From Simulation to Development in Multi Agent Systems: A JADE-based Approach

Abstract

Multi-agent systems (MAS) present an interesting approach to the efficient development of modular systems. MAS are composed by autonomous computational units called agents that are programmed to compete or work together, for instance in order to solve computational problems, engage in automatic negotiation or play computer games. Frameworks exist that aid the development of this class of systems and they range from mostly general-purpose frameworks to domain-specific in an array of different domains. Multi-agent-based simulations (MABS) are sometimes used on the course of development of a full-featured MAS - for instance, for testing purposes. However, most platforms for MAS development are not well suited for MABS development. JADE, a very popular MAS development framework allows the creation of seamless distributed agent systems and complies with FIPA standards for agent interaction. Unfortunately, its multi-threaded architecture falls short in delivering the necessary performance to run a local simulation with a large number of agents.

Repast is an agent-based simulation toolkit that allows creating simulations using rich GUI elements and real time agent statistics. Unlike JADE, though, Repast lacks much of the infrastructure for agent creation and interaction. Some authors proposed an integration of JADE and Repast by means of a middleware, essentially representing the agent in both and taking advantage of the features provided by the frameworks. The code generation tool proposed in this thesis should be capable of not only generating a Repast simulation from an existing JADE MAS, but also of creating a full featured JADE application from a Repast-based simulation. An implementation of FIPA's interaction protocols will be proposed for Repast as a means to deliver the mentioned conversion tool. The most immediate advantage over the previous approaches is that existing systems can make use of this tool to create appropriate simulations where agent interaction can be more easily analyzed. Furthermore, proficient programmers in one framework can quickly get started in the development for the complementary framework.