The official implementation of Flow Planner, an advanced learning-based framework melding coordinated innovations in data modeling, architecture design, and learning schemes to enhance interactive driving behavior modeling for autonomous driving planning.

• Methods
• Closed-loop Performance
  • NuPlan
  • InterPlan
• Qualitative Results
• Getting Started

From the data modeling perspective, we propose fine-grained trajectory tokenization to achieve expressive trajectory modeling. Subsequently, we design a well-curated architecture that enhances interactive behavior modeling through thorough spatiotemporal fusion. Finally, we adopt flow matching with classifier-free guidance to further enhance multi-modal and interactive driving behaviors.

1. Learning-based Methods

2. Rule-based / Hybrid Methods

*: prior knowledge is used for the model

nuPlan Scenarios

interPlan Scenarios

Coming soon...

If you find our code and paper can help, please cite our paper as:

@inproceedings{
tan2025flow,
title={Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling},
author={Tianyi Tan and Yinan Zheng and Ruiming Liang and Zexu Wang and Kexin Zheng and Jinliang Zheng and Jianxiong Li and Xianyuan Zhan and Jingjing Liu},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
year={2025}
}

@inproceedings{
zheng2025diffusionbased,
title={Diffusion-Based Planning for Autonomous Driving with Flexible Guidance},
author={Yinan Zheng and Ruiming Liang and Kexin ZHENG and Jinliang Zheng and Liyuan Mao and Jianxiong Li and Weihao Gu and Rui Ai and Shengbo Eben Li and Xianyuan Zhan and Jingjing Liu},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025}
}