【作者】 陈忠泽；

【导师】 林良明；

【作者基本信息】 上海交通大学 ， 精密仪器与机械， 2001， 博士

【摘要】 机器人的应用方式正在由部件式单元应用向系统式应用方向发展,多机器人系统在工业、军事、医学和航天等许多领域都有着广阔的应用前景。因此,多机器人协作理论问题已经成为机器人学研究的一个热点。研究多机器人协作问题的传统方法基本上是采用过程控制器的思维方式,这样的系统实际上由于将完成协作功能的模块“捆绑”在控制器中,从而使协作过程过多地依赖于原始信息,最终削弱了协作部分的协作策略决策能力。为此,本文将系统完成协作功能的部分(或者说过程)从协作过程控制器中剥离出来,用具有一定自主性(或者说“人”性)的Agent来实现,从而使得决策过程和决策结果只与它的知识库有关,使之成为一个协作策略专家系统,更专业地表达策略决策。这样协作的多机器人系统就可以抽象为一个多Agent系统,称之为协作多机器人Agent系统。本文以国家863计划与多机器人协作相关的主题为背景,着重对协作多机器人系统进行严格地定义和形式化建模,为协作系统的设计提供理论依据。全文内容分为六章:第一章是绪论,对协作多机器人系统研究的发展历史和现状进行了分析与研究,首次正式提出并论证了结合MAS(即多智能体系统)的理论与方法发展协作多机器人学的思想;提供了一个基于MAS的协作多机器人系统的概念框架。这一思想的提出为协作多机器人系统的设计提供了一条新的途径,充实了协作多机器人学的内容。同时,该设计方法与面向(Agent)对象的编程思想是一致的,使得协作系统易于软件实现。在多机器人智能体系统(MRAS:Multi-Robot Agent Systems)环境下,任务必须首先被精确定义和分解,才可以在执行过程中被各Agent协作完成。故而,第二章着重对(基于MAS的协作多机器人系统概念框架中的)任务模块的有关概念进行了形式化定义。这一章工作的意义在于,对“任务”的描述是从知识的层次进行的,包括任务的形式化模型建立,任务的知识推理语言构造以及它的知识表达及分解等。该方法与常规的基于Petri网等的任务描述有着显著的区别:前者摆脱了复杂的时序关系,可以兼顾任务模型建立方法的通用性及系统中Agent品质的多样性,可以在任务描述与自主Agent系统之间形成直接沟通,此外从知识层次描述任务有助于系统设计人员进行系统硬件尤其是传感器系统的设计。最后用积木世界问题对该章提出的部分主要概念做出了说明,并得到了积木世界任务的知识表达。第三章从历史唯物主义的哲学观点出发,对(多机器人Agent协作环境下的)Agent社会性进行了非形式化建模,在模型中主要对构成Agent社会性的若干概念(冲突和依赖、目标委托和目标采纳等)进行了比较深入地阐述;并对多Agent系统所包含的所有结构以及依赖关系结构在这些结构中的基础地位做了简单地说明。非形式化模型的建立对于建立协作Agent的形式化模型(见第四章与第五章)而言是基础性的。第四章在Agent社会性非形式化模型的基础上,对作为Agent行动演算推理的本源的信念(及其刷新机制)进行了形式化描述。主要内容包括,基于信念机的社会性Agent的信念模型的语法和语义的描述;信念机的信念计算(刷新)机制;一些常见的带有普遍性的信念之间的因果关系(推理规则)以及它们的适用范围等。信念的这种基于信念机的模型可以使Agent在环境发生变化时其信念得到相应刷新,从而始终保持信念的真实性。第五章建立了Agent行动演算的基于广义模态逻辑的形式化模型(在该模型中,Agent行动是以其信念和知识为基础,以愿望和目标等为原始推动力,以行动演算为行动策略决策而实现的),并实现了代码属性和语义容器属性在Agent行动演算形式化模型中的统一。从而,该模型不仅可以从宏观功能的角度,也可以从软件代码实现的角度表达机器人Agent实现行动演算的各个功能模块。本文最后还用积木世界问题作为一个简单的合作环境说明了协作过程中各Agent的信念集及其刷新情况,以及代表行动演算结果的各个状态的可行行动模态原子集;结果还表明,与Sarit Kraus等的基于商议的关于积木世界问题的协作方案的比较,本文的协作过程步骤简单,占用的计算时间较少。第六章是对全文工作的总结。包括本文的主要工作(创新点)以及在理论完善以及实践上需要进一步深入进行的工作。更多还原

【Abstract】 The application mode of robots is changing from cell to system one, which is of the need of practice as well as the evolution trend of robotics. And the theory of cooperative robotics cannot but to become a focus of robotics.Traditional methods to be used to deal with problems of cooperation are mostly from“control”point of views. Robots cooperate just through a kind of process Controller, which integrates into a course unit all levels of information processing and at last give a control signal. Based on a kind of control unit, the system can also accomplish actions in phase. But in fact, because of the amalgamation of cooperation and coordination module with the controller, cooperation & coordination rely excessively on primitive information. This leads to weakening its cooperation & coordination decision-making capability. Moreover, this kind of processing method corresponds precisely to structural programming style.With the development of cooperative robotics, the concept of multi-agent systems is becoming of growing importance, though we lack of an abstract, powerful, homogenous and simple conceptual framework to put them into practice, even more when multi-agent systems involve many components distributed over a wide and uncontrolled network of computers, information and competence. Our idea on dealing with cooperation came into being just under this environment. And in contrast to cybernetic method, our robot has a“mind”named agent, which is an autonomous and human one. Here an agent is actually an expert system on cooperative strategy, of which input material is from knowledge database and its output no longer directly depend on primitive information from sensors. This method is consistent with OOP (Object Oriented Programming) style. And this kind of agent based cooperative robot system is called Multi-Robot Agent System (MRAS). In this thesis we are to discuss a cooperative multi-robot agent systems from the following aspects:Chapter 1 At first, we give an analysis of foregone investigation on Cooperative Robotics. And then we will bring forward and demonstrate the viewpoint that the development tendency of Cooperative Robotics certainly will be combined with MAS. It is an approach to scientific cooperative robotics for us to use the theories and methodologies in MAS for reference. And then we will give a conceptual framework for MAS based cooperative robots system, which is consequently consist with AOP(Agent Oriented Programming) style from the software point of view.Chapter 2 This chapter we will address the problem of the control in distributed artificial systems how to decompose a task into subtasks, which will be allocated to distributed components, and have them cooperate to the solution of the initial problem. The notion of task is an essential component of knowledge based systems theories, so we deals with precise definition of task from the knowledge level point of view. A model framework is presented including a language for the knowledge reasoning. And in the end of this paper, we offered an example to illuminate concepts that were provided by the context. Compared with common Petri Net based method to be used to deal with task decomposing, the former gets rid of intricated time sequence relation exist among sub-tasks. And it has the advantages over compatibility of universality in modeling and multiformity in agent. And then achieves direct communication between task describing and autonomous agent. Furthermore, knowledge based task decomposing makes for easier system hardware designing.Chapter 3 This chapter deals with modeling informally for a social agent under Multi-Robot Agent Systems environment. The main reason for which we develop this modeling work is: the real foundation of all sociality is the individual social action and mind, one cannot reduce action at the collective level to action at the individual level unless one passes through the social character of the individual action. The most important two claims we will propose at first are: (1) Sociality emerges from the interference and dependence relation that presented among agents; (2) Goal-Delegation and Goal-Adoption are the basic ingredients of cooperation and organization. As far as interference and dependence relation be concerned, we expound emphatically the latter (i.e. dependence relation) from the following aspects: its role as the foundation of sociality, its premise and principles as well. Then we will illuminate Goal-Delegation and Goal-Adoption (including their motives and mechanisms). At the end of this chapter, we’ll recommend simply the notions of social organization and social structures.Chapter 4 In this chapter, we focus on setting up formal model of belief and its update mechanism. Contents included in this chapter are: formal model of social believes belong to an agent; and belief change mechanism exist in belief machine; and some familiar consequence relations exist between some items of believes and their support belief machines, etc.. All these jobs is for Agent Action Programming which we will be discussed in the next chapter.Chapter 5 In chapter 5, we are going to deal with setting up formal model of Agent Action Programming. We consider an agent as an entia of software code and its semantics container (it represents the semantics of all types of data and functions which construct agent software code) in the context. And we integrate successfully this two kinds of properties into our model. Finally, we let the BWE as an example of cooperation. The result indicates that,compared with negotiation based cooperative system put forward by Sarit Kraus, ours gains an advantage over simplicity and needs less CPU hours.Chapter 6 Chapter 6 is a summarize to this thesis. It includes main jobs done in this Ph. D. thesis and our future improve works.更多还原