Framework for Model-Based Design of Multi-agent Systems

Simulation models, Agent control language, Reactive and BDI agents, Model-based design

Artificial agents and multi-agent systems are attractive area which touches many other areas of interest including non-informatics ones like sociology or economy as well as many topics relating to computers, for example artificial intelligence or system design. When we have to develop a multi-agent system we need, among others, to build agent architectures together with some decision algorithms and communication protocols. Before such systems are realized in a real world we should sufficiently test and verify them. Presented text describes a new approach to building artificial agents and multi-agent systems using methodology of model-based design. In this methodology some models are used during the design process. The models are used for testing of particular elements behavior as well as for testing of behavior of the system as whole - in both cases by their simulation. Environment surrounding the elements is also simulated in the model and particular systems are tested whether they fulfills given objectives in proper way. When all the system elements work well in the model then they could be realized in some real environment. Our effort is to develop a tool that would allow model-based design of systems with artificial agents. For this reason we have been building application called T-Mass (Tool for Multi-agent System Development) which is aimed right on the model-based development of such systems and provides some important facilities for building rational agents. As a part of the tool we developed language called t-Sapi by which agents' behavior is controlled. Also we made two-phase synchronous algorithms for control of the model run. This paper shows how some popular agent architectures could be built with the t-Sapi language and how then they are used in the multi-agent simulation model. We also present some remarks about usage of the modeled agents and their consequent realization in real multi-agent applications.