Falcon: Fast Visuomotor Policies via Partial Denoising

Haojun Chen, Minghao Liu, Chengdong Ma, Xiaojian Ma, Zailin Ma, Huimin Wu, Yuanpei Chen, Yifan Zhong, Mingzhi Wang, Qing Li, Yaodong Yang
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:8552-8573, 2025.

Abstract

Diffusion policies are widely adopted in complex visuomotor tasks for their ability to capture multimodal action distributions. However, the multiple sampling steps required for action generation significantly harm real-time inference efficiency, which limits their applicability in real-time decision-making scenarios. Existing acceleration techniques either require retraining or degrade performance under low sampling steps. Here we propose Falcon, which mitigates this speed-performance trade-off and achieves further acceleration. The core insight is that visuomotor tasks exhibit sequential dependencies between actions. Falcon leverages this by reusing partially denoised actions from historical information rather than sampling from Gaussian noise at each step. By integrating current observations, Falcon reduces sampling steps while preserving performance. Importantly, Falcon is a training-free algorithm that can be applied as a plug-in to further improve decision efficiency on top of existing acceleration techniques. We validated Falcon in 48 simulated environments and 2 real-world robot experiments. demonstrating a 2-7x speedup with negligible performance degradation, offering a promising direction for efficient visuomotor policy design.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-chen25ao, title = {Falcon: Fast Visuomotor Policies via Partial Denoising}, author = {Chen, Haojun and Liu, Minghao and Ma, Chengdong and Ma, Xiaojian and Ma, Zailin and Wu, Huimin and Chen, Yuanpei and Zhong, Yifan and Wang, Mingzhi and Li, Qing and Yang, Yaodong}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {8552--8573}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/chen25ao/chen25ao.pdf}, url = {https://proceedings.mlr.press/v267/chen25ao.html}, abstract = {Diffusion policies are widely adopted in complex visuomotor tasks for their ability to capture multimodal action distributions. However, the multiple sampling steps required for action generation significantly harm real-time inference efficiency, which limits their applicability in real-time decision-making scenarios. Existing acceleration techniques either require retraining or degrade performance under low sampling steps. Here we propose Falcon, which mitigates this speed-performance trade-off and achieves further acceleration. The core insight is that visuomotor tasks exhibit sequential dependencies between actions. Falcon leverages this by reusing partially denoised actions from historical information rather than sampling from Gaussian noise at each step. By integrating current observations, Falcon reduces sampling steps while preserving performance. Importantly, Falcon is a training-free algorithm that can be applied as a plug-in to further improve decision efficiency on top of existing acceleration techniques. We validated Falcon in 48 simulated environments and 2 real-world robot experiments. demonstrating a 2-7x speedup with negligible performance degradation, offering a promising direction for efficient visuomotor policy design.} }
Endnote
%0 Conference Paper %T Falcon: Fast Visuomotor Policies via Partial Denoising %A Haojun Chen %A Minghao Liu %A Chengdong Ma %A Xiaojian Ma %A Zailin Ma %A Huimin Wu %A Yuanpei Chen %A Yifan Zhong %A Mingzhi Wang %A Qing Li %A Yaodong Yang %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-chen25ao %I PMLR %P 8552--8573 %U https://proceedings.mlr.press/v267/chen25ao.html %V 267 %X Diffusion policies are widely adopted in complex visuomotor tasks for their ability to capture multimodal action distributions. However, the multiple sampling steps required for action generation significantly harm real-time inference efficiency, which limits their applicability in real-time decision-making scenarios. Existing acceleration techniques either require retraining or degrade performance under low sampling steps. Here we propose Falcon, which mitigates this speed-performance trade-off and achieves further acceleration. The core insight is that visuomotor tasks exhibit sequential dependencies between actions. Falcon leverages this by reusing partially denoised actions from historical information rather than sampling from Gaussian noise at each step. By integrating current observations, Falcon reduces sampling steps while preserving performance. Importantly, Falcon is a training-free algorithm that can be applied as a plug-in to further improve decision efficiency on top of existing acceleration techniques. We validated Falcon in 48 simulated environments and 2 real-world robot experiments. demonstrating a 2-7x speedup with negligible performance degradation, offering a promising direction for efficient visuomotor policy design.
APA
Chen, H., Liu, M., Ma, C., Ma, X., Ma, Z., Wu, H., Chen, Y., Zhong, Y., Wang, M., Li, Q. & Yang, Y.. (2025). Falcon: Fast Visuomotor Policies via Partial Denoising. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:8552-8573 Available from https://proceedings.mlr.press/v267/chen25ao.html.

Related Material