This project will extend previous work on a decision theoretic approach for designing satisficing multiagent systems so that it is practical to apply the approach to large-scale, real-world sensor networks. The approach is based on distributed Bayesian networks and decentralized Markov decision processes. In terms of intellectual merit, this research addresses a critical need for formal techniques to support the design of multiagent systems. Approximate, satisficing approaches can reason about the appropriate tradeoffs between information quality and resource usage and, thus, reduce barriers to building large-scale sensor networks, as well as other large-scale multiagent systems. In terms of broader impact, the project could ultimately have a substantial positive societal impact by encouraging the development of sophisticated, large-scale sensor networks. Such large-scale sensor networks have numerous military and civilian applications. They can be used to enhance security and safety, as well as to make computerized systems more autonomous and flexible.
