Quantitative Evaluation of Automated Vehicles Operations under Adverse Weather Conditions

Abstract: The Pennsylvania Department of Transportation (PennDOT) recently passed Act 130, allowing highly automated vehicles (HAVs) to operate on public roads and opening the door for HAV companies to test and deploy their technology in Pennsylvania. HAV testing is often conducted in locations with consistently favorable weather, such as Arizona and Texas, where clear skies and predictable road conditions minimize complications. However, significant operational issues may arise when HAVs are deployed in regions with more challenging climates like Pennsylvania, where rain, fog, and snow are common. This study investigates the impact of adverse environmental conditions on key HAV functionalities, focusing specifically on perception and control modules. By examining how HAVs behave in more realistic, challenging settings, we aim to identify vulnerabilities in HAV technologies that could compromise safety and functionality in regions with unpredictable weather, and to guide state DOT like PennDOT in better setting technical criteria for market entry.

To conduct this evaluation, we constructed a simulation framework using CARLA, integrating computer vision and sensor models such as YOLO and PointPillars for the vehicle's perception system, and a PID controller for motion control. The performance degradation of these algorithms under various environmental disturbances—such as obscured objects and low visibility—was systematically measured to assess the overall impact on HAV performance. This comprehensive approach allows us to quantify the effects of adverse weather on HAV systems and propose solutions for improving the resilience of HAV operations under real-world conditions.