Pose-Graph SLAM Using Forward-Looking Sonar
Jie Li · Michael Kaess · Ryan M. Eustice · Matthew Johnson-Roberson
TL;DR
A real-time sonar-based SLAM system uses learned loop closure proposals and constraint-based data association to localize underwater robots during ship hull inspection.
BRIEF
Underwater robots inspecting ship hulls often rely on imaging sonar, since cameras struggle in murky water. Sonar images, though, have few distinctive features and easily produce false matches, which makes it hard to recognize previously visited places and correct drift. This work builds a real-time pose-graph SLAM system around a forward-looking sonar. A machine learning method proposes loop closures based on which image regions stand out, and a data association step checks matches against several available constraints. It is evaluated on real ship hull inspection data.
Abstract
This letter reports on a real-time simultaneous localization and mapping (SLAM) algorithm for an underwater robot using an imaging forward-looking sonar and its application in the area of autonomous underwater ship hull inspection. The proposed algorithm overcomes specific challenges associated with deliverable underwater acoustic SLAM, including feature sparsity and false-positive data association when utilizing sonar imagery. An advanced machine learning technique is used to provide saliency-aware loop closure proposals. A more reliable data association approach using different available constraints is also developed. Our evaluation is presented on the real-world data collected in a ship hull inspection application, which illustrates the system's performance and robustness.