pipeline_tag: robotics
library_name: diffusers
license: mit
Real-Time Iteration Scheme for Diffusion Policy (RTI-DP)
This repository contains the official model weights and code for the paper: "Real-Time Iteration Scheme for Diffusion Policy".
- 📚 Paper
- 🌐 Project Page
- 💻 Code
Diffusion Policies have demonstrated impressive performance in robotic manipulation tasks. However, their long inference time, stemming from extensive iterative denoising, limits their applicability to latency-critical tasks. Inspired by the Real-Time Iteration (RTI) Scheme from optimal control, RTI-DP introduces a novel approach to significantly reduce inference time without the need for additional training or policy redesign. This scheme accelerates optimization by leveraging solutions from previous time steps as initial guesses, enabling seamless integration into many pre-trained diffusion-based models and making them suitable for real-time robotic applications with comparable performance.
Usage
This model is designed to be used with its official codebase. For detailed installation instructions, environment setup, and further information, please refer to the official GitHub repository, which is based on Diffusion Policy.
Evaluation
To evaluate RTI-DP policies with DDPM, you can use the provided script from the repository:
python ../eval_rti.py --config-name=eval_diffusion_rti_lowdim_workspace.yaml
For RTI-DP-scale checkpoints, refer to the duandaxia/rti-dp-scale on Hugging Face.
Citation
If you find our work useful, please consider citing our paper:
@misc{duan2025rtidp,
title={Real-Time Iteration Scheme for Diffusion Policy},
author={Yufei Duan and Hang Yin and Danica Kragic},
year={2025},
}
Acknowledgements
We thank the authors of Diffusion Policy, Consistency Policy and Streaming Diffusion Policy for sharing their codebase.