TTT SLAM: A feature-based bathymetric SLAM framework

Abstract

This paper presents a new feature-based framework for bathymetric simultaneous localization and mapping (SLAM), called the TTT SLAM. In the frontend of the SLAM framework, we extract and match the handcrafted terrain gradient features designed for bathymetric data from submaps, and we employ the TEASER++ robust point cloud registration method to align overlapping submaps. In the backend, we utilize the Graduated Non-convexity Truncated Least Squares optimization method for pose graph optimization. To compare the performance of the terrain gradient features with existing point cloud features, we conducted feature matching experiments. Subsequently, we evaluated the performance of TTT SLAM and the existing bathymetric SLAM methods using two bathymetric datasets. The results demonstrate that the proposed terrain gradient features achieve a higher inlier rate compared to existing features, and TTT SLAM exhibits robustness and efficiency under different scenarios. To the best of our knowledge, the proposed TTT SLAM is the first bathymetric SLAM framework that utilizes handcrafted features directly extracted from the data collected by multibeam sonars.