Multi-Modal SLAM for Accurate Localisation in Self-similar Environments

Abstract

Regular inspection and maintenance (I&M) of road tunnels is critical for ensuring safe operation and maximising the infrastructure's longevity. Today's I&M operations are time-consuming and disruptive to normal operations, but advances within robotics, automation, and digitalisation promise significant productivity gains. Accurate and reliable localisation is key to achieving this, but poses significant challenges in tunnels due to the absence of GNSS signals and the self-similar nature of the environment. This paper presents a novel approach for achieving real-time high-accuracy localisation in tunnels such that it can be used for autonomous navigation. The proposed system implements a simultaneous localisation and mapping (SLAM) solution that integrates data from scanning LiDAR, camera and inertial measurement unit (IMU). We have developed a novel approach that fuses the information from these sensors at the feature level and jointly optimises over all constraints. This enables our system to overcome the degeneracy of typical SLAM solutions in self-similar environments such as tunnels. To evaluate the performance of the proposed system, experiments and autonomous missions were conducted in real tunnels, and comparisons were made against existing localisation methods. The results demonstrate that the proposed system achieves high accuracy and exhibits good robustness in challenging tunnel conditions.