Multilayered Framework for Securing Connected Autonomous Vehicle

Abstract

Connected autonomous vehicles (CAVs) hold the promise of not only enhancing functional safety but also improving mobility and the efficiency of transportation systems. The CAV is a cyber-physical system (CPS) that contains many networked electronic control units (ECUs), sensors, actuators, wireless interfaces, and an advanced driver assistant system (ADAS). Just like other CPS, CAVs rely on data gathered from the sensors, actuators, and software for critical decisionmaking to enhance efficiency, reliability, safety, and functionality. Besides, CAVs utilize connectivity to improve drivers’ and passengers’ experience by integrating built-in wireless interfaces like WiFi, Bluetooth, etc. The growing connectivity feature of modern vehicles is marking them more vulnerable to cyberattacks. Many researchers have successfully exploited the remote connectivityinduced attack surface. According to the recent industry report (1) on cyberattacks on CAVs indicated that more than 700 incidents were reported targeting vehicular systems between 2010-2020. Among them, in 27.63% of these incidents, attackers tried to control or manipulate the vehicle which could jeopardize passenger safety. Based on the literature, several intrusion detection-based solutions have been proposed to detect attacks on CAVs. While the solutions are effective against a certain range of attack vectors, they are limited in scope and effectiveness. For instance, existing solutions are unable to localize the attack. In addition, existing state-of-the-art require thousands of in-vehicular network (IVN) packets for intrusion detection. It is therefore important to develop reliable, robust, and real-time security solutions to safeguard CAVs by mitigating emerging cyber threats. This dissertation aims to address the aforementioned cybersecurity challenges of CAVs by developing a robust and reliable framework to safeguard against attacks at different points through a multi-layered framework. Each layer of the proposed solution aims at neutralizing cyberattacks on in-vehicle networks by breaking some critical links in the attack chain. The first layer of the proposed framework aims to protect IVNs by developing a sender identification algorithm that utilizes the unclonable signal attributes to fingerprint transmitting ECUs. The proposed framework is novel and efficient that leverages the uniqueness of physical signals to create images and uses a deep learning algorithm for attack detection and localization. The second layer aims to protect IVNs against firmware attacks using ECU behavioral fingerprinting through a data-driven graph theory-based approach. The proposed methodology takes advantage of a huge amount of IVN data to model the normal behavior of the network by using graph analytics and develops a network monitoring system to detect unusual behavior created by attackers. The effectiveness of the proposed multilayered framework is evaluated by conducting a series of experiments using bench testing and as well on vehicular public data. The experimental results suggest that the proposed multilayered framework is capable of detecting IVN message injection attacks with higher accuracy and can reliably localize the attacker on the network. Additionally, the thesis hypothesis and solution were validated by conducting market research through active participation in both the regional and final programs of the National Science Foundation (NSF) I-Corps. In the future, I plan to build a prototype of the proposed framework and deploy it in actual vehicles for rigorous field-testing.