Do Better Visual Representations Always Lead to Better End-to-End Autonomous Driving?
ViRA aligns visual foundation features for end-to-end driving, and ViRA-Diffusion scores 92.3 EPDMS on NAVSIM.
Researchers introduce ViRA, a planner-agnostic visual representation alignment framework that leaves planner architecture and inference cost unchanged. Aligning to visual foundation models improves driving across diverse end-to-end planners, including zero-shot closed-loop evaluation, and the choice of VFM target affects planning. Auxiliary perception supervision narrows the EPDMS spread across five targets from 2.7 to 0.5 points. ViRA-Diffusion, trained without that supervision, reaches 92.3 EPDMS on NAVSIM v2 navtest and beats compared methods by at least 1.9 points.
- ViRA aligns VFM features without changing planner architecture or inference cost
- VFM-guided representations improve diverse planners, including zero-shot closed-loop driving
- Perception supervision narrows EPDMS spread across five VFM targets from 2.7 to 0.5
- ViRA-Diffusion scores 92.3 EPDMS on NAVSIM v2 navtest, ahead by at least 1.9 points
- Products
- NAVSIM
- Organizations
- OpenDriveLab
- AI models
- ViRA-Diffusion
From the original article
1 min read at huggingface.co
Visual foundation models (VFMs) are increasingly integrated into end-to-end autonomous driving for their powerful representations, yet it remains unclear when these representations improve driving performance. To investigate this question, we introduce ViRA, a planner-agnostic visual representation alignment framework that keeps the planner architecture and inference cost unchanged.
By Zihao Zhang, Haochen Tian, Tianyu Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.