,T Fukuda,Cooperative Search Using Genetic Algorithm
    Basedon Local Information,IEEEIROS93,1993:1110-1115
    27. 范 永,谭 民,MRCS中机器人控制体系框架结构,控制与决策,Vol.15,No.3,2000:325-328
    28. 韩慧莲,杨风暴,分布交互仿真及其关键技术,测试技术学报,Vol.12,No.1,1998:32-36
    29. 惠天舒,童 军,陈宗基,李伯虎,分布交互仿真技术的发展,系统仿真学报,Vol.9,No.4,1997:1-7
    30. 王 硕,张 斌,谭 民,曹志强,多自主移动机器人计算机仿真系 统的实现与设计,系统仿真学报,Vol.14,No.2,2002:225-228
    31. 王超越,谈大龙等,一个多智能体机器人协作装配系统[J],高技术通信,1998,8(7):6-10
    32. 战 强,多移动机器人运动规划 [D],哈尔滨:哈尔滨工业大学,1999
    33. Tucker Balch,Ronald C Arkin,Behavior-Based Formation Control for Multirobot Teams [J],IEEE Transactions on Robotics and Automation,1998, 14(6)
    34. R.S.Sylett and D.P.Barnes,A multi-robot architecture for planetary
     rovers,Proceedings,5th ESA Workshop on Space robotics, ASTRA '98,
    3.6-4
    35. Gaofeng Huang,Andrew Lim,A Hybrid Genetic Algorithm for
    three-Index Assignment Problem,Congress on Evolutionary Computation,2003
    36. 焦吉成,对匈牙利算法的改进及其应用实例,山东冶金,Vol.22,No.2,2000:55-57
    37. 王 鹏,伊 鹏,金德鹏,曾烈光,匈牙利算法在输入排队调度仿真中的应用研究,计算机应用,Vol.23,No.7,2003:4-6
    38. 赵 升,对分配问题求解方法的改进,郑州工业大学学报,Vol.19,No.3,1998:92-95
    
    39. 吴晓涛,孙增圻,用遗传算法进行路径规划,清华大学学报(自然科
    学版),1995,35:14-19
    40. 周 林,张 文,娄寿春,赵 杰,空防对抗仿真系统概要设计,系
    统工程与电子技术,Vol.24,No.3,2002:111-113
    41. 赵 超,彭可茂,柳嘉润,申公璋等,战术任务综合控制系统分布式实时仿真系统设计,系统仿真学报,Vol.14,No.1,2002:60-63

© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号

地址:北京市海淀区学院路29号 邮编:100083

电话:办公室:(+86 10)66554848;文献借阅、咨询服务、科技查新:66554700