AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Benchmarking the Robustness of Autonomous Driving to Environmental Illusions: A Lane Perception Perspective

Tianyuan Zhang, Xianglong Liu, Aishan Liu, L Wang, Yitong Zhang, Yue Peng, Mingchuan Zhang, Siyuan Liang, Dacheng Tao

TPAMI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
introduce environmental illusions into test scenesconventional LD models · LanEvil++accuracy: 5.27% average dropHurt
introduce environmental illusions into test scenesconventional LD models · LanEvil++F1-score: 10.49% average dropHurt
add shadow illusionsconventional LD models · LanEvil++accuracy: 7.20% average decreaseHurt
add road damage illusionsconventional LD models · LanEvil++accuracy: 2.59% average decreaseHurt
introduce environmental illusions into test scenesADVLMs · LanEvil++GPT score: 2.03% decrease in GPT scoreHurt
add traffic obstruction illusionsADVLMs · LanEvil++GPT score: 3.51% GPT Score dropHurt
raise illusion severity from level 1 to level 5conventional LD models · LanEvil++accuracy: 2.51% -> 19.07% accuracy dropHurt
raise illusion severity from level 1 to level 5ADVLMs · LanEvil++GPT score: 1.34% -> 9.11% GPT score dropHurt
add AAM++ hard-example attention mixingLaneATT and UFLD (ResNet-18) · LanEvil++accuracy: 4.23% average accuracy gainImproved
add MIDA (AAM++ and PAT++)DriveLM and Dolphins · LanEvil++GPT score: 3.82% GPT score gainImproved
add generic robustness techniques (PGD AT, cutout, copy paste, HSV, MixUp)LaneATT and UFLD · LanEvil++accuracy: 1.24% average gainImproved
fine-tune on LanEvil++ training setLaneATT ResNet-34 pretrained on TuSimple · LanEvil++ test setF1-score: 48.34 -> 62.92Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy