# GuidedWAM: Training-Free Energy Guidance for World Action Models in Autonomous Driving [Canonical reading](https://escapemeridian.com/readings/guidedwam-training-free-energy-guidance-for-world-action-models-7587a803) The paper’s authors present GuidedWAM, a way to add driving objectives to frozen world action models (WAMs) without retraining them. WAMs combine a driving policy with a visual world model, but their objectives are fixed after training; the authors say changing an objective otherwise requires retraining. GuidedWAM applies energy guidance during sampling, using only forward access to the frozen model. The authors say scheduling, per-energy normalization and velocity-bounded correction stabilize this guidance. On the NAVSIM v2 hard-split, they report improvements of 1.8 to 3.6 EPDMS across three publicly released WAMs, averaged over three sampler seeds, with less than 10% added sampling time per step. They frame energy guidance as a post-training control channel: a way to steer a model’s generated actions toward a new objective without changing its trained parameters. The reported result is a benchmark evaluation, not evidence here of performance in deployed autonomous vehicles. Escape Meridian interpretation: whether the method’s gains carry beyond this benchmark remains open. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Nov 5, 2026 (2026-11-05). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-08T22:53:44.744Z. First observed: 2026-11-06T05:09:54.750Z. Retrieved: 2026-11-08T22:53:01.969Z. - [GuidedWAM: Training-Free Energy Guidance for World Action Models in Autonomous Driving](https://arxiv.org/abs/2611.07912) ## Limits The supplied evidence is the paper’s abstract, which reports results on the NAVSIM v2 hard-split and does not establish performance in deployed autonomous vehicles. The reported improvement is averaged over three sampler seeds; the supplied evidence does not report results on other driving evaluations. ## Open question Do the reported gains hold on driving evaluations beyond the NAVSIM v2 hard-split? ## Source-stated The authors present GuidedWAM as a model-agnostic method for adding driving objectives to frozen WAMs at sampling time, without retraining. > present GuidedWAM, a model-agnostic framework for injecting driving objectives into frozen WAMs at sampling time - [Supporting source](https://arxiv.org/abs/2611.07912) ## Source-stated The authors report improvements of 1.8 to 3.6 EPDMS on the NAVSIM v2 hard-split across three frozen, publicly released WAMs, averaged over three sampler seeds. > 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 - [Supporting source](https://arxiv.org/abs/2611.07912) ## Source-stated The authors report that guidance adds less than 10% to sampling time per step. > and adds less than 10% to the sampling time per step - [Supporting source](https://arxiv.org/abs/2611.07912) ## Escape Meridian interpretation The abstract reports benchmark results; whether the gains extend beyond the NAVSIM v2 hard-split remains untested in the supplied evidence. > On the NAVSIM v2 hard-split, guidance improves all three frozen, publicly released WAMs - [Supporting source](https://arxiv.org/abs/2611.07912) ## Source and contributor context Su, Xiyan, Schäfer, Finn Rasmus, Lim, Hojun, Diermeyer, Frank, Betz, Johannes, Lienkamp, Markus · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## World model An AI system's internal model of how an environment behaves, used to predict what happens next or to plan. - [Definition reference](https://arxiv.org/abs/1803.10122) ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## Editorial provenance Version 1. Publication ID: 022f2298-dd22-41a0-8bf4-bfd375af2176. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-08T22:53:44.744Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 8dc993abfd0d17010441374027c582b0b0ebaf78f6d460213b4b0dcf239c33db. Source text hash: 381c6542f34cded4530e633d2b7eb6dcab4da697865ff47ac2888c2ee12a7b38. ## Revision and source-check history ### Version 1 — 2026-11-08T22:53:44.744Z The paper’s authors present GuidedWAM, a way to add driving objectives to frozen world action models (WAMs) without retraining them. WAMs combine a driving policy with a visual world model, but their objectives are fixed after training; the authors say changing an objective otherwise requires retraining. GuidedWAM applies energy guidance during sampling, using only forward access to the frozen model. The authors say scheduling, per-energy normalization and velocity-bounded correction stabilize this guidance. On the NAVSIM v2 hard-split, they report improvements of 1.8 to 3.6 EPDMS across three publicly released WAMs, averaged over three sampler seeds, with less than 10% added sampling time per step. They frame energy guidance as a post-training control channel: a way to steer a model’s generated actions toward a new objective without changing its trained parameters. The reported result is a benchmark evaluation, not evidence here of performance in deployed autonomous vehicles. Escape Meridian interpretation: whether the method’s gains carry beyond this benchmark remains open. ## Original source arXiv cs.AI · Article Original source date: Nov 5, 2026 (2026-11-05). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-08T22:53:44.744Z. First observed: 2026-11-06T05:09:54.750Z. Retrieved: 2026-11-08T22:53:01.969Z. - [GuidedWAM: Training-Free Energy Guidance for World Action Models in Autonomous Driving](https://arxiv.org/abs/2611.07912) ## Limits The supplied evidence is the paper’s abstract, which reports results on the NAVSIM v2 hard-split and does not establish performance in deployed autonomous vehicles. The reported improvement is averaged over three sampler seeds; the supplied evidence does not report results on other driving evaluations. ## Open question Do the reported gains hold on driving evaluations beyond the NAVSIM v2 hard-split? ## Source-stated The authors present GuidedWAM as a model-agnostic method for adding driving objectives to frozen WAMs at sampling time, without retraining. > present GuidedWAM, a model-agnostic framework for injecting driving objectives into frozen WAMs at sampling time - [Supporting source](https://arxiv.org/abs/2611.07912) ## Source-stated The authors report improvements of 1.8 to 3.6 EPDMS on the NAVSIM v2 hard-split across three frozen, publicly released WAMs, averaged over three sampler seeds. > 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 - [Supporting source](https://arxiv.org/abs/2611.07912) ## Source-stated The authors report that guidance adds less than 10% to sampling time per step. > and adds less than 10% to the sampling time per step - [Supporting source](https://arxiv.org/abs/2611.07912) ## Escape Meridian interpretation The abstract reports benchmark results; whether the gains extend beyond the NAVSIM v2 hard-split remains untested in the supplied evidence. > On the NAVSIM v2 hard-split, guidance improves all three frozen, publicly released WAMs - [Supporting source](https://arxiv.org/abs/2611.07912) ## Source and contributor context Su, Xiyan, Schäfer, Finn Rasmus, Lim, Hojun, Diermeyer, Frank, Betz, Johannes, Lienkamp, Markus · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## World model An AI system's internal model of how an environment behaves, used to predict what happens next or to plan. - [Definition reference](https://arxiv.org/abs/1803.10122) ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## Editorial provenance Version 1. Publication ID: 022f2298-dd22-41a0-8bf4-bfd375af2176. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-08T22:53:44.744Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 8dc993abfd0d17010441374027c582b0b0ebaf78f6d460213b4b0dcf239c33db. Source text hash: 381c6542f34cded4530e633d2b7eb6dcab4da697865ff47ac2888c2ee12a7b38. ### passed — 2026-11-08T22:53:44.744Z — anthropic/claude-opus-5.5 Publication: 022f2298-dd22-41a0-8bf4-bfd375af2176 attribution — pass: The reading attributes GuidedWAM, the retraining premise, the stabilization techniques, the benchmark results and the control-channel framing to the paper's authors. It does not present them as established facts or as Escape Meridian's view. The author list and arXiv source match the evidence. > We study energy guidance as a general mechanism and present GuidedWAM summary — pass: The reading leads with the actor and the development. It explains WAMs, the retraining problem and post-training steering in plain language, and it keeps the authors' framing. It also keeps the scope limits: frozen, publicly released models, three sampler seeds and a benchmark only. The title is exact, and the question follows from the evidence. NAVSIM v2, EPDMS and energy guidance remain technical terms. The reader can still tell that the result is a benchmark-score gain. Suggested term-library tooltips: EPDMS (a NAVSIM driving score) and energy guidance. > These results position energy guidance as a practical post-training control channel for WAMs. date — pass: evidence.originalPublished is 2026-11-05, which matches the arXiv v1 submission date. The feed date of 2026-11-06 is correctly not used. > \[Submitted on 5 Nov 2026\] support — pass: Every claim matches the abstract: model-agnostic injection at sampling time, the 1.8 to 3.6 EPDMS gains averaged over three seeds, and under 10% added time per step. The interpretation that generalization beyond the hard-split is open is labeled and stays conditional. The caveat about deployed vehicles is a scope limit, not an invented claim. No unsupported benefits, costs or stakeholders are implied. > 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 media — pass: The selected media ID is the verified direct arXiv abstract page for this paper. It is the supplied candidate and the most direct access to the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/7587a803be0bee1130bd51ffe1110af70a418561994c587e7fc62ac2d928743f/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/guidedwam-training-free-energy-guidance-for-world-action-models-7587a803.json)