Enhancing the Robustness of LiDAR-based Object Detection under Disappearing Attacks

Autonomous driving systems rely on LiDAR-based 3D object detection to identify obstacles. Recent studies have shown that detectors are susceptible to disappearing attacks, leading to missed detections and potential vehicle collisions. However, improving the adversarial robustness of 3D object detection against such attacks remains an open question. Our work seeks to bridge this gap by proposing an effective defense strategy for 3D object detection under both black-box and white-box disappearing attacks. Specifically, we formulate the problem of defending against disappearing attacks in 3D object detection as a bilevel min-max problem and introduce a novel approach, DART, to train machine learning models robust to disappearing attacks. Our extensive evaluations demonstrate that DART outperforms general adversarial defense methods, reducing the attack success rate by over 85% while achieving a better balance between robustness and performance, and meeting the real-time requirements of autonomous vehicles.

I. Introduction

LiDARs, which capture high-resolution 3D point clouds (PCs) to provide shape and depth information for object detection, are crucial for guaranteeing the performance and safety of autonomous vehicles (AVs). LiDAR-based 3D object detection has achieved superior performance owing to advancements in deep learning technology. However, deep neural networks are vulnerable to adversarial examples [1]-[8]. Prior studies have explored the security of LiDARs and introduced disappearing attacks [9]-[12] to investigate vulnerabilities in 3D object detection [13]-[16]. Attackers typically manufacture adversarial objects around the target to make it invisible to detectors, potentially causing severe safety incidents, e.g. collisions and casualties, as illustrated in Fig. 1.