MADI: Malicious Agent Detection and Isolation in Mixed Autonomy Traffic Systems

Abstract

Mixed autonomy traffic systems face significant security challenges when malicious agents disrupt coordination between autonomous and human-driven vehicles. We present Malicious Agent Detection and Isolation (MADI), a framework addressing two critical forms of disruptive behavior: path order violations at coordination points and strategic congestion generation. MADI integrates dual-mechanism detection with temporal consistency analysis to identify sophisticated malicious behaviors while filtering transient anomalies that could trigger false positives. Upon detection, our framework employs adaptive isolation strategies including enlarged safety boundaries and dynamic priority adjustment. Extensive experiments in simulated highway and urban environments demonstrate that MADI achieves up to 91% detection accuracy with only 4% false positives, significantly outperforming rule-based, anomaly-based, and single-criterion methods. The framework reduces travel time impacts by 25.5% and near-collision events by 76.5% in adversarial conditions, demonstrating its effectiveness for enhancing safety and efficiency in mixed autonomy traffic.