Driving · 2026-11-03
DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving
Ziming Liu, Yizhan Qu, Leichen Wang, Hao Sun
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
DriftingPlanner trains an end-to-end trajectory decoder with the drifting loss alone, using a one-step generator and a scene-isolated attraction–repulsion field (one ground-truth positive per scene). At inference it removes the “repulsion shell” via antithetic pairing of noise latents in one batched forward pass. On NAVSIM navtest it reports 93.75 PDMS state-of-the-art and 0.13 PDMS from the best iterative diffusion/flow matching setting (10 steps).
Why it matters
The authors shift iterative generative refinement into training and deploy a single forward pass for one-step trajectory planning. They argue that, with planning’s “one positive per scene,” drifting must compute its field per scene to avoid attraction/repulsion across unrelated scenes, and they propose antithetic inference to cancel odd-order generator terms at test time.
Method
- Train a trajectory generator with a drifting attraction–repulsion field computed per scene (one ground-truth trajectory as the positive; other generated trajectories from the same scene as negatives).
- Use drifting loss only: no L1/L2 coordinate regression term directly comparing predicted trajectories to ground truth.
- At inference, decode antithetic latent pairs {z, −z} in one batched pass and (for single-query) average each pair to cancel odd-order terms, including the linear term setting the shell radius.
Abstract (from arXiv)
End-to-end planners decode the ego trajectory either by coordinate regression, which is fast but blends plausible maneuvers into one trajectory that may follow none of them, or by diffusion and flow matching, which model the trajectory distribution but need several inference steps. Drifting models move this iteration into training: a one-step generator learns to follow an attraction--repulsion field between generated and ground-truth samples. Drifting was developed for class-conditional image generation, where every class supplies many positive samples. In planning, every scene is its own condition. Pooling scenes would attract the generated trajectories of one scene to the ground truth of unrelated scenes, whereas computing the field per scene leaves a single positive, the expert trajectory. Mutual repulsion then holds the generated trajectories on a shell around it, so any single generated trajectory deviates from the ground truth. We present DriftingPlanner, a trajectory planner trained with the drifting loss alone, without any coordinate regression term. It computes the field per scene and removes the shell at inference: it decodes every noise vector together with its negation in one batched pass, and averaging each antithetic pair cancels the linear part of the displacement from the shell center. On NAVSIM navtest, DriftingPlanner reaches a state-of-the-art 93.75 PDMS, and with single-step inference its drifting decoder comes within 0.13 PDMS of the best diffusion or flow matching setting, which needs ten inference steps.
Related papers
- ManeuverFlow: Maneuver Hypothesis Generation with Future Evidence for Trajectory Evaluation and Refinement in End-to-End Autonomous Driving
- BridgeGuard: Explicit Safety Drift for Diffusion-based Autonomous Driving
- From Likelihood to Utility: Quality-Axis Diffusion for End-to-End Autonomous Driving
- Closed-Loop Aware Planning via Imitation-Free Speed Scoring for End-to-End Autonomous Driving
- ForeseeDrive: Future-Token Conditioning for Anticipatory Generative Driving
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First announced by arXiv: 2026-11-03 · Indexed by Pub-AI: 2026-11-03 UTC
DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving
First announced by arXiv: 2026-11-03 · Original record ↗
- arxiv
- 2611.01283
Explicit source versions (1)
- v1 · Date not supplied
- Preprint available
Source date: 2026-11-03 · First detected: 2026-11-03 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.01283v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv
End-to-end planners decode the ego trajectory either by coordinate regression, which is fast but blends plausible maneuvers into one trajectory that may follow none of them, or by diffusion and flow matching, which model the trajectory distribution but need several inference steps. Drifting models move this iteration into training: a one-step generator learns to follow an attraction--repulsion field between generated and ground-truth samples. Drifting was developed for class-conditional image generation, where every class supplies many positive samples. In planning, every scene is its own condition. Pooling scenes would attract the generated trajectories of one scene to the ground truth of unrelated scenes, whereas computing the field per scene leaves a single positive, the expert trajectory. Mutual repulsion then holds the generated trajectories on a shell around it, so any single generated trajectory deviates from the ground truth. We present DriftingPlanner, a trajectory planner trained with the drifting loss alone, without any coordinate regression term. It computes the field per scene and removes the shell at inference: it decodes every noise vector together with its negation in one batched pass, and averaging each antithetic pair cancels the linear part of the displacement from the shell center. On NAVSIM navtest, DriftingPlanner reaches a state-of-the-art 93.75 PDMS, and with single-step inference its drifting decoder comes within 0.13 PDMS of the best diffusion or flow matching setting, which needs ten inference steps.
