Ground-optimized SLAM with Hierarchical Loop Closure Detection in Large-scale Environment

Accurate and reliable localization and mapping are crucial prerequisites for autonomous driving to achieve path planning. However, in large-scale complex dynamic environments, the traditional LiDAR loop closure detection methods that rely on radius search can easily result in false negatives, leading to the inability to correct accumulated errors effectively. To address this issue, a hierarchical LiDAR descriptor loop closure detection strategy is proposed in this paper, which detects invalid loop closures and has good viewpoint invariance. We integrate this strategy into an advanced LiDAR Inertial tightly coupled SLAM framework. In addition, to reduce the drift in the vertical direction during mapping, we introduce a ground marking algorithm and construct corresponding ground constraints in the back-end optimization. Our proposed method is evaluated on the MulRan dataset, and the experimental results show that our method could achieve lower accumulated errors than competing methods.