Skip to Main content Skip to Navigation

Machine learning for intrusion detection systems in autonomous transportation

Abstract : Despite all the different technological innovations and advances in the automotive field, autonomous vehicles are still in the testing phase. Many actors are working on several improvements in many domains to make autonomous cars the safest option. One of the important dimensions is cybersecurity. Autonomous vehicles will be prone to cyberattacks, and criminals might be motivated to hack into the vehicles' operating systems, steal essential passenger data, or disrupt its operation and jeopardize the passenger's safety. Thus, cybersecurity remains one of the biggest obstacles to overcome to ensure vehicles safety and the contribution that this technology can bring to society. Indeed, the actual and future design and implementation of Autonomous Vehicles imply many communication interfaces, In-vehicle communication of the embedded system, Vehicle-to-X (V2X) communications between the vehicle and other connected vehicles and structures on the roads. Even though the cybersecurity aspect is incorporated by design, meaning that the system needs to satisfy security standards (anti-virus, firewall, etc.), we cannot ensure that all possible breaches are covered. The Intrusion Detection System (IDS) has been introduced in the IT world to assess the state of the network and detect if a violation occurs. Many experiences and the history of IT have inspired the cybersecurity for autonomous vehicles. Nevertheless, autonomous vehicles exhibit their own needs and constraints. The current state of vehicles evolution has been made possible through successive innovations in many industrial and research fields. Artificial Intelligence (AI) is one of them. It enables learning and implementing the most fundamental self-driving tasks. This thesis aims to develop an intelligent invehicle Intrusion detection system (IDS) using machine learning (ml) from an automotive perspective, to assess and evaluate the impact of machine learning on enhancing the security of future vehicle intrusion detection system that fits in-vehicle computational constraints. Future In-vehicle network architecture is composed of different subsystems formed of other ECUs (Electronic Controller Units). Each subsystem is vehicles. Our primary focus is on In-vehicle communication security. We conduct an empirical investigation to determine the underlying needs and constraints that in-vehicle systems require. First, we review the deep learning literature for anomaly detection and studies on autonomous vehicle intrusion detection systems using deep learning. We notice many works on in-vehicle intrusion detection systems, but not all of them consider the constraints of autonomous vehicle systems. We conduct an empirical investigation to determine the underlying needs and constraints that in-vehicle systems require. We review the deep learning literature for anomaly detection, and there is a lack of tailored study on autonomous vehicle intrusion detection systems using Deep Learning (DL). In such applications, the data is unbalanced: the rate of normal examples is much higher than the anomalous examples. The emergence of generative adversarial networks (GANs) has recently brought new algorithms for anomaly detection. We develop an adversarial approach for anomaly detection based on an Encoding adversarial network (EAN). Considering the behaviour and the lightweight nature of in-vehicle networks, we show that EAN remains robust to the increase of normal examples modalities, and only a sub-part of the neural network is used for the detection phase. Controller Area Network (CAN) is one of the mostused vehicle bus standards designed to allow microcontrollers and devices to communicate. We propose a Deep CAN intrusion detection system framework. We introduce a Multi-Variate Time Series representation for asynchronous CAN data. We show that this representation enhances the temporal modelling of deep learning architectures for anomaly detection.
Document type :
Complete list of metadata
Contributor : Abes Star :  Contact
Submitted on : Tuesday, November 30, 2021 - 11:40:21 AM
Last modification on : Friday, January 28, 2022 - 3:00:07 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03456817, version 1


Elies Gherbi. Machine learning for intrusion detection systems in autonomous transportation. Technology for Human Learning. Université Paris-Saclay, 2021. English. ⟨NNT : 2021UPASG037⟩. ⟨tel-03456817⟩



Les métriques sont temporairement indisponibles