Optimized Management of Networked Sensors in the Presence of Communication, Collection, and Processing Latencies

We propose to develop algorithms to manage multiple networked sensors embedded in a system with significant latencies. Sources of latency can include limited bandwidth in communication channels; limited capabilities for processing collected data, especially imagery; and data collection times. Latencies must be accounted for by the sensor manager so that it can appropriately hedge. By using sensors with shorter latencies to cue sensors with longer latencies, the sensor resources required per target can be reduced. To model sensor latencies and other characteristics, we propose to use Bayesian networks. This modeling framework is already being used by the Missile Defense Agency on other programs and provides a principled means for modeling systems. Until recently, performing calculations with such a network model has been impeded by the lack of computationally efficient algorithms, especially in the case of continuous states. Lately, a number of exact and approximate algorithms have been developed for efficiently performing calculations with such a model. We propose to incorporate them into our sensor management algorithms. Specifically, we propose to work within the framework of approximate dynamic programming to develop a sensor manager that uses advanced Bayesian net inferencing algorithms to evaluate expected rewards resulting from a sensor policy.