Off-line and Real-time Optimization for Sensor Management

Sensor management presents a problem of great importance to the Department of Defense. New sensors, harder-to-detect platforms, increasingly sophisticated threats, and the need to reduce friendly fire incidents through accurate target identification are increasing the need to effectively and rapidly collect and exploit sensor data so that appropriate actions can be taken. However, the complexity of real world sensor data and scenarios has challenged efforts to automatically manage sensors. Specifically, ad hoc approaches, as well as theoretically sound approaches, have focused on on-line computation of option prioritization and schedules. Because of the real-time nature of sensor management, these approaches have been "myopic," i.e., they have focused their computations on the immediate rewards of decisions without looking ahead to see the longer-term consequences of the decisions. The objectives of the proposed project is to design and document an effective software architecture for non-myopic sensor management and sense task scheduling. This real-time sensor manager will combine state-of-the-art incremental decision-theoretic on-line methods with off-line learning of plausible actions and search control heuristics.