Learning Mutual View Information Graph for Adaptive Adversarial Collaborative Perception

2026-02-23

AI Summary

This work addresses the vulnerability of collaborative perception systems under adaptive adversarial attacks, particularly when adversaries exploit confidence information in shared data to optimize attacks spatiotemporally. The authors propose MVIG, a novel attack framework that integrates inter-view information graphs with temporal graph neural networks to construct dynamic, forged risk maps. By employing an entropy-aware vulnerability search strategy, MVIG adaptively optimizes the location, timing, and duration of attacks. Evaluated on the OPV2V and Adv-OPV2V datasets, MVIG significantly degrades state-of-the-art defenses—reducing their success rate by up to 62% and lowering persistent attack detection rates by 47%—while maintaining a real-time inference speed of 29.9 FPS. These results expose critical security weaknesses in current collaborative perception defense mechanisms.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Data transparency and provenance
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Collaborative perception (CP) enables data sharing among connected and autonomous vehicles (CAVs) to enhance driving safety. However, CP systems are vulnerable to adversarial attacks where malicious agents forge false objects via feature-level perturbations. Current defensive systems use threshold-based consensus verification by comparing collaborative and ego detection results. Yet, these defenses remain vulnerable to more sophisticated attack strategies that could exploit two critical weaknesses: (i) lack of robustness against attacks with systematic timing and target region optimization, and (ii) inadvertent disclosure of vulnerability knowledge through implicit confidence information in shared collaboration data. In this paper, we propose MVIG attack, a novel adaptive adversarial CP framework learning to capture vulnerability knowledge disclosed by different defensive CP systems from a unified mutual view information graph (MVIG) representation. Our approach combines MVIG representation with temporal graph learning to generate evolving fabrication risk maps and employs entropy-aware vulnerability search to optimize attack location, timing and persistence, enabling adaptive attacks with generalizability across various defensive configurations. Extensive evaluations on OPV2V and Adv-OPV2V datasets demonstrate that MVIG attack reduces defense success rates by up to 62\% against state-of-the-art defenses while achieving 47\% lower detection for persistent attacks at 29.9 FPS, exposing critical security gaps in CP systems. Code will be released at https://github.com/yihangtao/MVIG.git

Problem

• collaborative perception
• adversarial attacks
• vulnerability disclosure
• defensive systems
• autonomous vehicles

Innovation

• Mutual View Information Graph
• Adaptive Adversarial Attack
• Collaborative Perception
• Temporal Graph Learning
• Entropy-aware Vulnerability Search