LiDAR-based Traversability Map Construction Method for Off-Road Environments

Unmanned ground vehicles operating in complex off-road environments face challenges such as unstable ground segmentation and limited semantic representation in traversability maps. This paper presents a LiDAR-based method for constructing dense traversability maps. A dual-channel ground segmentation framework is developed by combining block-wise plane fitting and gradient-based filtering, achieving a balance between precision and recall under challenging terrain conditions. Local geometric descriptors, including flatness, slope, and multi-scale terrain roughness, are designed to quantify terrain traversability. A Traversability Index (TI) is defined to provide a vehicle-aware assessment of terrain passability. By integrating multi-frame point cloud fusion with Bayesian Kernel Inference (BKI), a dense 3D voxel map enriched with traversability semantics is generated. Experimental results demonstrate that the proposed approach achieves high stability and real-time performance in both typical off-road and urban scenarios, offering reliable environmental support for path planning and decision-making in autonomous driving.