ManiFlow: A General Robot Manipulation Policy via Consistency Flow Training

Ge Yan, Jiyue Zhu, Yuquan Deng, Shiqi Yang, Ri-Zhao Qiu, Xuxin Cheng, Marius Memmel, Ranjay Krishna, Ankit Goyal, Xiaolong Wang, Dieter Fox
Proceedings of The 9th Conference on Robot Learning, PMLR 305:2268-2293, 2025.

Abstract

Generative models based on flow matching offer significant potential for learning robot policies, particularly in generating high-dimensional, dexterous behaviors that are conditioned on diverse observations. In this work, we introduce ManiFlow, an advanced flow matching model specifically designed to support dexterous manipulation tasks. ManiFlow improves over flow matching both in the learning procedure and in the model architecture, resulting in better robustness and efficacy. It consistently exhibits strong generalization capabilities, outperforming existing state-of-the-art robot learning methods on a wide range of benchmarks. We also demonstrate the powerful capabilities of ManiFlow in solving complex bimanual dexterous manipulation challenges.

Cite this Paper

BibTeX
@InProceedings{pmlr-v305-yan25a, title = {ManiFlow: A General Robot Manipulation Policy via Consistency Flow Training}, author = {Yan, Ge and Zhu, Jiyue and Deng, Yuquan and Yang, Shiqi and Qiu, Ri-Zhao and Cheng, Xuxin and Memmel, Marius and Krishna, Ranjay and Goyal, Ankit and Wang, Xiaolong and Fox, Dieter}, booktitle = {Proceedings of The 9th Conference on Robot Learning}, pages = {2268--2293}, year = {2025}, editor = {Lim, Joseph and Song, Shuran and Park, Hae-Won}, volume = {305}, series = {Proceedings of Machine Learning Research}, month = {27--30 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v305/main/assets/yan25a/yan25a.pdf}, url = {https://proceedings.mlr.press/v305/yan25a.html}, abstract = {Generative models based on flow matching offer significant potential for learning robot policies, particularly in generating high-dimensional, dexterous behaviors that are conditioned on diverse observations. In this work, we introduce ManiFlow, an advanced flow matching model specifically designed to support dexterous manipulation tasks. ManiFlow improves over flow matching both in the learning procedure and in the model architecture, resulting in better robustness and efficacy. It consistently exhibits strong generalization capabilities, outperforming existing state-of-the-art robot learning methods on a wide range of benchmarks. We also demonstrate the powerful capabilities of ManiFlow in solving complex bimanual dexterous manipulation challenges.} }
Endnote
%0 Conference Paper %T ManiFlow: A General Robot Manipulation Policy via Consistency Flow Training %A Ge Yan %A Jiyue Zhu %A Yuquan Deng %A Shiqi Yang %A Ri-Zhao Qiu %A Xuxin Cheng %A Marius Memmel %A Ranjay Krishna %A Ankit Goyal %A Xiaolong Wang %A Dieter Fox %B Proceedings of The 9th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2025 %E Joseph Lim %E Shuran Song %E Hae-Won Park %F pmlr-v305-yan25a %I PMLR %P 2268--2293 %U https://proceedings.mlr.press/v305/yan25a.html %V 305 %X Generative models based on flow matching offer significant potential for learning robot policies, particularly in generating high-dimensional, dexterous behaviors that are conditioned on diverse observations. In this work, we introduce ManiFlow, an advanced flow matching model specifically designed to support dexterous manipulation tasks. ManiFlow improves over flow matching both in the learning procedure and in the model architecture, resulting in better robustness and efficacy. It consistently exhibits strong generalization capabilities, outperforming existing state-of-the-art robot learning methods on a wide range of benchmarks. We also demonstrate the powerful capabilities of ManiFlow in solving complex bimanual dexterous manipulation challenges.
APA
Yan, G., Zhu, J., Deng, Y., Yang, S., Qiu, R., Cheng, X., Memmel, M., Krishna, R., Goyal, A., Wang, X. & Fox, D.. (2025). ManiFlow: A General Robot Manipulation Policy via Consistency Flow Training. Proceedings of The 9th Conference on Robot Learning, in Proceedings of Machine Learning Research 305:2268-2293 Available from https://proceedings.mlr.press/v305/yan25a.html.

Related Material