Intermittent VIO-Assisted LiDAR SLAM Against Degeneracy: Recognition and Mitigation

Light detection and ranging (LiDAR) simultaneous localization and mapping (SLAM) is widely used in intelligent vehicles for environment perception, self-localization, and mapping. However, in specific environments such as long corridors, highways, or caves, LiDAR SLAM may fail to perceive changes in the surrounding environment, leading to an inability to estimate motion, known as LiDAR degeneracy. LiDAR degeneracy significantly affects the accuracy of LiDAR localization. To address this issue, this article proposes a LiDAR degeneracy recognition algorithm based on a dynamic threshold and mitigates LiDAR degeneracy by introducing intermittent visual-inertial odometry (VIO). We model dynamic thresholds from position accuracy demands through simple mathematical derivations, avoiding heuristic tuning. We also propose an intermittent VIO that operates only during LiDAR degeneracy, achieving comparable or better position performance with minimal computational resource utilization. To improve the intermittent VIO, we use LiDAR for fast initialization of the VIO and visual feature depth association, and we also introduce LiDAR points as features in VIO motion estimation. Through experiments on simulated and public datasets, our method accurately identifies LiDAR degeneracy and effectively mitigates it. Experimental results show that our method outperforms state-of-the-art methods with lower average CPU usage while achieving similar or higher accuracy.