Non-myopic sensor scheduling for low radiation risk tracking using mixed POMDP

Abstract

This paper addresses a non-myopic sensor-scheduling problem of how to select and assign active sensors for trading off the tracking accuracy and the radiation risk, where the radiation risk is incurred by the fact that the emission energy originating from active sensors for target tracking can be intercepted by the enemy target. This problem is formulated as a mixed partially observable Markov decision process (POMDP) composed of a continuous-state POMDP for target tracking and a discrete-state POMDP for emission control. Based on the idea of foresight optimization, the long-term accuracy reward is evaluated by the combination of unscented transformation sampling and Kalman filtering, whereas the long-term radiation cost is derived from hidden Markov model filter. Because the problem can be converted into a decision tree, a branch and bound algorithm is developed for problem solution. A simulation example illustrates the effectiveness of our approach.