[ Japanese Version ]

Agent Framework

To support designing complex mechanisms for agents, we have developed two agent description languages, RXF and MiLog. RXF is a reflective constraint logic programming language and its runtime environment. MiLog is a logic-based mobile agent framework that is strongly focused on developing WWW-based intelligent information agents.

[ MiLog , MiPage , iML | RXF ]


Cooperation among intelligent software is on of the most important issues on software research. We call the software 'agent'. To support designing complex mechanisms for agents, we have developed an agent description language RXF. Improvements of the information technology give us easily access to massive amount of information. It causes another problem 'information overload'; it is difficult to find needed information from vast amount of information garbage. Supporting information gathering is an important issue on the world. We are tackling to realize 'information gathering support agent' by using our technologies on designing agents. Information agent is an intelligent software that supports information-gathering tasks of a user. The main difficulties are scalability and programmability. Scalability is important because we want to gather many types of information from the Internet. Programmability makes it easy to customize and adjust behaviors of agents. To deal with dynamically changing information, we have to adjust the agent to information changes. To utilize network and computers is difficult for a single agent. Our goal is to design information-gathering support systems by using cooperative agents. To realize these issues, we are developing a mobile cooperative agent framework 'MiLog'. On the MiLog framework, we are integrating 'Anytime Migration' technology that enhances reactivity of agents. The framework provides a logic programming language to support designing complex behaviors of agents. Many libraries are also integrated for easy design of accessing and analyzing information on the WWW. 'Anytime Migration' is implemented by radically new logic engine that realizes both efficiency and flexibility of the agent migration. We are developing support tools for designing user interface modules to make our agents a real application. Our approach will dramatically improve the design process of intelligent information gathering applications. We are currently developing applications based on the framework. Some of them are used in our lab.


The aim of developing MiPage framework is to support agent development process by integrating HTML-based user interface design method and logic-based behavior description method. Agents built by MiPage are mobile agents based on MiLog, and also agents are WWW applications which can serve many functionalities via HTTP. Users can access these agents by using WWW browsers such as Netscape or IE. MiPage is implemented by a MiLog application which compiles a MiPage document to a MiLog source code. MiPage agents use the MiLog framework for their runtime environment.


In our previous implementation, GUIs are implemented by using Java and Java to MiLog communication interface. There are two problems using this approach, the usability and the performance. On GUI programming, there are events established by the user's operations. The programmer may need some codes to handle and pass the event to appropriate prolog queries. At this situation, the system do not provide strong mobility on the Java programming even if the MiLog framework provides strong mobility on prolog programming. The state of GUIs are preserved by using the Java serialization technique. The problem is the performance of the Java serialization is low and some GUI components (especially built by Swing components) cannot preserved by Java serialization. On the iML framework, all GUIs are represented by using first order logic in compact form. These represented GUIs are operated seamlessly by prolog programs. On the iML framework, we provide a visual tool to design GUI layouts.