Receding Horizon Control under Uncertainty Using Optimal Input Design and the Unscented Transform

This paper presents a framework for receding horizon control under measurement or sensor uncertainty. Concepts from optimal input design (OID) are combined with the unscented transform (UT) developed for nonlinear estimation. UT algorithms represent probability distributions by a set of representative sample points that capture the first and second moments of the distribution. Using these sample points, the effects of nonlinear operators on a probability distribution can be approximated. This approximation can be used to calculate open-loop feedback control cost functions. Optimal input design enables receding horizon controllers to calculate and optimize a measure of the sensitivity of the estimation process to system inputs, thus enabling improvement of the state estimates during the control process