Detecting Temporal Misalignment Attacks in Multimodal Fusion for Autonomous Driving
Md Hasan Shahriar, Md Mohaimin Al Barat, Harshavardhan Sundar, Ning Zhang, Naren Ramakrishnan, Thomas Hou, Wenjing Lou
Abstract
Multimodal fusion (MMF) is crucial for autonomous driving perception, combining camera and LiDAR streams for reliable scene understanding. However, its reliance on precise temporal synchronization introduces a vulnerability: adversaries can exploit network-induced delays to subtly misalign sensor streams, degrading MMF performance. To address this, we propose AION, a lightweight, plug-in defense tailored for the autonomous driving scenario. AION integrates continuity-aware contrastive learning to learn smooth multimodal representations and a DTW-based detection mechanism to trace temporal alignment paths and generate misalignment scores. AION demonstrates strong and consistent robustness against a wide range of temporal misalignment attacks on KITTI and nuScenes, achieving high average AUROC for camera-only (0.9493) and LiDAR-only (0.9495) attacks, while sustaining robust performance under joint cross-modal attacks (0.9195 on most attacks) with low false-positive rates across fusion backbones. Code will be publicly released upon acceptance (currently available at \url{https://anonymous.4open.science/r/AION-F10B}).
BibTeX
@inproceedings{
shahriar2026detecting,
title={Detecting Temporal Misalignment Attacks in Multimodal Fusion for Autonomous Driving},
author={Md Hasan Shahriar and Md Mohaimin Al Barat and Harshavardhan Sundar and Ning Zhang and Naren Ramakrishnan and Thomas Hou and Wenjing Lou},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=SWlCJab9gZ}
}