In the monolithic programming model, increasingly capable systems require increasingly complex software. Multi-agent systems acheive sophisticated capability through complex interactions, not increasingly complex software. As such, modularity, reconfigurability, and extensibility are most easily achieveable and components can largely be tested in isolation. However, most implementations of multi-agent systems to not take advantage of modularity and reconfigurability because the depend too heavily on the forsite of the designer at design time. Reconfiguration is typically a time-consuming manual process that often involves changes to the components themselves. The creation of a general multi-agent software architecture that can learn from its own interaction with the world, evaluate its performance, and fine-tune itself to better achieve its goals would find inherent use in the distributed system, real-time control, and proxy computing arenas. We propose the distributed system supporting port-based agents to be such an architecture.
We are using mobile robots as a test bed, and we have integrated our work with the CyberRAVE environment.