A Novel Generative AI-Based Framework for Resilient False Data Injection Attack Mitigation in Autonomous Vehicles

Abstract

Every year, millions of people die or are injured due to road traffic accidents; the most common reasons for these accidents are overspeeding, fatigued driving, distracted driving, and drugs or narcotics. To reduce these risks, several research studies are taking place in vehicle automation. The impacts of Autonomous vehicles (AVs) on contemporary transportation depend on how they can change the status of transportation, including human errors, inefficiencies, and traffic congestion. However, like all sophisticated IT systems, AVs remain vulnerable to remote compromise, which can pose significant risks to their safety, security, and users’ personal privacy. The most severe threats are disc-shaped False Data Injection (FDI) attacks. These are false or intentionally submitted data designed to influence critical decisions or the operation of an AV. Such a data grab could compromise AV users in a very significant way. Thus, efficient schemes to prevent and detect false data injection while maintaining minimal intervention in the vehicle dynamics are desirable. In the case of FDI attacks, the data loss rate is reduced. Despite the challenges, Generative Artificial Intelligence (Generative AI) presents a novel approach to address the problem of malicious data filtration while processing only accurate and clean data in real time. The existing FDI detection techniques have shortcomings such as output delay and high algorithmic cost, with large demand for machine resources and high latency, which contradict the requirements of intelligent driving. This paper focuses on the crucial role played by Generative AI in assisting detection and fraud data injection defense in AVs. The method adopted depends on the use of AI algorithms that can consider interference  (resource restrictions). In this paper, a generative-AI-based approach was incorporated for anomaly detection and FDI attack mitigation in AVs. AI techniques were applied to raw inputata to avoid high latency and keep a low rate of error. With the proposed classes of incoming data being legitimate, partially compromised, or fully compromised, it is possible to delineate good data, slightly compromised data, and totally bad data. This classification improves AV safety while maintaining its performance level. In general, on comparing the presented solution with existing methodologies, it can be seen that more accurate results are obtained by the presented solution, and it also provides higher reliability, with less computation. Thus, substantial testing and validation verify that the proposed system does not diminish data coherence, with an acceptable compromise of system performance. In addition, this approach can be generalized for detection of other threats in AVs by adding more input variables and adjusting the Generative AI model for counteracting various types of malicious data injection. Thus, the proposed framework can be regarded as significant progress towards ensuring the sustainability of autonomous transportation systems against modern cyber threats.