The robust regression approach in EKF for underwater CLMA problem

The extended Kalman filter(EKF) is used extensively for contact location and motion analysis (CLMA) from measurements contaminated by Gaussian noise. However, the additive noise in the time delay measurements is often characterized as heavy-tailed non-Gaussian causing too much tracking error or even divergence of the filter working under the Gaussian error assumption. This paper discusses using the robust regression approach in conjunction with the extended Kalman filter (RrEKF) to improve filter performance. The state estimate in the filter is done using robust regression. We use proposals by Schwappe and Welsch (1980) in the regression process. Monte Carlo simulation results involving many heavy-tailed contaminated observation noise levels demonstrate the robustness of the estimation procedure