PCMF: SLAM-based Global Positioning for Autonomous Surface Vessels with
Prior Maps in Harbor Environments
Abstract
This paper presents Point Cloud Map Fitting (PCMF), a SLAM-based method
for robust global positioning of autonomous surface vehicles in harbor
environments. To address the challenge of precise positioning in
unstructured and dynamic environments, a LiDAR sensor provides accurate
measurements of the surroundings, which are fitted to a prior map using
rotational initial alignment and G-ICP detail registration. The initial
alignment process improves the accuracy of subsequent registration by
leveraging corner features of the environment to address the limitations
associated with large transformations in ICP registration. The resulting
position estimates enable redundant positioning estimated by an
Unscented Kalman Filter sensor fusion algorithm together with GNSS. The
pipeline is tested in the field with multiple locations, tides, and
weather conditions to demonstrate efficient position estimation in real
time. The multimodal positioning system is stress-tested through the
manipulation of GNSS signals in a simulated cyber-attack, which serves
to demonstrate the edge case robustness of the system.