Mutual information based tracking with mobile sensors

Abstract

In order to utilize mobile sensor nodes in a sensing and estimation problem, one must carefully consider the optimal placement of those sensor nodes and simultaneously account for the cost incurred in moving the sensor nodes. We present an approximate dynamic programming approach to a tracking problem with mobile sensor nodes. We utilize mutual information as the objective for optimal sensor placement. We show how a constrained dynamic programming approach allows us to balance estimation quality against mobility costs. However this constrained optimization problem is NP-hard. We present a set of approximations that allow this dynamic program to be solved with polynomial complexity in the number of sensors. We present a greedy multiple time step planning algorithm that greedily selects the most informative paths over a fixed planning horizon. These approximation algorithms are verified via simulation to give a comparative analysis of estimate quality and mobility costs.