Stochastic Dynamic Programming for Farsighted Sensor Management

Today's tactical aircraft employ sophisticated onboard sensors which can be managed in several dimensions, including spatial focus, sensing modality, and temporal scheduling. Effective mission operations, such as air interdiction and counter air, require efficient sensor control and scheduling that accounts for the limited availability of sensor resources to support all current sensor tasks, as well as the need to plan for a variety of future contingencies. This effort develops a theoretical foundation for sensor management based on stochastic dynamic programming, leading to nonmyopic algorithms that utilize statistical track and identification information from an aircraft's sensor fusion system to predict the effects of future sensor management decisions. Novel open-loop feedback and closed-loop feedback solution methodologies are proposed which are adaptive and even-driven, continually replanning on the basis of the most recent information. Closed-loop feedback approaches improve upon open-loop feedback by using explicit stochastic models to hedge against uncertain future events. We further propose to investigate an innovative approach, neuro-dynamic programming, for computing closed-loop feedback solutions. We will evaluate the algorithms, and select one for demonstration in a prototype sensor management algorithm operating on a simple air interdiction scenario.