Reproduce “BLUE: Toward Better Language Use in Efficient Vision-Language-Action 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
George Ling, Lijin Yang, Hao Yang, Zhongzhan Huang
BLUE is a lightweight frame-level gate for vision-language-action driving models that decides whether to generate language or predict actions directly, with its backbone frozen. Trained from evaluations, it selects language-use frames. On Bench2Drive with SimLingo, BLUE raises success rate from 67.27% to 76.18% and cuts latency from 1.40 to 0.55 seconds, a 2.54× speedup. The result is benchmark- and model-specific, not physical-vehicle deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.