Sensor control for search and identification of Markov objects

In this paper, we discuss stochastic control approaches to sensor control problems for the purposes of locating and classifying objects that can enter and leave areas of interest, and there are many objects to interrogate. Noisy sensors with limited energy can choose to interrogate areas to find and identify objects while they are present in the scenario, and can use different modes to either search or identify objects. The goal is to identify objects appearing in the scenario as soon as they are present. Although the resulting stochastic control problem is a partially observed Markov decision problem with combinatorially large action and state spaces, we develop an approximate stochastic control formulation based on relaxing constraints concerning the utilization of sensor energy, and obtain an efficient algorithm for generating near-optimal sensor control decisions. The resulting algorithm is illustrated in a simple scenario with a single sensor observing multiple areas of interest.