Skip to Main Content
This paper presents comprehensive testing and analysis of a particle filter (PF) framework for real-time terrain navigation on an autonomous underwater vehicle (AUV). The goal is to obtain georeferenced localization for an AUV navigation system using reference bathymetry maps and bathymetric measurements in lieu of Global Positioning System (GPS) updates or acoustic localization methods such as long baseline (LBL). The algorithms are tested using navigation and sensor data collected during a long-distance run of the midsize autonomous reconfigurable vehicle (MARV) AUV, during which the Inertial Navigation System (INS) accumulated significant horizontal position error (>;100 m). Run data sets over areas of both high and low bathymetric relief are used to compare performance with different altitude sensors, including multibeam and single-beam sonar, and four-beam Doppler velocity logger (DVL) sonar. A comparison of established techniques for dealing with tide levels is also presented. The technique of adding a third dimension to the PF state vector to compensate for changing water levels allowed the estimator to work consistently in areas of low bathymetric relief, where relative terrain profiling methods broke down and even slight errors in tide prediction models otherwise caused significant position errors. The effects of decreased vertical resolution and accuracy of the reference map is also investigated. To address problems of insufficient sample size, a novel PF resampling technique is presented in which jittering noise added to duplicate samples is scaled based on the particle cloud volume. This technique allows the PF to consistently recover a position fix over a search area of several kilometers without needing to significantly increase the number of particles used.