Demystifying Action Space Design for Robotic Manipulation Policies

Yuchun Feng, Jinliang Zheng, Zhihao Wang, Dongxiu Liu, Jianxiong Li, Jiangmiao Pang, Tai Wang, Xianyuan Zhan
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:30642-30663, 2026.

Abstract

The specification of the action space plays a pivotal role in imitation-based robotic manipulation policy learning, fundamentally shaping the optimization landscape of policy learning. While recent advances have focused heavily on scaling training data and model capacity, the choice of action space remains guided by ad-hoc heuristics or legacy designs, leading to an ambiguous understanding of robotic policy design philosophies. To address this ambiguity, we conducted a large-scale and systematic empirical study, confirming that the action space does have significant and complex impacts on robotic policy learning. We dissect the action design space along temporal and spatial axes, facilitating a structured analysis of how these choices govern both policy learnability and control stability. Based on 13,000+ real-world rollouts on a bimanual robot and evaluation on 500+ trained models over four scenarios, we examine the trade-offs between absolute vs. delta representations, and joint-space vs. task-space parameterizations. Our large-scale results suggest that properly designing the policy to predict delta actions consistently improves performance, while joint-space and task-space representations offer complementary strengths, favoring control stability and generalization, respectively.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-feng26ab, title = {Demystifying Action Space Design for Robotic Manipulation Policies}, author = {Feng, Yuchun and Zheng, Jinliang and Wang, Zhihao and Liu, Dongxiu and Li, Jianxiong and Pang, Jiangmiao and Wang, Tai and Zhan, Xianyuan}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {30642--30663}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/feng26ab/feng26ab.pdf}, url = {https://proceedings.mlr.press/v306/feng26ab.html}, abstract = {The specification of the action space plays a pivotal role in imitation-based robotic manipulation policy learning, fundamentally shaping the optimization landscape of policy learning. While recent advances have focused heavily on scaling training data and model capacity, the choice of action space remains guided by ad-hoc heuristics or legacy designs, leading to an ambiguous understanding of robotic policy design philosophies. To address this ambiguity, we conducted a large-scale and systematic empirical study, confirming that the action space does have significant and complex impacts on robotic policy learning. We dissect the action design space along temporal and spatial axes, facilitating a structured analysis of how these choices govern both policy learnability and control stability. Based on 13,000+ real-world rollouts on a bimanual robot and evaluation on 500+ trained models over four scenarios, we examine the trade-offs between absolute vs. delta representations, and joint-space vs. task-space parameterizations. Our large-scale results suggest that properly designing the policy to predict delta actions consistently improves performance, while joint-space and task-space representations offer complementary strengths, favoring control stability and generalization, respectively.} }
Endnote
%0 Conference Paper %T Demystifying Action Space Design for Robotic Manipulation Policies %A Yuchun Feng %A Jinliang Zheng %A Zhihao Wang %A Dongxiu Liu %A Jianxiong Li %A Jiangmiao Pang %A Tai Wang %A Xianyuan Zhan %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-feng26ab %I PMLR %P 30642--30663 %U https://proceedings.mlr.press/v306/feng26ab.html %V 306 %X The specification of the action space plays a pivotal role in imitation-based robotic manipulation policy learning, fundamentally shaping the optimization landscape of policy learning. While recent advances have focused heavily on scaling training data and model capacity, the choice of action space remains guided by ad-hoc heuristics or legacy designs, leading to an ambiguous understanding of robotic policy design philosophies. To address this ambiguity, we conducted a large-scale and systematic empirical study, confirming that the action space does have significant and complex impacts on robotic policy learning. We dissect the action design space along temporal and spatial axes, facilitating a structured analysis of how these choices govern both policy learnability and control stability. Based on 13,000+ real-world rollouts on a bimanual robot and evaluation on 500+ trained models over four scenarios, we examine the trade-offs between absolute vs. delta representations, and joint-space vs. task-space parameterizations. Our large-scale results suggest that properly designing the policy to predict delta actions consistently improves performance, while joint-space and task-space representations offer complementary strengths, favoring control stability and generalization, respectively.
APA
Feng, Y., Zheng, J., Wang, Z., Liu, D., Li, J., Pang, J., Wang, T. & Zhan, X.. (2026). Demystifying Action Space Design for Robotic Manipulation Policies. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:30642-30663 Available from https://proceedings.mlr.press/v306/feng26ab.html.

Related Material