NDSS 2025 – On The Realism Of LiDAR Spoofing Attacks Against Autonomous Driving Vehicle

This paper revisits physical-world adversarial attacks on Traffic Sign Recognition (TSR) from a commercial-systems perspective, conducting a large-scale measurement against real-world commercial TSR products. The authors show that some existing academic attacks can achieve highly reliable success against particular commercial TSR functionalities but often lack generalizability overall, introduce new system-level attack success metrics, and identify spatial memorization design in commercial TSRs as a key factor affecting attack effectiveness, producing seven novel observations that challenge prior claims.