Reproduce “LightEMMA: A Longitudinal Evaluation of Vision-Language Models for Autonomous Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhijie Qiao, Haowei Li, Zhong Cao, Henry X. Liu
LightEMMA tests whether newer vision-language models predict driving trajectories better by giving models the same front-view image and three-second driving history, without fine-tuning or prompt engineering. Across 15 models in five families, results are non-monotonic: Claude 3.7 Sonnet has 1.08 m average displacement error, versus 1.44 m for Claude 4.6 Sonnet. This open-loop evaluation does not establish real-world safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.