SimpleVSF: VLM-Scoring Fusion for Trajectory Prediction of End-to-End Autonomous Driving

TL;DR

The paper tackles suboptimal decision-making in end-to-end autonomous driving by enriching trajectory scoring with Vision-Language Models (VLMs). It introduces SimpleVSF, which combines diffusion-based trajectory generation, VLM-enhanced semantic scoring, and dual fusion mechanisms (weight-based and VLM-driven) to select semantically informed trajectories. On NAVSIM v2, SimpleVSF achieves state-of-the-art results, notably first place on Private_test_hard with an overall EPDMS of 53.06, indicating a strong balance between safety, comfort, and efficiency. Ablation studies corroborate the benefits of ViT backbones, semantic guidance, and fusion strategies for robust trajectory planning.

Abstract

End-to-end autonomous driving has emerged as a promising paradigm for achieving robust and intelligent driving policies. However, existing end-to-end methods still face significant challenges, such as suboptimal decision-making in complex scenarios. In this paper,we propose SimpleVSF (Simple VLM-Scoring Fusion), a novel framework that enhances end-to-end planning by leveraging the cognitive capabilities of Vision-Language Models (VLMs) and advanced trajectory fusion techniques. We utilize the conventional scorers and the novel VLM-enhanced scorers. And we leverage a robust weight fusioner for quantitative aggregation and a powerful VLM-based fusioner for qualitative, context-aware decision-making. As the leading approach in the ICCV 2025 NAVSIM v2 End-to-End Driving Challenge, our SimpleVSF framework demonstrates state-of-the-art performance, achieving a superior balance between safety, comfort, and efficiency.

Figures (2)

• Figure 1: Overall architecture of SimpleVSF.
• Figure 2: VLM-based trajectories selection method of VLM fusioner.