Learning Adversarial Attacks on Adaptive Traffic Signal Control Systems Under Cooperative Perception

Explore a work-in-progress research presentation examining cybersecurity vulnerabilities in modern adaptive traffic signal control systems that utilize cooperative perception with connected and automated vehicles (CAVs). Discover how researchers from Purdue University developed a novel reinforcement learning-based black-box adversarial attack framework to test the security of state-of-the-art traffic control systems. Learn about the multi-action proximal policy optimization (multi-PPO) algorithm used to train attacker agents capable of generating fake CAVs and their falsified vehicle detection data. Understand how these sophisticated attacks can deceive learning-based traffic control systems, with experimental results showing a significant 62.5% increase in average vehicle delay when fake CAVs inject falsified detection information. Gain insights into the growing cybersecurity challenges facing intelligent transportation systems as they become increasingly integrated with advanced sensing and communication technologies, and examine the implications for traffic management in an era of connected and automated vehicles.