DRESUN was developed to explore the implications of having agents with more sophisticated evidential representations and control capabilities than the agents that were used in our earlier research with the Distributed Vehicle Monitoring Testbed (DVMT) (e.g., [Lesser and Corkill 1983] and [Durfee and Lesser 1991]). Because of the agent limitations, that research did not adequately address several important issues that may arise when sharing information among DSA agents, including: representing incomplete and inconsistent information from other agents, and determining what information is needed by other agents to resolve local uncertainties and global inconsistencies. Furthermore, overall agent activities were not driven by an explicit need to produce local solutions that were globally consistent.
DRESUN agents are based on the RESUN sensor interpretation framework. RESUN maintains explicit representations of the reasons why its interpretation hypotheses are uncertain, which are referred to as SOUs (source of uncertainty statements). Interpretation is viewed as an incremental process of resolving SOUs until an acceptable solution is found. DRESUN extends the RESUN representation with a group of global consistency SOUs (or GSOUs) that represent uncertainty due to unresolved questions about the global consistency of local solutions. Communication among the agents results from resolving the GSOUs and the GSOUs can be used to drive the agents to produce local solutions that are globally consistent.
DRESUN's approach to distributed interpretation is based on the functionally accurate, cooperative (FA/C) paradigm for distributed problem solving ([Lesser and Corkill 1981] and [Lesser 1991]). This approach was proposed for applications in which tasks are naturally distributed but in which the distributed subproblems are not independently solvable. In the FA/C approach, agents produce tentative, partial results based on local information and then exchange these results with the other agents. The constraints that exist among the agents' subproblems are then exploited to resolve the local uncertainties and global inconsistencies that occur due to the lack of accurate, complete, and up-to-date local information.
The key issue for FA/C agents is that of subproblem interdependencies (interactions): agents must communciate and exchange information in order to solve their own local subproblems (local area interpretations) as well as the global/system problem (global interpretation/assessment). A critical assumption of the FA/C approach is that agents can produce satisfactory global solutions without "excessive" communication among the agents. However, DSA tasks can present several sources of difficulty for information sharing: agents' local evidence may lead to solutions that are globally inconsistent; agent beliefs (interpretations of local data) are uncertain and imprecise; interpretations are complex structures (so agents will not usually have complete views of other agents beliefs due to communication and processing limitations); and beliefs are constantly being revised due to new data and further processing (so agents will not usually have up-to-date views of the other agents beliefs and responses to inquiries are received in a context that may have changed from when the request was made).
A major focus of work with DRESUN has been on issues related to modeling the beliefs/evidence of other agents. Experiments showed that extensions to the model of external evidence were necessary to effectively utilize inter-agent communication of incomplete and conflicting evidence, and evidence at multiple levels of abstraction. These extensions give agents the abilitity to represent the uncertainties that occur when DRESUN agents exchange such information, reformulate hypotheses to efficiently pursue alternatives, and numerically evaluate satisfaction of the global termination criteria. The extensions enhance the flexibility of the agents by improving their ability to evaluate the current state of agent beliefs, make better (local) use of incomplete information from another agent, and determine precisely what information is needed to resolve global inconsistencies. This helps limit communication among the agents by supporting incremental, directed communication of information.
The DRESUN architecture is one of the basic elements of our research into a generic agent architecture for real-time distributed situation assessment.
"Resolving Global Inconsistency in Distributed Sensor Interpretation: Modeling Agent Interpretations in DRESUN," N. Carver, V. Lesser, and Q. Long, Proceedings of the 12th International Workshop on Distributed Artificial Intelligence, May, 1993 (a revised and expanded version is available as Technical Report 93-75, Department of Computer Science, University of Massachusetts).
"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).