Seafloor Elevation Trends as Environmental Features for Data Association in Underwater Simultaneous Localization and Mapping
While critical to the Simultaneous Localization and Mapping (SLAM), process data association is often unreliable, especially so in a noisy, dynamic, underwater environment. This thesis presents a novel approach for data association that enhances underwater SLAM on autonomous underwater vehicles (AUV) using side-scan sonars. It does this by jointly associating the relative position of a landmark to the AUV with the seafloor elevation gradients surrounding the landmark. The local elevation gradients are extracted from the same side-scan sonar images as the landmarks. Seafloor gradients are relatively stable environmental features compared to the much smaller landmarks which can be subject to movement and positional changes over time due to currents and shifting bottom cover. This concept was found to yield correct associations when implemented and validated in post-processing of data using a hardware-in-the-loop AUV simulator and side-scan sonar data from earlier trials. The algorithm has been installed on an IVER3 AUV and in-water trials are validating this concept in a real world setting.