Rao-Blackwellised Point-Mass Smoothers for a Class of Conditionally Linear Dynamic Models

The paper deals with the state estimation of nonlinear stochastic dynamic systems. The stress is laid on the numerical solution to the Bayes' rule considering a class of conditionally linear Gaussian models typically appearing in navigation. In particular, three novel Rao-Blackwellised smoothers are proposed, where the nonlinear part of the model is solved by a computationally expensive point-mass smoother, whereas the conditionally linear part is solved by a set of linear smoothers. The proposed smoothers offer a tradeoff between the computational complexity and smoothing performance. The properties of the smoothers are theoretically analysed and discussed.