Iterated statistical linear regression for Bayesian updates

This paper deals with Gaussian approximations to the posterior probability density function (PDF) in Bayesian nonlinear filtering. In this setting, using sigma-point based approximations to the Kalman filter (KF) recursion is a prominent approach. In the update step, the sigma-point KF approximations are equivalent to performing the statistical linear regression (SLR) of the (nonlinear) measurement function with respect to the prior PDF. In this paper, we indicate that the SLR of the measurement function with respect to the posterior is expected to provide better results than the SLR with respect to the prior. The resulting filter is referred to as the posterior linearisation filter (PLF). In practice, the exact PLF update is intractable but can be efficiently approximated by carrying out iterated SLRs based on sigma-point approximations. On the whole, the resulting filter, the iterated PLF (IPLF), is expected to outperform all sigma-point KF approximations as demonstrated by numerical simulations.