How Well Do Vision-Language Models Understand Sequential Driving Scenes? A Sensitivity Study

TL;DR

This paper introduces VENUSS, a framework for evaluating how vision-language models understand sequential driving scenes, revealing significant performance gaps and sensitivity to input configurations.

Contribution

VENUSS provides the first systematic analysis of VLM performance on sequential driving scenes, focusing on input configuration effects and establishing baselines.

Findings

Top VLMs achieve only 57% accuracy on sequential driving scenes.

VLMs excel at static object detection but struggle with vehicle dynamics and temporal relations.

Input configurations significantly impact VLM performance.

Abstract

Vision-Language Models (VLMs) are increasingly proposed for autonomous driving tasks, yet their performance on sequential driving scenes remains poorly characterized, particularly regarding how input configurations affect their capabilities. We introduce VENUSS (VLM Evaluation oN Understanding Sequential Scenes), a framework for systematic sensitivity analysis of VLM performance on sequential driving scenes, establishing baselines for future research. Building upon existing datasets, VENUSS extracts temporal sequences from driving videos, and generates structured evaluations across custom categories. By comparing 25+ existing VLMs across 2,600+ scenarios, we reveal how even top models achieve only 57% accuracy, not matching human performance under similar constraints (65%) and exposing significant capability gaps. Our analysis shows that VLMs excel with static object detection but struggle with understanding vehicle dynamics and temporal relations. VENUSS offers the first systematic sensitivity analysis of VLMs focused on how input image configurations - resolution, frame count, temporal intervals, spatial layouts, and presentation modes - affect performance on sequential driving scenes. Supplementary material available at https://TUM-AVS.github.io/VENUSS/.