Neural Network-Driven Reliability Analysis in Safety Evaluation of LiDAR-Based Automated Vehicles: Considering Highway Vertical Alignments and Adverse Weather Conditions

Abstract

Complex road environments threaten the safe operation of automated vehicles. Among these, adverse weather conditions and road geometries have particularly significant impacts. This study investigates LiDAR-based automated vehicles (LAVs) driving safety on vertical curved roads in adverse weather. A key methodology involves constructing a failure function that incorporates both the available sight distance (ASD) and the required stopping sight distance (RSD). This function is analyzed using a combined approach of neural networks and Monte Carlo simulations to quantitatively evaluate and generalize the reliability of LAVs under various conditions. The results reveal that variations in weather conditions and vertical curve radii significantly impact the ASD of LAVs, while the influence of speed is relatively minor. Notably, dense fog and rainfall can substantially reduce LAVs’ ASD on vertical curves. Furthermore, the vehicle automation level and speed have a significant impact on driving safety, emphasizing the need for road and operational domain design tailored to LAVs under adverse weather conditions and vertical curve radii.