Generative AI Powered Shadow-Resilient Real-Time Pothole Detection in Autonomous Vehicles

With the rise of Industry 4.0, the integration of Artificial Intelligence (AI) into autonomous systems is transforming modern transportation, prioritizing safety, efficiency, and real-time decision-making. Autonomous Vehicles (AVs) heavily rely on sensor data and AI-driven perception systems to navigate roads, detect hazards, and make split-second trajectory adjustments. However, pothole detection in shadowed environments remains a critical challenge, as traditional models often misclassify road defects due to varying illumination conditions. To address this, we propose a pothole detection framework powered by Generative AI (GAI) that is shadow-resilient and enhances AV perception in complex lighting scenarios. Our approach utilizes Cycle-Consistent Generative Adversarial Network (CycleGAN)-based shadow synthesis to generate realistic training data and lightweight You Only Look Once (YOLO) variants to achieve high-precision real-time detection. Furthermore, cooperative perception through Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication ensures timely alerts and cloud-based updates, enabling AVs to dynamically adjust their navigation. Experimental results show that our model achieves a mean Average Precision (mAP) of 0.883 while maintaining an inference speed of 2.3 ms per frame using YOLOv11n, with strong generalization in real-world conditions, achieving 0.864 mAP under diverse on-road environments, including shadowed potholes and varying illumination. By integrating AI perception, cooperative intelligence, and cloud-based adaptability, our framework establishes a new benchmark for robust AV navigation in the era of Industry 4.0.