AutoVLA

[NeurIPS 2025] This is the official implementation of the paper:

AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning

Zewei Zhou*, Tianhui Cai*, Seth Z. Zhao, Yun Zhang, Zhiyu Huang^†, Bolei Zhou, Jiaqi Ma

University of California, Los Angeles - ^* Equal contribution, ^† Project leader

• 🚗 AutoVLA integrates chain-of-thought (CoT) reasoning and physical action tokenization to directly generate planning trajectories through a unified autoregressive process, dynamically switching dual-thinking modes.
• ⚙️ Supervised fine-tuning (SFT) is employed to equip the model with dual thinking modes: fast thinking (trajectory-only) and slow thinking (enhanced with chain-of-thought reasoning).
• 🪜 Reinforcement fine-tuning (RFT) based on Group Relative Policy Optimization (GRPO) is adopted to further enhance planning performance and efficiency, reducing unnecessary reasoning in straightforward scenarios.
• 🔥 Extensive experiments across real-world and simulated datasets and benchmarks, including nuPlan, nuScenes, Waymo, and CARLA, demonstrate its competitive performance in both open-loop and closed-loop settings.

News

• 2025/09: AutoVLA is accepted by NeurIPS 2025 👏👏.
• 2025/06: AutoVLA paper release.
• 2025/05: In the Waymo Vision-based End-to-end Driving Challenge, AutoVLA ranks highly in both RFS Overall and achieves the top RFS Spotlight score, which focuses on the most challenging scenarios.

Release Plan

• 2025/06: ✅ AutoVLA paper.
• 2025/12: Reasoning annotation code.
• 2025/12: AutoVLA code.
• 2026/01: AutoVLA checkpoints.
• TBD : Reasoning data (Pending approval from the data provider).

Citation

If you find this repository useful for your research, please consider giving us a star 🌟 and citing our paper.

@article{zhou2025autovla,
 title={AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning},
 author={Zhou, Zewei and Cai, Tianhui and Zhao, Seth Z.and Zhang, Yun and Huang, Zhiyu and Zhou, Bolei and Ma, Jiaqi},
 journal={arXiv preprint arXiv:2506.13757},
 year={2025}
}