Terrain aided navigation for autonomous underwater vehicles with coarse maps

Abstract

Terrain aided navigation (TAN) is a form of geophysical localization technique for autonomous underwater vehicles (AUVs) operating in GPS-denied environments. TAN performance on sensor-rich AUVs has been evaluated in sea trials. However, many challenges remain before TAN can be successfully implemented on sensor-limited AUVs, especially with coarse maps. To improve TAN performance over coarse maps, a Gaussian process (GP) is proposed for the modeling of bathymetric terrain and integrated into the particle filter (GP-PF). GP is applied to provide not only the bathymetric value prediction through learning a set of bathymetric data from coarse maps but also the variance of the prediction. As a measurement update, calculated on bathymetric deviation is performed through the PF to obtain absolute and bounded positioning accuracy. Through the analysis of TAN performance on experimental data for two different terrains with map resolutions of 10–50 m, both the ability of the proposed model to represent the actual bathymetric terrain with accuracy and the effect of the GP-PF for TAN on sensor-limited systems in suited terrain are demonstrated. The experiment results further verify that there is an inverse relationship between the coarseness of the map and the overall TAN accuracy in rough terrains, but there is hardly any relationship between them in relatively flat terrains.