Knowledge representation in goal-motivated decision making for intelligent active sonar

Abstract

We consider knowledge representation in a goal-driven model for intelligent, or cognitive, active sonar designed to improve system performance and reduce sonar operator burden. The intelligent sonar uses observations of the environment (broadly defined) and predictions of future behavior to make decisions that leverage system resources to address a set of goals. These goals are created and managed by the system itself with possible input from the sonar operator. Decisions take the form of action selections, where actions may include transmit waveform, ping interval, illuminated region, etc. Each action is assigned a utility score, or metric, based on how well it addresses each system goal, with weighting according to how simultaneous (possibly competing) goals are prioritized by the system and/or operator. Metrics for evaluating candidate actions rely on signal processing models for localization and tracking in active sonar. To illustrate the impact of the structure of utility metrics on system behavior, we study the performance and decision-making behavior of the intelligent sonar system in simulated scenarios in which the system is tasked with addressing multiple search and track goals simultaneously. The effect of the chosen metric structure on the impact of goal prioritization and operator input is explored and discussed.