A Generic Agent Architecture for Real-Time Distributed Situation Assessment


Research into a generic agent architecture for real-time distributed situation assessment (DSA) holds much promise because it would greatly facilitate the development of a variety of military and commercial applications of DSA and related systems. Examples of such applications include: distributed sensor networks, distributed network diagnosis, distributed information retrieval and digital libraries, distributed perceptual processing for cooperating robots/autonomous vehicles, and computer-supported cooperative work on situation analysis tasks. DSA will also be an integral component of many distributed planning and scheduling systems.

Three years ago, we initiated an ambitious, long-term effort to develop a generic architecture for large-scale distributed real-time situation assessment systems. Significant progress has been made toward this goal through the development of the major components that are necessary for such an architecture and through the investigation of approaches for formally analyzing DSA systems and applications. We see the key issues as how to organize both local agent and network-wide problem solving so that the agents can cooperate effectively to produce answers of appropriate quality within fixed deadlines, using limited communication bandwidth, and have their performance degrade gracefully as sensors, communication links, and processors fail.

Our recent work has emphasized the development of an agent model that can support sophisticated coordination strategies in the context of real-time constraints. The task database maintains a representation of the local agent's tasks and the tasks of other agents, with information about the relations among the tasks. Local task information is provided by the interpretation system module, which is also reponsible for low-level control decisions, or when there are no relevant task relationships (subproblem interactions) or real-time constraints. When there are relevant task relationships and real-time constraints, the real-time scheduler uses the task information and interacts with the interpretation module to select appropriate tasks/actions. We are currently working on the integration of the interpretation module with the other modules and expect the further development of this model will be an important focus of continued research.

Sources of Additional Information

This work is being done in conjunction with Professor Victor Lesser at the University of Massachusetts. It is supported by the Dept. of the Navy, Office of the Chief of Naval Research, under Grant No.N00014-95-1-1198.

This project encompasses several subprojects:

The DRESUN Testbed for Distributed Situation Assessment

TAEMS: A Framework for Task Analysis, Environment Modeling, and Simulation

Generic Coordination Strategies for Agents (GPGP)

Design-to-time Real-time Scheduling

A Framework for Analysis of Sophisticated Control

Additional relevant information may be found from: The Distributed Artificial Intelligence Group at UMass.

Relevant publications

"The DRESUN Testbed for Research in FA/C Distributed Situation Assessment: Extensions to the Model of External Evidence," N. Carver and V. Lesser, Proceedings of the International Conference on Multiagent Systems, June, 1995.

"Sophisticated Cooperation in FA/C Distributed Problem Solving Systems," N. Carver and V. Lesser, Proceedings of AAAI-91, 191--198, 1991 (also available as Technical Report 91-23, Department of Computer Science, University of Massachusetts).

Norman Carver's home page.