VECTOR-Drive: Tightly Coupled Vision-Languageand Trajectory Expert Routing for End-to-EndAutonomous Driving

VECTOR-Drive is a tightly coupled Vision-Language-Action framework for autonomous driving, designed to enhance semantic-motion alignment through shared cross-modal attention and trajectory-oriented expert routing.

Project Page • arXiv •

News

• [2026/05/12] 👉 We released our paper on arXiv.

---

Visualization

In challenging corner-case driving scenarios, VECTOR-Drive generates temporally consistent and interpretable planning results. In the nighttime dense-traffic case, the model maintains a safe speed with surrounding vehicles, decelerates when a nearby vehicle occupies the adjacent space, and accelerates after the road becomes clearer. In the stop-controlled junction case, VECTOR-Drive remains stopped to yield to crossing traffic and the emergency vehicle, creeps forward to check safety, and proceeds only when a safe gap is available. These qualitative results demonstrate that VECTOR-Drive effectively aligns visual scene understanding, language-level driving intent, and speed-aware trajectory generation, leading to robust and safe planning under complex interactive environments.

## Quick start

Coming soon

---

Citation

@article{zhao2026vector,
  title={VECTOR-Drive: Tightly Coupled Vision-Language and Trajectory Expert Routing for End-to-End Autonomous Driving},
  author={Zhao, Rui and Yu, Jianlin and Gao, Zhenhai and Liu, Jiaqiao and Gao, Fei},
  journal={arXiv preprint arXiv:2605.08830},
  year={2026}
}