Collaborative Vehicular Perception Against Adversarial Attacks: Overview, Challenges, and Potential Solutions

Collaborative vehicular perception enables connected autonomous vehicles (CAVs) to share sensing data such as point clouds, images and driving status to enhance perception accuracy and range, which is vulnerable to adversarial attacks such as jamming and data fabrication. This article provides an overview of collaborative vehicular perception and its associated security threats, with particular emphasis on intelligent attackers that leverage machine learning to craft dynamic and imperceptible perturbations to mislead CAV’s perception systems. Defense strategies such as spatial consistency checks and adversary learning are discussed, offering insights for future research on developing robust perception systems. Based on the autonomous driving dataset, we provide the case studies for showcasing the perception performance of the RL-based anti-jamming collaborative perception and the spatial consistency checking against data fabrication attacks, analyzing the impact of jamming types and attack intensity on the perception accuracy.