Title:Who is in Control? Practical Physical Layer Attack and Defense for mmWave based Sensing in Autonomous Vehicles

Abstract:With the wide bandwidths in millimeter wave (mmWave) frequency band that results in unprecedented accuracy, mmWave sensing has become vital for many applications, especially in autonomous vehicles (AVs). In addition, mmWave sensing has superior reliability compared to other sensing counterparts such as camera and LiDAR, which is essential for safety-critical driving. Therefore, it is critical to understand the security vulnerabilities and improve the security and reliability of mmWave sensing in AVs. To this end, we perform the end-to-end security analysis of a mmWave-based sensing system in AVs, by designing and implementing practical physical layer attack and defense strategies in a state-of-the-art mmWave testbed and an AV testbed in real-world settings. Various strategies are developed to take control of the victim AV by spoofing its mmWave sensing module, including adding fake obstacles at arbitrary locations and faking the locations of existing obstacles. Five real-world attack scenarios are constructed to spoof the victim AV and force it to make dangerous driving decisions leading to a fatal crash. Field experiments are conducted to study the impact of the various attack scenarios using a Lincoln MKZ-based AV testbed, which validate that the attacker can indeed assume control of the victim AV to compromise its security and safety. To defend the attacks, we design and implement a challenge-response authentication scheme and a RF fingerprinting scheme to reliably detect aforementioned spoofing attacks.