Is Your Autonomous Vehicle Safe? Understanding the Threat of Electromagnetic Signal Injection Attacks on Traffic Scene Perception

TL;DR

This work addresses the vulnerability of camera-based perception in autonomous vehicles to Electromagnetic Signal Injection Attacks (ESIA), which distort captured images and can mislead AI decisions. It introduces a scalable ESIA simulation method and a corresponding simulated attack dataset, enabling extensive robustness evaluations across diverse driving scenarios without costly real-world attacks. Experiments show that perception performance, including traffic object detection and drivable area segmentation, degrades progressively with attack severity, with distinct patterns across weather, scene, and time-of-day conditions, highlighting concrete safety risks such as contraflow driving. The study provides a practical simulation and evaluation framework to guide the development of robust, secure perception systems and motivates future work on multi-sensor fusion and mitigation strategies for safer autonomous driving.

Abstract

Autonomous vehicles rely on camera-based perception systems to comprehend their driving environment and make crucial decisions, thereby ensuring vehicles to steer safely. However, a significant threat known as Electromagnetic Signal Injection Attacks (ESIA) can distort the images captured by these cameras, leading to incorrect AI decisions and potentially compromising the safety of autonomous vehicles. Despite the serious implications of ESIA, there is limited understanding of its impacts on the robustness of AI models across various and complex driving scenarios. To address this gap, our research analyzes the performance of different models under ESIA, revealing their vulnerabilities to the attacks. Moreover, due to the challenges in obtaining real-world attack data, we develop a novel ESIA simulation method and generate a simulated attack dataset for different driving scenarios. Our research provides a comprehensive simulation and evaluation framework, aiming to enhance the development of more robust AI models and secure intelligent systems, ultimately contributing to the advancement of safer and more reliable technology across various fields.

Figures (6)

• Figure 1: A general process of autonomous driving systems consists of three modules: Perception, Planning, and Control. Electromagnetic signal injection attacks (ESIA) can manipulate the output image of the autonomous vehicle's cameras, hiding the motorbike from being detected, potentially leading to an accident.
• Figure 2: The simulation process generates arbitrary adversarial patterns, i.e., color strips, in an RGB image.
• Figure 3: Example real (left) and simulated (right) attack images with different levels of attack severity.
• Figure 4: Similar performance ($\Delta$mAP50) between (a) real attack images and (b) simulated attack images across different object detection models.
• Figure 5: Model attention variations across attack intensities in the case of "driving against traffic”.