Terrain relative navigation using bathymetric slope profile with a Rao-Blackwellized particle filter for long range AUVs

This project designs, develops, and validates a Rao-Blackwellized particle filter (RBPF) for use on long-range AUVs. Most systems currently in use rely on high resolution a-priori maps of less than 10 meter grid spacing and can only compute positional accuracy as good as the map's resolution. This study uses an RBPF with the slope of the bathymetry as its measurement and 5 additional linear states for estimating attitude and velocity biases and errors to achieve higher position accuracy than the underlying accuracy of low resolution maps. This technique will allow the use of more accessible maps and eliminates the tidal influence on the water column height during the time of measurement. The RBPF estimates non-linear states for global position making it unnecessary for the AUV to surface for a GPS fix or have unbounded error drift that occurs in inertial navigation systems (INS) and Doppler velocity log (DVL) navigation. In areas with 5m bathymetry grids the RBPF estimated the position to within 3–5 meters from the true position where the INS estimate was 95m off from true. Areas of very little terrain variability decreased the performance of the RBPF and it only came within 5m–55m from true where the INS was off by 10m–45m. For a medium resolution map with 15m grid spaces the RBPF was 10m–35m from the true position, which is only 1 to 2 reference measurement grid spaces from the true position. Finally, for a low resolution map with a grid size of 55 meters the RBPF's estimated position was only 7 meters off in the Easting direction and 22 meters off in the Northing direction. The INS position estimate for this run was 67 meters from the true position. This result not only supports that the RBPF estimate exceeds the INS during long-range missions but also that it was capable of estimating the position to 1/7 of a reference map grid cell.