DR‐ICP: Robust LiDAR SLAM Registration for Degenerate Tunnels

Light Detection and Ranging (LiDAR) has been extensively deployed in structured environments, such as urban roadways and indoor facilities, owing to its superior ranging accuracy and robust anti‐interference capabilities. However, its performance degrades significantly in unstructured environments like underground tunnels, in such scenarios, the scarcity of distinctive geometric features and insufficient spatial constraints frequently induce degeneracy in LiDAR‐based Simultaneous Localization and Mapping (SLAM), leading to severe registration drift. To address this challenge, this paper proposes a Degeneracy‐Robust Iterative Closest Point (DR‐ICP) method that integrates point cloud normal analysis with multi‐scale curvature constraints. The proposed approach first utilizes a degeneracy detection mechanism to ascertain the occurrence and direction of degeneracy by exploiting statistical properties of point cloud normals. Building on this detection, an adaptive multi‐scale strategy is applied to extract subtle concave‐convex and linear features from optimal neighborhoods. Furthermore, a weighted registration scheme ensures reliable alignment in degenerate regions by enhancing the representation of valid features. Extensive simulations and real‐world tunnel experiments demonstrate that the proposed DR‐ICP effectively identifies degenerate directions and achieves superior robustness and positioning accuracy in complex underground scenarios.