Real-Time Anomaly Detection in Connected Autonomous Vehicles: A Data-Driven Approach

The impressive advancements in intelligent and interconnected transportation systems are largely attributed to the emergence of smart vehicles. However, the rise of vehicle connectivity, while offering clear benefits, also brings about significant safety issues that can jeopardize the lives of people. One of the challenging but essential tasks in this field is Anomaly Detection, Identification, and Recovery (ADIR) on connected autonomous vehicles (CAVs). The emergence of recent technologies that enable real-time and onboard processing of large data has made it possible to implement novel approaches in Anomaly Detection (AD) for complex systems. With the increased use of Artificial Intelligence (AI) in solving complex engineering problems in the face of anomaly detection, there has been significant increased use of these methods for systems that experience external disruptive intruding attacks. This chapter outlines some preliminary experimental and proofs-of-concept research work by using a deep autoencoder approach to verify the feasibility of AD task on a platoon of vehicles that are dynamically coupled through communication and feedback control moving in a rectilinear formation in a leader-follower architecture. The AD strategy was implemented on a set of Quanser Cars (QCars) within the Laboratory for Autonomous Resilient Systems (LARS) at the University of Windsor.