Identifying and Patching Vulnerabilities of Camera-LiDAR Based Autonomous Driving Systems

Details

• Creators:
  Xie, Cihang ; Kantarcioglu, Murat ; Cardenas, Alvaro
• Corporate Creators:
  National Center for Transportation Cybersecurity and Resiliency (TraCR) National University Transportation Center (UTC) ; Clemson University ; University of California, Santa Cruz
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  Autonomous Vehicles Cameras Computer Security Laser Radar Machine Learning Physical Phenomena Safety Sensors
• Publication/ Report Number:
  7
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report, 01/01/2024 - 12/31/2024
• Corporate Publisher:
  National Center for Transportation Cybersecurity and Resiliency (TraCR) National University Transportation Center (UTC)
• Abstract:
  Autonomous driving systems rely on advanced perception models to interpret their surroundings and make real-time driving decisions. Among these, Bird’s Eye View (BEV) perception has emerged as a critical component, offering a unified 3D representation from multi-camera and sensor inputs. While BEV-based models have gained traction in industry-leading platforms, their security vulnerabilities remain largely underexplored in adversarial machine learning research. This study provides a multi-dimensional security analysis of BEV perception models, focusing on adversarial threats in both vision-only and multi-sensor fusion architectures. We examine the susceptibility of state-of-the-art models - including BEVDet, BEVDet4D, DAL, and BEVFormer—to adversarial attacks targeting their detection and decision-making capabilities. Unlike traditional adversarial research that primarily misleads perception models at the classification level, this study investigates real-world attack scenarios where adversaries can manipulate perception to cause practical disruptions, such as inducing traffic congestion or triggering unsafe vehicle behaviors. Our findings reveal significant security risks in BEV-based perception, with both vision-only and sensor-fusion models vulnerable to adversarial perturbations. Attack transferability across architectures further highlights the urgency of developing robust defense mechanisms to ensure the reliability of self-driving technology. This work underscores the need for adversarially resilient perception models to safeguard the future of autonomous driving.
• Format:
  PDF
• Funding:
  69A3552344812, 69A3552348317
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:5de38b75d50efe22ce97a1a51ab1d4fa5d5bc009352a4c5a40f7910f23aa91659dbd960a035fbfef067f82400d23bcdef9ca16794e3d60882479fe6a61fd5d8a
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/86152/dot_86152_DS1.pdf
• File Type:
  [PDF - 1.71 MB ]