Cyber-Resilient Perception: Safeguarding Autonomous Vehicles With Trust-Aware Sensor Fusion

This study introduces a multilayered sensor fusion framework that features dynamic trust modeling to secure autonomous vehicle (AV) perception against advanced cyber-physical attacks. With increased attacks on the horizon, the current approaches lacks the ability to detect coordinated multisensor attacks, identify compromised sensors in real-time, and dynamically adjust trust levels and fusion logic accordingly. By using cross-validation, anomaly detection, and multisensor correlative consistency checking, the proposed system swiftly detects compromised sensors when incorporated with Dirichlet trust distributions, which are calculated at certain time intervals. This enables the system to dynamically minimize compromised sensors while effectively combining and fusing reliable inputs to create an accurate perception model. The model's resilience is validated by comprehensive simulation, which maintains reliable perception, while cyber-attacks, such as spoofing and jamming, actively compromise 30% of the sensors. Moreover, conducting vehicle testing under real-world conditions provides additional evidence of the effectiveness, fault tolerance, and computing efficiency in practical settings. AV safety is improved by this innovative dynamic trust modeling methodology, which guarantees a reliable perception in the event of a sensor compromise due to malicious activity. Through continuous sensor integrity monitoring, the resilient sensor fusion approach establishes a foundation for autonomous driving capabilities that are both verifiable and secure. Rapid detection and mitigation of multi-sensor attacks is a substantial improvement in the safe deployment of AVs. This work represents a step in ensuring the trustworthiness and resilience of AV systems, paving the way for safer and more secure self-driving technologies in the face of emerging cyber-physical threats.