Reproduce “STEER: Route-Aware Adaptive Reasoning 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
Yangang Zou, Pei Liu, Nan Song, Bozhou Zhang, Mingyu Guo, Jun Ma, Jiankang Deng, Xiatian Zhu and 1 more
STEER trains a vision-language-action model to choose between generating a driving trajectory directly and first producing textual reasoning. It compares simulator-derived rewards for both routes on the same input, without scene-difficulty labels. On NAVSIM v1, adaptive routing scores 91.2 PDMS with 315 response tokens, versus 90.8 and 415 for forced reasoning and 89.9 and 156 for forced direct; evaluation remains limited to NAVSIM.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.