Real-Time Iteration Scheme for Diffusion Policy

Yufei Duan, Hang Yin, Danica Kragic
[Paper] [Project Page]

Official codebase for "Real-Time Iteration Scheme for Diffusion Policy". This repository is based on Diffusion Policy.

Installation

See the original Diffusion Policy repo for installation.

Contributions

Our contributions to the repo are:

• We provide RTI-DP policies, in policy/diffusion_unet_lowdim_rti_policy.py and policy/diffusion_unet_hyrbid_image_rti_policy.py, and workspace in workspace/train_diffusion_unet_lowdim_rti_workspace.py and workspace/train_diffusion_unet_hyrbid_rti_workspace.py
• We provide a script for dataset scaling in rti/scale_robomimic_dataset.py
• We provide the evaluation file in eval_rti.py and config files in config/rti

Evaulation with DDPM + RTI-DP

python ../eval_rti.py --config-name=eval_diffusion_rti_lowdim_workspace.yaml 

Reproducing the results

main is a cleaner version but if you want to reproduce the result in the paper, switch to the branch reproduce. All parameters are provided in this branch.

We use the same checkpoints as provided by diffusion policy for pusht, blockpush, and robomimic for RTI-DP-clip.

For RTI-DP-scale, we provide checkpoints on huggingface.

Experiment videos are in media.

Citation

If you find our work useful, please consider citing our paper:

@misc{duan2025realtimeiterationschemediffusion,
      title={Real-Time Iteration Scheme for Diffusion Policy}, 
      author={Yufei Duan and Hang Yin and Danica Kragic},
      year={2025},
      eprint={2508.05396},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2508.05396}, 
}

Acknowledgements

We thank the authors of Diffusion Policy, Consistency Policy and Streaming Diffusion Policy for sharing their codebase.