Evaluating the impact of vehicle automation on the safety of highway design: A 3D risk assessment approach using reliability theory
本研究利用三维激光雷达数据和可靠性理论，评估了不同自动化水平的车辆在高速公路弯道上的视距风险，发现自动化总体上降低风险，但在某些弯道因速度和减速度假设而风险升高，为基础设施改进和算法优化提供了依据。
Empirical quantification of how automation affects road safety, particularly whether current highway designs can accommodate autonomous vehicles (AV), remains under-researched. This study addresses the gap using a 3D risk assessment framework that integrates reliability theory with mobile Light Detection and Ranging (LiDAR) data to examine the interaction between sight distance and vehicle automation. Using 308 curves along a rural highway in Canada, a three-phase analysis was conducted. First, available sight distance (ASD) was estimated using 2D analytical formulas and a voxel-based 3D LiDAR approach. Second, a reliability-based risk assessment compared ASD with the stopping sight distance (SSD) requirements of three vehicle types representing increasing automation levels—human-driven vehicles (HDVs), transition-stage AVs, and fully developed AVs—using probability of non-compliance (P nc ) as a risk index. Finally, sensitivity analyses assessed the effects of operational parameters and sensor configurations on risk levels. Results show that the 3D method provided a more realistic and context-sensitive evaluation of ASD and associated obstructions than the 2D method, and was therefore used in the risk assessment. Overall risk decreased with increasing automation, although some scenarios showed elevated risk for fully developed AVs. Segment- and curve-level analysis attribute these increases to higher speeds and gentler deceleration rates assumed on sharper curves. Sensitivity analyses show that higher deceleration rates substantially reduce AV risk, while increased sensor height offers limited benefits. Overall, this study demonstrates the value of LiDAR-based assessment and P nc as a quantitative risk index, enabling identification of critical locations to guide highway infrastructure improvements and AV algorithm refinement.