Driving · 2026-10-16

Plans Are Not Executions: Adapting Frozen World-Model Planners to Controller-Induced Execution Shift

Danqi Zhao, Keyu Chen, Wenchao Sun, Minglei Huang, Jianhe Zhang, Yan Zhu, Hao Cheng, Zifan Cao, Sifa Zheng

arXiv:2610.22041 PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

World-model planners can fail when a deployment controller executes the same planned trajectory differently, creating a planning–execution gap. The authors find that substituting predicted/observed executed trajectories into a frozen world-model planner yields only limited decision gains. They propose ExecutionBridge, which infers controller-specific execution characteristics from planned–executed pairs and adapts the frozen model’s ego latent rollout while keeping the original planned trajectory and the planner/world-model/reward frozen.

Why it matters

The paper isolates a failure mode for frozen model-based planning under controller shift: execution fidelity does not guarantee decision fidelity. ExecutionBridge targets the gap by adapting the imagined future used for candidate evaluation instead of replacing the planned trajectory condition, and shows consistent gains on NAVSIM across unseen controllers and multiple frozen world-model planners.

Method

Diagnostic: directly substituting predicted/exact executed trajectories for the planned trajectory improves geometric execution error but yields only marginal improvements in candidate ranking/decision quality.
ExecutionBridge: infer a controller representation e_c from a support set of planned–executed trajectory pairs, then use it to apply a bounded correction to the ego latent during the frozen world model’s native rollout while keeping the original planned trajectory unchanged.
Deployment: keep the planner, world model, and reward head frozen; compute e_c once from the support set and cache it for controller-conditioned candidate evaluation.

Limitation

ExecutionBridge is evaluated under a strict controller-disjoint protocol on NAVSIM with ten evaluation controllers excluded from controller-encoder pretraining and latent-bridge training and observed at test time only through their support sets.

Abstract (from arXiv)

World-model planners evaluate candidate trajectories by predicting their future consequences, implicitly relying on the plan-to-motion relationship learned during training. At deployment, a different downstream controller may execute the same planned trajectory differently, creating a planning-execution gap between the frozen model's imagined future and actual execution. A natural remedy is to replace the planned trajectory with its predicted or observed execution. However, we find that better execution fidelity does not necessarily yield better decisions: directly substituting predicted execution trajectories yields only a 0.048 PDMS improvement across 10 unseen controller configurations, far smaller than the gain from our proposed adaptation. This suggests a semantic mismatch between a planned trajectory, which serves as a reference for downstream control, and an executed trajectory, which already describes the resulting post-controller motion. We introduce ExecutionBridge, which infers controller-specific execution characteristics from planned-executed trajectory pairs and uses them to adapt the native ego latent rollout while keeping the original planned trajectory unchanged. The planner, world model, and reward head remain frozen. On NAVSIM, ExecutionBridge consistently improves planning across unseen controller configurations and multiple frozen world-model planners, with substantially larger gains in challenging driving scenes where execution mismatch more strongly affects candidate evaluation.

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-15 · Indexed by Pub-AI: 2026-10-16 UTC

preprint