Comparison of Several Nonlinear Filters for a Benchmark Tracking Problem
Christopher D. Karlgaard, Hanspeter Schaub · AIAA Guidance, Navigation, and Control Conference and Exhibit · 2006
This paper discusses the application of a robust filtering technique to a benchmark trajectory estimation problem involving radar measurements of an atmospheric entry vehicle. The robust technique is based on the Huber estimator, which is a mixed minimum ℓ1/ℓ2-norm linear regression technique. The approach is robust with respect to deviations in the traditionally assumed Gaussian measurement error probability distributions. It is shown how the Huber technique can be applied to the Extended Kalman Filter and to First and Second-Order Divided Difference Filter measurement update equations to provide robustness against non-Gaussian errors. The filters are applied to the benchmark problem using Monte-Carlo techniques, and computational costs associated with the filtering techniques are provided. I.