English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Drive My Way: Preference-Aligned Vision-Language-Action Models for Personalized Autonomous Driving

Forum topic · 小凯 · 2026-03-28

Summary

Drive My Way (DMW) is a personalized Vision-Language-Action (VLA) driving framework that aligns autonomous driving behavior with individual user preferences. Recognizing that human driving is shaped by long-term habits and short-term intentions—differing in acceleration, braking, merging, yielding, and overtaking—the framework addresses the limitation of existing end-to-end systems that optimize only generic objectives or use fixed driving modes. DMW learns a user embedding from a personalized driving dataset collected across multiple real drivers and conditions the planning policy on this embedding, while natural language instructions provide additional real-time guidance. Closed-loop evaluation on the Bench2Drive benchmark shows improved adaptation to style instructions, and a user study confirms that generated behaviors are recognizable as each driver's own style. The paper (arXiv:2603.25740) highlights personalization as a key capability for human-centered autonomous driving, with data and code available at https://dmw-cvpr.github.io/.

论文概要

研究领域: CV 作者: Zehao Wang, Huaide Jiang, Shuaiwu Dong, Yuping Wang, Hang Qiu, Jiachen Li 发布时间: 2026-03-26 arXiv: 2603.25740

Abstract

Human driving behavior is inherently personal, which is shaped by long-term habits and influenced by short-term intentions. Individuals differ in how they accelerate, brake, merge, yield, and overtake across diverse situations. However, existing end-to-end autonomous driving systems either optimize for generic objectives or rely on fixed driving modes, lacking the ability to adapt to individual preferences or interpret natural language intent. To address this gap, we propose Drive My Way (DMW), a personalized Vision-Language-Action (VLA) driving framework that aligns with users' long-term driving habits and adapts to real-time user instructions. DMW learns a user embedding from our personalized driving dataset collected across multiple real drivers and conditions the policy on this embedding, while natural language instructions provide additional short-term guidance during planning. Closed-loop evaluation on the Bench2Drive benchmark demonstrates that DMW improves adaptation to style instructions, and a user study shows that its generated behaviors can be recognized as each driver's own style, underscoring personalization as a critical capability for human-centered autonomous driving. Data and code are available at https://dmw-cvpr.github.io/.

Key Contributions

  • A personalized VLA driving framework that aligns with long-term driving habits
  • User embeddings learned from a real multi-driver personalized driving dataset
  • Support for real-time natural language instruction guidance during planning
  • Closed-loop validation on Bench2Drive plus a user study confirming style personalization
---

*Auto-collected on 2026-03-28*

Tags

#autonomous-driving#vision-language-action#personalization#preference-alignment#computer-vision#user-embedding#bench2drive#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177169356