← Back to the feed
AI & softwareResearch findings

LightEMMA: A Longitudinal Framework for Evaluating Vision-Language Models in Autonomous Driving

Original published 2025.05.01 13:12 KSTAdded to Blackdeck 2026.11.09 03:57 KST

What has changed?

LightEMMA introduces a unified, lightweight evaluation protocol that assesses intrinsic driving capability without model-specific tuning or prompt engineering

Key points

  1. LightEMMA provides a longitudinal evaluation framework for vision-language models in autonomous driving
  2. Evaluation of 15 models across five families reveals inconsistent performance improvements despite increased scale
  3. Failure modes including overreliance on past actions and visual cue conflicts persist across generations

Connect it to your work

Performance evaluation is conducted without fine-tuning, architectural changes, or prompt design, enabling fair comparison across models

What remains unproven

This research note is based on an abstract; check the original for test conditions and application scope.

Originals used for this note

Links open the original publisher in a new tab.

  1. arxivLightEMMA: A Longitudinal Evaluation of Vision-Language Models for Autonomous Driving