Tunnel-LIVO: LiDAR-Inertial-Visual Odometry for Degenerate Tunnel Environments

In large tunnel environments such as the water diversion and spillway tunnels in hydropower stations, various SLAMs are prone to degradation. In this paper, we propose Tunnel-LIVO, an odometry designed specifically for tunnels, which integrates both LIO and VIO subsystems. In contrast to existing advanced odometrys, our approach is specifically tailored for scenarios characterized by single-directional translational degradation. By constructing a visual map aligned with the uninformative direction, it improves odometry stability under such challenging conditions. The VIO subsystem utilizes point cloud normal vectors to extract the tunnel degradation direction, and we construct a map-building strategy based on the contribution of the degraded direction. Additionally, we construct a sparse direct frame-to-map alignment method that favors the degraded direction. The LIO subsystem constructs a scan-to-map point-to-plane matching residual and constrains pose estimation along the degraded direction. Experiments conducted in the real-world water diversion and spillway tunnels of a hydropower station demonstrate that our system exhibits robust performance, even when other advanced odometry systems degrade.