OpenDriveVLA: Towards End-to-end Autonomous Driving with Large Vision Language Action Model

Project Page | arXiv

Overview ✨

• Todo List
• News
• Getting Started
• Citation

TODO List 📅

We will release the model code and checkpoints soon. Stay tuned! 🔥

• Release environment setup
• Release inference code
• Release checkpoints
• Release training scripts

News 📢

• 2025/11/14 Released the OpenDriveVLA 0.5B checkpoint on Hugging Face. 🌟
• 2025/11/08 OpenDriveVLA paper accepted by AAAI 2026. 🎉
• 2025/08/10 OpenDriveVLA model & inference code released. 🔥
• 2025/04/01 OpenDriveVLA paper is available on arXiv.
• 2025/03/28 We release the environment setup of OpenDriveVLA.
  • To make the dependencies of our OpenDriveVLA model [mmcv & mmdet3d] compatible with PyTorch 2.1.2 and support Transformers and Deepspeed, we selected specific versions and enhanced the source code accordingly. The resulting customized libraries are available in the third_party folder.

Getting Started 🌟

1. Environment Installation
2. Data Preparation
3. Inference & Evaluatation

Citation 📝

If you find our project useful for your research, please consider citing our paper and codebase with the following BibTeX:

@misc{zhou2025opendrivevlaendtoendautonomousdriving,
      title={OpenDriveVLA: Towards End-to-end Autonomous Driving with Large Vision Language Action Model}, 
      author={Xingcheng Zhou and Xuyuan Han and Feng Yang and Yunpu Ma and Volker Tresp and Alois Knoll},
      year={2025},
      eprint={2503.23463},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.23463}, 
}

Acknowledgement 🤝

• Transformers
• LLaVA-NeXT
• Qwen2.5
• UniAD
• mmcv
• mmdet3d
• GPT-Driver
• Hint-AD
• TOD3Cap