A SLAM-based approach for underwater mapping using AUVs with poor inertial information

This paper presents a SLAM-based approach for creating maps of underwater terrain using AUVs with poor inertial information. The initial motivating application for this work was mapping in the non-inertial frame of a free-drifting Antarctic iceberg, but poor inertial information can also occur if low-cost, high drift inertial instrumentation is used in standard mapping tasks, or if DVL bottom lock is lost during the mission. This paper presents a SLAM-based approach in which features are extracted from concatenated multibeam data and descriptors are created, allowing these features to be compared against past terrain as the vehicle traverses the area. There have been a number of previous research efforts that used feature-based SLAM techniques for underwater mapping, but they have generally made assumptions or relied on sensors that are inconsistent with this paper's motivating application, such as a flat bottom, the availability of visual imagery, or manmade fiducial markers. The method presented here uses natural terrain, is robust to water turbidity, and can be used in areas with vertical terrain like the walls of canyons and icebergs. Results are presented on data collected from Monterey Canyon using a vehicle with a high-grade IMU but that lost DVL bottom lock during the mapping run.