Information-Theoretic Sensor Motion Control for Distributed Estimation
Allison D. Ryan, Hugh F Durrant-Whyte, J. Karl Hedrick · 2007
Estimate uncertainty is a clear metric for sensing problems, but is not traditionally used for optimal control of mobile sensors because it is difficult to model how it is affected by sensor motion. This work develops a multiple-step receding horizon cost for sensor motion control based on minimization of expected entropy of the estimate distribution. The structure of the cost function is analyzed and used to upper bound the degree of coupling between sensors. The contribution is a multiple step prediction of the estimate entropy incorporating probabilistic sensor and target motion models and a decomposition for its decentralized calculation. Multiple-step receding horizon control has the potential to provide better performance than single-step optimization in the cases of delayed payoff or sensor motion constraints, and has not generally been implemented for non-Gaussian models. An example is developed based on a team of unmanned air vehicles carrying vision sensors, and initial simulation results confirm the accuracy of the predictive cost calculation.