Are Self-Driving Cars Secure? Evasion Attacks Against Deep Neural Networks for Steering Angle Prediction

Abstract

Deep Neural Networks (DNNs) have tremendous potential in advancing the vision for self-driving cars. However, the security of DNN models in this context leads to major safety implications and needs to be better understood. We consider the case study of steering angle prediction from camera images, using the dataset from the 2014 Udacity challenge. We demonstrate for the first time adversarial testing-time attacks for this application for both classification and regression settings. We show that minor modifications to the camera image (an L2 distance of 0.82 for one of the considered models) result in mis-classification of an image to any class of attacker's choice. Furthermore, our regression attack results in a significant increase in Mean Square Error (MSE) -by a factor of 69 in the worst case. 1

Software Verification and Validation of Safe Autonomous Cars: A Systematic Literature Review

“…Reference [38] suggests further research and experiments (with popular datasets such as MNIST) on using affine transformations to find adversarial examples. Although there are extensive research efforts on how to find adversarial examples in DNNs, how to avoid these attacks is still an open issue [24].…”

Towards Low-Latency Service Delivery in a Continuum of Virtual Resources: State-of-the-Art and Research Directions

“…Self-driving cars (Sec. VIII-C) Maqueda, A. I., et al [171] Steering Prediction 2018 Chen, S., et al [172] Testing platform 2019 Chernikova, A., et al [173] Security 2019 Ndikumana, A., et al [174] Caching for MEC 2020 Extended Reality (XR) (Sec. VIII-D) Liu, Y., et al [175] MEC-assisted VR 2018 Doumanoglou, A., et al [176] Quality of Experience 2018 van der Hooft, J., et al [177] Adaptive VR services 2019 van der Hooft, J., et al [178] Point cloud compression 2019 Industrial IoT (IIoT) (Sec.…”

The Driver Time Memory Car-Following Model Simulating in Apollo Platform with GRU and Real Road Traffic Data

“…e mean square error (MSE) (21) was used to construct the loss function and the Adam optimizer [24][25][26], due to its excellent performance in most cases, was adopted.…”