Incorporating prior maps with Bathymetric Distributed Particle SLAM for improved AUV navigation and mapping

We propose a method to improve the geo-referenced accuracy and self-consistency of bathymetric maps generated by autonomous underwater vehicles (AUVs), where the navigation solution is prone to drift when GPS or other methods of absolute positioning are unavailable. This is accomplished using a non-feature based approach to simultaneous localization and mapping (SLAM) that utilizes a 2D grid structure to represent the map and a distributed particle filter to track the uncertainty in the vehicle state. Our method does not need to explicitly identify features in the surrounding environment or apply complicated matching algorithms to our bathymetry, as is commonly done when performing terrain aided navigation. In this work we demonstrate how a prior low-resolution map generated by a surface vessel in a standard gridded form can be readily integrated into our approach to bathymetric SLAM to additionally enforce consistency between the prior map and the AUV bathymetry. We illustrate the proposed approach using data from recent survey work undertaken off the coast of the Tasman Peninsula in South Eastern Tasmania, Australia. The results achieved by the bathymetric distributed particle SLAM (BPSLAM) filter are shown to improve the maps and trajectories when compared to dead reckoning fused with USBL observations.