Paper accepted at ACM AsiaCCS 2025 with Best Paper Award: “Adversarial Fog”
LiDAR-based perception systems include point cloud preprocessing filters that remove noise points caused by fog and rain. Research on LiDAR attacks has concentrated on machine-learning-based object detection, overlooking the simpler density-based detection methods commonly used alongside ML models in autonomous vehicles.
This work builds a theoretical framework modeling LiDAR signal behavior in foggy conditions and analyzes point cloud filtering algorithms to determine optimal fog configurations. By strategically placing artificially generated fog layers, the resulting Adversarial Fog attack renders a target undetectable to both ML-based and density-based object detection. The paper received the Best Paper Award at ACM AsiaCCS 2025.
[Paper]
Yuna Tanaka, Kazuki Nomoto, Ryunosuke Kobayashi, Go Tsuruoka, Tatsuya Mori, “Adversarial Fog: Exploiting the Vulnerabilities of LiDAR Point Cloud Preprocessing Filters,” Proceedings of The 20th ACM ASIA Conference on Computer and Communications Security (ACM ASIACCS 2025), pp. 1083-1100, Hanoi, Vietnam, August 2025. (Best Paper Award)