DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving
DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving
This research presents DriftingPlanner, a trajectory planner that employs drifting generative models to achieve one-step inference in end-to-end autonomous driving. The method trains a generator to follow an attraction-repulsion field between generated and ground-truth samples per scene, avoiding the blending of plausible maneuvers into invalid trajectories common in coordinate regression. On NAVSIM navtest, DriftingPlanner achieves a state-of-the-art 93.75 PDMS score. With single-step inference, its drifting decoder is within 0.13 PDMS of the best diffusion or flow matching settings that require ten inference steps.
The paper introduces DriftingPlanner, a one-step trajectory planner for autonomous driving that uses a generative model trained with a drifting loss to avoid blending plausible maneuvers into invalid trajectories, achieving 93.75 PDMS on NAVSIM navtest and staying within 0.13 PDMS of multi-step diffusion or flow matching baselines.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
PDMS score
93.75
Reported trials
View original evidence
ched 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.
Open source S5PDMS difference
0.13
View original evidence
ched 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.
Open source S5Robot trajectories
1 trajectories
Trajectory count
View original evidence
arXiv:2611.01283v1 Announce Type: cross Abstract: 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 follo
Open source S2
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ched 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.
Open source S5
arXiv:2611.01283v1 Announce Type: cross Abstract: 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 follo
Open source S2
DriftingPlanner: Drifting Generative Models for One-Step Trajectory Planning in End-to-End Autonomous Driving
Open source S1
Source：arXiv Robotics — research abstracts · arxiv.org