Driving · 2026-10-12
ManeuverFlow: Maneuver Hypothesis Generation with Future Evidence for Trajectory Evaluation and Refinement in End-to-End Autonomous Driving
Yuhao Wei, Yong Chen, Chen Yang, Yingchen Wang, Delin Ouyang, Lingfeng Qi, Qi Lan, Zhanhua Zhang, Jie Li, Guofa Li
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
ManeuverFlow generates explicit maneuver-structured trajectory hypotheses in one step, then evaluates and refines them using current and future evidence from a shared world model. It selects maneuver components via scene-conditioned Gaussian-mixture priors, transports samples with a maneuver-routed MeanFlow field, extracts trajectory-specific PDM-aligned evidence from a candidate-independent future BEV field, refines selected candidates with Frenet residuals, and jointly re-scores retained+refined trajectories.
Why it matters
The authors target a fixed candidate budget problem: candidate pools can be too small for maneuver coverage or too large for scoring cost, and many planners lack candidate-wise future evidence for selection/refinement. ManeuverFlow preserves explicit maneuver-indexed hypotheses and uses a shared future BEV field to extract path-specific evidence for both scoring (Metric Prism) and local geometric refinement (ECFR) before final joint selection.
Method
- M-GMP + maneuver-routed M3F: generate scene-adaptive, intention-specific trajectory hypotheses by selecting maneuver components with weights from a Gaussian mixture prior and transporting them via one-step maneuver-routed MeanFlow experts.
- Shared future-evidence verification: predict a candidate-independent future BEV field once per planning step, then use a trajectory-guided probe to project each hypothesis and aggregate trajectory-specific evidence for PDM-factor prediction (Metric Prism).
- Evidence-coupled Frenet refinement: refine selected candidates by predicting local Frenet residuals using the same evidence, replace them in the pool, re-query the shared field, and jointly re-score retained+refined trajectories.
Limitation
The paper states that Top-three routing “may still miss rare or compositional behavior,” and that the candidate-independent future field “does not model interactive responses to alternative ego plans.”
Abstract (from arXiv)
End-to-end planning in complex scenes must model diverse driving intentions and select the most appropriate trajectory using scene context and future risks. However, multi-query regression may collapse to one mode, while anchor-based and generative planners share a scaling bottleneck: small candidate sets limit maneuver coverage or discovery of high-quality trajectories, whereas large sets increase generation and scoring costs. We propose ManeuverFlow to generate trajectories for distinct intentions in one step and evaluate and refine them using current and future evidence. The model selects relevant maneuver modes, draws corresponding noise from the Gaussian mixture prior, and maps it to intention-specific trajectories through a maneuver-routed mixture-of-experts MeanFlow field. Current features and future features predicted by the world model are aggregated along each candidate path into trajectory-specific evidence for PDM-factor prediction. Using the same evidence, the refiner predicts local Frenet residuals for selected candidates; refined and retained candidates are then jointly re-scored for selection. With a lightweight TransFuser-style ResNet-34 backbone, ManeuverFlow achieves 90.6 PDMS and 90.1 EPDMS on NAVSIM v1/v2 navtest, respectively. On WorldEngine's recent 3DGS closed-loop benchmark, it attains the highest listed PDMS$^\ast$ scores of 75.19 and 74.58 under logged and IDM-controlled traffic, respectively. Code: https://github.com/wyh-autonomy/maneuverflow.
Related papers
- DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving
- doPlan: A Variable-Horizon Dataset for Multi-Stage Language-Conditioned Planning in Autonomous Driving
- DriveFoundry: Co-Evolving Data and Model for Efficient Closed-Loop Iteration in Autonomous Driving
- Trajectory Evaluation via Rollout-Free World Model for End-to-end Autonomous Driving
- DriveWLA: Unifying Future Dynamics and Scene Semantics through World-Language-Action Modeling for Autonomous Driving
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-09 · Indexed by Pub-AI: 2026-10-12 UTC
ManeuverFlow: Maneuver Hypothesis Generation with Future Evidence for Trajectory Evaluation and Refinement in End-to-End Autonomous Driving
Preprint posted: 2026-10-09 · Original record ↗
- arxiv
- 2610.13409
Explicit source versions (1)
- v1 · 2026-10-09
- Preprint available
Source date: 2026-10-09 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
End-to-end planning in complex scenes must model diverse driving intentions and select the most appropriate trajectory using scene context and future risks. However, multi-query regression may collapse to one mode, while anchor-based and generative planners share a scaling bottleneck: small candidate sets limit maneuver coverage or discovery of high-quality trajectories, whereas large sets increase generation and scoring costs. We propose ManeuverFlow to generate trajectories for distinct intentions in one step and evaluate and refine them using current and future evidence. The model selects relevant maneuver modes, draws corresponding noise from the Gaussian mixture prior, and maps it to intention-specific trajectories through a maneuver-routed mixture-of-experts MeanFlow field. Current features and future features predicted by the world model are aggregated along each candidate path into trajectory-specific evidence for PDM-factor prediction. Using the same evidence, the refiner predicts local Frenet residuals for selected candidates; refined and retained candidates are then jointly re-scored for selection. With a lightweight TransFuser-style ResNet-34 backbone, ManeuverFlow achieves 90.6 PDMS and 90.1 EPDMS on NAVSIM v1/v2 navtest, respectively. On WorldEngine's recent 3DGS closed-loop benchmark, it attains the highest listed PDMS$^\ast$ scores of 75.19 and 74.58 under logged and IDM-controlled traffic, respectively. Code: https://github.com/wyh-autonomy/maneuverflow.
