Adversarial Robustness of Deep Sensor Fusion Models
This work addresses security vulnerabilities in sensor fusion for autonomous vehicles, though it is incremental as it focuses on experimental analysis without proposing new methods.
The study investigated the adversarial robustness of deep camera-LiDAR fusion models for 2D object detection in autonomous driving, finding that fusion models are generally more accurate and robust than single-sensor networks, but adversarial training can be detrimental or ineffective in boosting robustness.
We experimentally study the robustness of deep camera-LiDAR fusion architectures for 2D object detection in autonomous driving. First, we find that the fusion model is usually both more accurate, and more robust against single-source attacks than single-sensor deep neural networks. Furthermore, we show that without adversarial training, early fusion is more robust than late fusion, whereas the two perform similarly after adversarial training. However, we note that single-channel adversarial training of deep fusion is often detrimental even to robustness. Moreover, we observe cross-channel externalities, where single-channel adversarial training reduces robustness to attacks on the other channel. Additionally, we observe that the choice of adversarial model in adversarial training is critical: using attacks restricted to cars' bounding boxes is more effective in adversarial training and exhibits less significant cross-channel externalities. Finally, we find that joint-channel adversarial training helps mitigate many of the issues above, but does not significantly boost adversarial robustness.