SIGMA POINT KALMAN FILTER FOR UNDERWATER TERRAIN-BASED NAVIGATION

Abstract

Precise underwater navigation is crucial in a number of marine applications. Navigation of most autonomous underwater vehicles (AUVs) is based on inertial navigation. Such navigation systems drift off with time and external fixes are needed. This paper concentrates on terrain-based navigation, where position fixes are found by comparing measurements with a prior map. Nonlinear Bayesian methods like point mass and particle filters are often used for this problem. Such methods are often computationally demanding. The sigma point Kalman filter (or unscented Kalman filter) is a nonlinear filter that does not resort to local linearizations, representing the probability densities using a few deterministically chosen sigma points. The sigma point Kalman filter is not as computationally demanding as the aforementioned methods, and using real AUV data, the accuracies obtained are comparable to those of the point mass and particle filters.