Improved Progressive Gaussian Filtering Using LRKF Priors

In this paper, we present a new extension to the Progressive Gaussian Filter (PGF) to improve its performance in certain situations. The PGF is a nonlinear Gaussian-assumed filter. Its basic idea is to gradually take the likelihood of the measurement into account when performing the measurement update step. Because the filter does not assume the prior and the measurement to be jointly Gaussian distributed, it outperforms commonly used LRKFs in strongly nonlinear scenarios. However, it suffers from suboptimality in cases where the model is close to linear, as the PGF does not contain the Kalman filter as a special case. To remedy this issue, we propose an extension where the PGF is combined with an LRKF.