Recent advancements in the automotive field have significantly increased the level of complexity and connectivity of modern vehicles. In this context, the topic of cybersecurity becomes extremely relevant, as a successful attack can have an impact in terms of safety on the car navigation, potentially leading to harmful behavior. Risk assessment is typically performed using discrete input and output scales, which can often lead to an identical output in terms of risk evaluation despite the inputs presenting non-negligible differences. This work presents a novel fuzzy-logic-based methodology to assess cybersecurity risks which takes attack feasibility and safety impact as input factors. This technique allows us explicitly model the uncertainty and ambiguousness of input data, which is typical of the risk assessment process, providing an output on a more detailed scale. The fuzzy inference engine is based on a set of control rules expressed in natural language, which is crucial to maintaining the interpretability and traceability of the risk calculation. The proposed framework was applied to a case study extracted from ISO/SAE 21434. The obtained results are in line with the traditional methodology, with the added benefit of also providing the scatter around the calculated value, indicating the risk trend. The proposed method is general and can be applied in the industry independently from the specific case study.

This paper addresses the verification of neural network robustness against perturbations of input data in vision-based end-to-end autonomous driving systems. The main contributions of this work are: i) We provide a comprehensive analysis of current neural-network-based perception and decision-making components in autonomous vehicles, highlighting their susceptibility to attacks or perturbation. ii) We develop and implement a novel framework for systematically evaluating and verifying the robustness of neural networks against such perturbations, focusing on imperceptible image modifications. iii) We present empirical results demonstrating our verification framework’s effectiveness in identifying weaknesses and providing probabilistic guarantees on network behaviour. Our work underscores the critical need for robust verification methods to ensure the reliability and safety of autonomous driving systems, paving the way for safer integration of neural networks in autonomous vehicles.