Anomaly-based Intrusion Detection System for Autonomous Vehicles

Abstract

The rapid development of autonomous vehicles has yielded significant advancements and benefits for the transportation industry. However, despite the extensive progress in autonomous vehicle technology, the emphasis has primarily been on improving vehicular autonomy rather than addressing security concerns. As a result, autonomous vehicles remain vulnerable to various cyber threats and sensor attacks. This thesis seeks to bridge this gap by developing an Intrusion Detection System (IDS) based on behavior analysis of an unmanned autonomous ground vehicle. The primary objective is to design and implement a comprehensive IDS capable of accurately detecting the attacks targeted at the perception system of autonomous vehicles. To achieve this, the thesis introduces a framework based on Anomaly Behavior Analysis that relies on temporal features extracted from a physics-based autonomous vehicle model to capture the normal static and dynamic behaviors of vehicular perception. It employs a combination of model-based techniques and machine-learning algorithms to distinguish between normal and abnormal behaviors. The study focuses on two specific perception system attacks: Global Positioning System (GPS) spoofing and depth camera blinding. These attacks were performed on autonomous vehicle testbeds, enabling the collection of real-world vehicular data that encompasses both normal and abnormal behaviors. Using these datasets, the anomaly-based perception attack detection system is developed and evaluated. Unlike the existing approach of utilizing separate IDSs to detect the perception system attacks, our proposed anomaly-based IDS exhibits a high level of accuracy in detecting both attack types, thereby reducing system complexity. Furthermore, the datasets generated during this research are the first of their kind and have been made publicly available for the research community to assess their IDSs effectively.