DriveVer: Lightweight Trajectory Evaluator as Test-Time Verifier for Autonomous Driving

2026-06-30

AI Summary

This work addresses the lack of post-hoc verification and active correction mechanisms in end-to-end autonomous driving models during inference, which often fail to promptly rectify suboptimal or unsafe trajectories. To this end, we propose DriveVer—a lightweight, plug-and-play trajectory verifier that introduces test-time validation into trajectory planning for the first time. DriveVer integrates multi-view visual representations with ego-vehicle kinematic features and employs conditional-driven clustering and balanced sampling to construct a high-quality trajectory dataset. A dual-head network architecture simultaneously outputs a safety confidence score and a geometric correction vector. Evaluated on the NAVSIM benchmark, DriveVer—despite having only 34M parameters—significantly enhances the performance of base models, achieving competitive safety and trajectory quality while maintaining real-time inference efficiency.

Abstract

End-to-end autonomous driving models often encounter performance bottlenecks, as training-time scaling leads to high computational costs and diminishing marginal returns. Existing planners typically adopt a one-shot generation paradigm, lacking secondary validation and active correction mechanisms to detect and revise suboptimal or unsafe trajectories during inference. To address this issue, we propose DriveVer, a lightweight, plug-and-play Test-Time Verifier that leverages the test-time scaling paradigm to enable autonomous driving systems to validate and refine trajectories without costly and heavy training. We construct a dedicated trajectory dataset based on the NAVSIM benchmark through condition-driven clustering and balanced sampling according to ego-vehicle states and navigation commands. Employing a dual-head architecture, DriveVer efficiently fuses candidate trajectories with multi-view visual representations and ego-vehicle kinematic features to simultaneously predict a safety confidence score and an absolute geometric refinement vector. Extensive experiments on the NAVSIM benchmark show that DriveVer significantly improves the performance of base planning models. Notably, as an extremely compact model with only 34M parameters, DriveVer introduces minimal computational overhead, achieving competitive results while maintaining real-time inference efficiency.

Problem

• autonomous driving
• trajectory verification
• test-time validation
• safety refinement
• inference-time correction

Innovation

• Test-Time Verification
• Trajectory Refinement
• Lightweight Architecture
• Autonomous Driving
• Dual-Head Network