A system theoretic approach to constructing test beds for multi-agent systems

As the number of multi-agent systems grows so does the need for testing agents in virtual dynamic environments. A system theoretic approach to constructing test beds for multi-agent systems is presented. The formalism is rooted in DEVS and describes agents and their environments as reflective time triggered automata. The hierarchical compositional model design is complemented by an abstract simulator to support the parallel, discrete event execution of the reflective automata. This theoretical exploration has brought into being the simulation system JAMES, a Java-Based Agent Modeling Environment for Simulation. It constitutes a framework for constructing test beds, in which the effects of different agent strategies, e.g. referring to deliberation, mobility, and interaction, can be experimentally analyzed. The model design allows describing and embedding agents that deliberately change the overall systems' interaction and composition structure, e.g. by moving from one interaction context to another. The execution layer realizes the abstract simulator in a distributed environment. To reduce communications over the net, a move at the model level from one interaction context to another is answered by migrating processes at the execution layer. Thus in JAMES, modeling and simulation layer are coined equally by, an agent-based perspective.