Information-theoretic tracking control based on particle filter estimate

The contribution of this work is a control formulation for a mobile sensor to track a target using an information-theoretic cost function based on a particle filter estimate of the target state. The particle filter representation fully models the non-linearity and limited field of view of the sensor and is able to search for a lost target by updating the estimate to eliminate areas which have been searched. An entropy calculation is developed which reflects the uncertainty of the particle filtering density for the purpose of tracking, and is then combined with a sampling method to predict the expected entropy of the target state estimate under a proposed control. When sensor motion is constrained, such as for a fixed-wing aircraft, a long planning horizon can provide better performance than single step planning approaches. Exact prediction of information-theoretic costs for non-linear models is not generally feasible in real time, and so approximate methods will be required to predict the expected estimate entropy for receding horizon control. Simulation results demonstrate the accuracy of the prediction method and the effectiveness of the information-theoretic control. Initial experimental results verify the appropriateness of the particle filter for tracking a mobile target from an unmanned aircraft.