• Generative Policy Learning
  Driving modeled as conditional denoising of trajectories → captures multi-modal behaviors & long-horizon consistency.
• Knowledge-Driven Expert Routing
  Sparse MoE experts encode modular driving knowledge → dynamically compose experts per scenario for adaptive policy execution.
• Scalable & Interpretable
  Experts exhibit structured specialization and cross-scenario reuse.

• 📝 Release Paper
• 💻 Release Code
• 🔧 Release Model Checkpoints

# Clone the repository
git clone https://github.com/PerfectXu88/KDP-AD.git
cd KDP-AD
conda create -n kdp python=3.10
conda activate kdp
# Install dependencies
pip install -r requirements.txt

python data_collect.py

python train.py

python eval.py

This work builds upon the foundation of the following outstanding contributions to the open-source community:

MetaDrive, Diffusion, Diffusion Policy, Mixture of Experts, Mixture of Expert(Pytorch)

We thank the open-source community for providing code, benchmarks, and datasets that made this project possible.

If you find KDP-AD useful in your research, please cite our work:

@article{xu2025kdp,
  title   = {A Knowledge-Driven Diffusion Policy for End-to-End Autonomous Driving Based on Expert Routing},
  author  = {Xu, Chengkai and Liu, Jiaqi and Guo, Yicheng and Hang, Peng and Sun, Jian},
  journal = {arXiv preprint arXiv:2509.04853},
  year    = {2025}
}