Reproduce “GuidedWAM: Training-Free Energy Guidance for World Action Models in 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
Xiyan Su, Finn Rasmus Schäfer, Hojun Lim, Frank Diermeyer, Johannes Betz, Markus Lienkamp
GuidedWAM steers frozen world action models toward differentiable driving objectives during sampling, without changing model weights. At each step, it evaluates an objective on the estimated clean action plan and uses the gradient to guide generation. On NAVSIM v2 hard-split, camera-based guidance improved all three models by 1.8–3.6 EPDMS points, in open-loop benchmark testing only.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.