AI/ML-based intrusion detection systems (IDSs) and misbehavior detection systems (MDSs) have shown great potential in identifying anomalies in the network traffic of networked autonomous systems. Despite the vast research efforts, practical deployments of such systems in the real world have been limited. Although the safety-critical nature of autonomous systems and the vulnerability of learning-based techniques to adversarial attacks are among the potential reasons, the lack of objective evaluation and feasibility assessment metrics is one key reason behind the limited adoption of these systems in practical settings. This survey aims to address the aforementioned limitation by presenting an in-depth analysis of AI/ML-based IDSs/MDSs and establishing baseline metrics relevant to networked autonomous systems. Furthermore, this work thoroughly surveys recent studies in this domain, highlighting the evaluation metrics and gaps in the current literature. It also presents key findings derived from our analysis of the surveyed papers and proposes guidelines for providing AI/ML-based IDS/MDS solution approaches suitable for vehicular network applications. Our work provides researchers and practitioners with the needed tools to evaluate the feasibility of AI/ML-based IDS/MDS techniques in real-world settings, with the aim of facilitating the practical adoption of such techniques in emerging autonomous vehicular systems.

翻译：基于人工智能/机器学习（AI/ML）的入侵检测系统（IDS）与异常行为检测系统（MDS）在识别网络化自主系统流量中的异常方面展现出巨大潜力。尽管已有大量研究成果，但这些系统在实际环境中的部署仍然十分有限。虽然自主系统的安全关键特性以及学习技术对对抗性攻击的脆弱性是潜在原因之一，但缺乏客观评估与可行性评估指标是限制这些系统在实践场景中得以广泛应用的关键因素。本综述旨在通过深入分析基于AI/ML的IDS/MDS，并为网络化自主系统建立基线评估指标，以弥补上述不足。此外，本文全面梳理了该领域的最新研究，重点分析了现有文献中的评估指标及其不足。基于对综述论文的分析，本文提出了关键发现，并为适用于车载网络应用的AI/ML驱动的IDS/MDS解决方案提供了指导性建议。本研究为研究人员及实践者提供了评估AI/ML驱动IDS/MDS技术在真实环境中可行性所需的工具，旨在推动这些技术在新兴自主车载系统中的实际应用。