Exploring Adversarial Robustness of LiDAR-Camera Fusion Model in Autonomous Driving

Our study assesses the adversarial robustness of LiDAR-camera fusion modelsin 3D object detection. We introduce an attack technique that, by simply addinga limited number of physically constrained adversarial points above a car, canmake the car undetectable by the fusion model. Experimental results reveal thateven without changes to the image data channel, the fusion model can bedeceived solely by manipulating the LiDAR data channel. This finding raisessafety concerns in the field of autonomous driving. Further, we explore how thequantity of adversarial points, the distance between the front-near car and theLiDAR-equipped car, and various angular factors affect the attack success rate.We believe our research can contribute to the understanding of multi-sensorrobustness, offering insights and guidance to enhance the safety of autonomousdriving.