Learning sensor-detection policies

Tactical aircraft pilots frequently perform complex sequential support tasks to obtain accurate and timely integrated sensor information about the local environment. When workloads are heavy, offloading these sensor-support tasks to an automated sensor management system would enhance performance and situational awareness. Reinforcement learning, a family of machine learning techniques, offers a way to learn to conduct sensor-support tasks despite sensor complexities by mapping a situation to an action. This paper applies reinforcement learning to a simplified target-detection policy, compares simulated performances of the learned technique to that of an optimal and an uninformed detection policy, and draws conclusions for future research directions