Exploiting Rauch-Tung-Striebel formulae for IMM-based smoothing of Markovian switching systems

This paper presents a suboptimal fixed-interval smoothing algorithm for nonlinear Markovian switching systems. The posterior smoothed mean and covariance of the system state are approximated by combining in a backward-time recursive process the statistics produced by a forward-time Interacting Multiple Model filter. Each recursion of the backward-time process consists of a smoothing step based on Rauch-Tung-Striebel formulae and of a specific interaction step to allow mode cooperation. A simulated case study in the field of target tracking illustrates the method.