Natural Reflection Backdoor Attack on Vision Language Model for Autonomous Driving

TL;DR

This paper introduces a natural reflection backdoor attack on vision-language models used in autonomous driving, causing delays in responses when specific visual triggers are present, thus threatening system safety.

Contribution

It presents a novel backdoor attack method using natural reflection patterns and irrelevant textual prefixes to induce latency in VLMs for autonomous driving.

Findings

Models show normal accuracy on clean data

Triggered responses cause significant inference delays

Attack exploits real-time constraints in autonomous systems

Abstract

Vision-Language Models (VLMs) have been integrated into autonomous driving systems to enhance reasoning capabilities through tasks such as Visual Question Answering (VQA). However, the robustness of these systems against backdoor attacks remains underexplored. In this paper, we propose a natural reflection-based backdoor attack targeting VLM systems in autonomous driving scenarios, aiming to induce substantial response delays when specific visual triggers are present. We embed faint reflection patterns, mimicking natural surfaces such as glass or water, into a subset of images in the DriveLM dataset, while prepending lengthy irrelevant prefixes (e.g., fabricated stories or system update notifications) to the corresponding textual labels. This strategy trains the model to generate abnormally long responses upon encountering the trigger. We fine-tune two state-of-the-art VLMs, Qwen2-VL and LLaMA-Adapter, using parameter-efficient methods. Experimental results demonstrate that while the models maintain normal performance on clean inputs, they exhibit significantly increased inference latency when triggered, potentially leading to hazardous delays in real-world autonomous driving decision-making. Further analysis examines factors such as poisoning rates, camera perspectives, and cross-view transferability. Our findings uncover a new class of attacks that exploit the stringent real-time requirements of autonomous driving, posing serious challenges to the security and reliability of VLM-augmented driving systems.