Community Builder: Structuring Agent Architectures to Facilitate Domain Tasks

摘要

The results of this work include (1) Community Builder, a design methodology to assist software designers in designing mixed initiative, multi- agent intelligent decision support systems (DSSs); (2) Development of architectures for multi-agent decision support systems in several task domains and (3) Dyna-plan, a reusable framework describing the generic reasoning cycle used in most dynamic and uncertain environments. As the complexity of computing needs continually increases, multi-agent systems are becoming indispensable as approaches for making complex systems modular and manageable. However, designing effective organizations for multi-agent systems is far from simple. Each time an agent-based system is designed to support a new task, designing an effective architectural structure for the system of agents that works effectively for the task is difficult and time-consuming. The aim of the Community Builder and methodology is to provide software designers with a descriptive design methodology that will help them to identify the domain specific constraints on the agent architecture, and to organize the agents accordingly. The aim of the DynaPlan framework is to capture some of the structure common to many planning under uncertainty domains so that other developers can re-use some of the knowledge that we have gained in building related domains.