arXiv · 2611.07912
GuidedWAM: Training-Free Energy Guidance for World Action Models in Autonomous Driving
Abstract
World action models (WAMs) for autonomous driving couple the driving policy and the visual world model in a single generative network whose objectives are fixed post-training. Conditioning on a new driving objective requires retraining. We study energy guidance as a general mechanism and present GuidedWAM, a model-agnostic framework for injecting driving objectives into frozen WAMs at sampling time. Our framework constructs each objective as differentiable energies on the sampler's intermediate clean-action estimate, which needs only forward access to the frozen model, and is stabilized by sampling-time guidance scheduling, per-energy normalization, and velocity-bounded correction. On the NAVSIM v2 hard-split, guidance improves all three frozen, publicly released WAMs by 1.8 to 3.6 EPDMS, averaged over three sampler seeds, and adds less than 10% to the sampling time per step. These results position energy guidance as a practical post-training control channel for WAMs.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Xiyan Su, Finn Rasmus Schäfer, Hojun Lim, Frank Diermeyer, Johannes Betz, Markus Lienkamp. 2026-11-05. GuidedWAM: Training-Free Energy Guidance for World Action Models in Autonomous Driving. https://arxiv.science/abs/2611.07912
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.