Research On Role-Based Multi-Agent System Development Method

Agent-Oriented Software Engineering (AOSE), Multi-agent Systems (MAS) and Agent-Oriented Programming (AOP) attract more and more attention in software development field. AOSE is even regarded as a revolution by some researchers. Meanwhile, Agent-Based Social Simulation (ABSS) attracts social researchers' great interests too, e.g. RoboCup has already become an ideal test-bed for both ABSS and MAS. Besides, more MAS applications are emerging, which include many important military applications, such as Joint Grid-Based Integrated Targeting system developed by USA in 2001.Although the research on agent has been extended over about 20 years, no MAS development method has becomes the standard one. Resarchers have developed more than 20 different methods, but none of them gives an underlying thought about how to develop an MAS development method. This thesis proposes a way of developing a MAS development method. Conceptual model, product model and process model are three essential elements of a method. Based on this belief, an MAS development method, RoMAS, has is designed. The main work includes:1. The first question we need to answer is "what are agents?". To address this question, many definitions have been proposed by researchers. The main consideration is about agent's characteristics such as autonomous, social, reactive, etc. But they failed to give a constructive definition that can tell how to construct an autonomous, social, reactive agent. In this thesis, the underlying homogeneousness between human beings in a society and software agents in a MAS is presented in this thesis, based on which a new definition and architecture for agent is proposed. An agent is an instance of some roles. Such agents can be more appropriate for dynamic, open system's analysis, design and implementation. This understanding and definition lead to the MAS conceptual model proposed in this thesis. The conceptual model applies models, i.e. role, agent, interaction protocol and so on to construct MAS. We argue in this thesis that such a conceptual model will be more suitable for open and dynamic MAS than others.2. Another important question we need to answer is "How does an agent behave?". To address this question, the associated structure model and behavior model of MAS are described, which compose a full product model. Since UML is the de facto accepted modeling language for object-oriented software methodologies, we propose an UML-based graphical modeling language to describe such models. We extend UML to model roles, agents, interaction protocols, role organization diagrams, agent state diagrams, agent role-changing diagrams and role interaction diagrams.3. To complete our work, we need to design a method that guides MAS designersthrough requirement analysis, system design and implementation, i.e. process model. Many researchers have brought about quite a lot of theories about the description of agent's mental attitudes, internal behaviors and so on. However, there still leaves much to think about the agent systems' software engineering process and methods. In this field, some valuable and important works have been put forward, such as MaSE, Gaia, MESSAGE/UML, etc. Each of them pays attention to one or several aspects of agent system, but few of them give enough thought to the dynamic properties of agent system's role-based organization. Most of these works begin with analyzing system's static role models and transforming them into agent models, and then succeeding make detail design and later implementation. In this thesis, a MAS modeling method RoMAS (Role-based MAS modeling method) based on these definitions is presented by means of the case of RoboCup simulation football team case, including its graphical modeling language and the modeling process. A formal description of RoMAS processes is presented in this thesis, which provides formal and strict semantic for the graphically presented processes.4. Based on ArgoUML, a modeling tool is developed by our team. It supports graphically representation for role, agent, interaction protoco...