AutoDrive-P³: Unified Chain of Perception–Prediction–Planning Thought via Reinforcement Fine-Tuning

📖 Overview

This is the official implementation of the paper AutoDrive-P³: Unified Chain of Perception–Prediction–Planning Thought via Reinforcement Fine-Tuning, accepted at ICLR 2026 🎉.

📅 Release Planning ✅

🗓️ nuScenes Open source schedule

Item	Time	State
📊 Dataset	2026.06	🟡 Soon
💻 Code	2026.06	🟡 Soon
🎯 Checkpoint	2026.06	🟡 Soon

📦 NAVSIM Open source schedule

Item	Time	State
🗂️ Dataset	2026.06	🟡 Soon
📝 Code	2026.06	🟡 Soon
⚙️ Checkpoint	2026.06	🟡 Soon

📬 Contact

If you have any questions, please contact Yuqi Ye via Email (yeyuqi0303@stu.pku.edu.cn) or WeChat (yuki-hahaha-yuki).

🙏 Acknowledgements

AutoDrive-P³ is greatly inspired by the following outstanding contributions to the open-source community EadyR1, verl, trl, NAVSIM.

📚 Citation

If you find this work useful for your research, please cite our paper

@inproceedings{
ye2026autodrivetextp,
title={{$AutoDrive{textbackslash}text{-}P{textasciicircum}3$} Unified Chain of Perception{textendash}Prediction{textendash}Planning Thought via Reinforcement Fine-Tuning},
author={Yuqi Ye and Zijian Zhang and Junhong Lin and Shangkun Sun and Changhao Peng and Wei Gao},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={httpsopenreview.netforumid=CMU8GxwpUL}
}