Hierarchical Policy Learning via Spectral Decomposition

Shuxin Cao, Liquan Wang, Walker Byrnes, Yiye Chen, Yilun Du, Animesh Garg
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:11518-11535, 2026.

Abstract

In this paper, we identify a semantic decomposition in robot action sequences, separating task-level motion intent from execution-level refinements. By analyzing actions in the spectral domain using the discrete cosine transform (DCT), we observe that low-frequency components capture global motion trajectories, while high-frequency components encode precise timing, alignment, and contact behaviors. Motivated by this structure, we propose Causal Spectral Policy (CSP), which models action generation as a causal coarse-to-fine process: coarse motion is predicted from observation and language, and fine corrections are generated conditionally on the realized trajectory. Across simulation and real-world evaluations, CSP consistently outperforms strong baselines on precision-sensitive manipulation tasks. Additionally, we propose human-inspired teleoperation noise injection as a data augmentation method under which our approach demonstrates strong robustness to noisy demonstrations

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-cao26u, title = {Hierarchical Policy Learning via Spectral Decomposition}, author = {Cao, Shuxin and Wang, Liquan and Byrnes, Walker and Chen, Yiye and Du, Yilun and Garg, Animesh}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {11518--11535}, 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/cao26u/cao26u.pdf}, url = {https://proceedings.mlr.press/v306/cao26u.html}, abstract = {In this paper, we identify a semantic decomposition in robot action sequences, separating task-level motion intent from execution-level refinements. By analyzing actions in the spectral domain using the discrete cosine transform (DCT), we observe that low-frequency components capture global motion trajectories, while high-frequency components encode precise timing, alignment, and contact behaviors. Motivated by this structure, we propose Causal Spectral Policy (CSP), which models action generation as a causal coarse-to-fine process: coarse motion is predicted from observation and language, and fine corrections are generated conditionally on the realized trajectory. Across simulation and real-world evaluations, CSP consistently outperforms strong baselines on precision-sensitive manipulation tasks. Additionally, we propose human-inspired teleoperation noise injection as a data augmentation method under which our approach demonstrates strong robustness to noisy demonstrations} }
Endnote
%0 Conference Paper %T Hierarchical Policy Learning via Spectral Decomposition %A Shuxin Cao %A Liquan Wang %A Walker Byrnes %A Yiye Chen %A Yilun Du %A Animesh Garg %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-cao26u %I PMLR %P 11518--11535 %U https://proceedings.mlr.press/v306/cao26u.html %V 306 %X In this paper, we identify a semantic decomposition in robot action sequences, separating task-level motion intent from execution-level refinements. By analyzing actions in the spectral domain using the discrete cosine transform (DCT), we observe that low-frequency components capture global motion trajectories, while high-frequency components encode precise timing, alignment, and contact behaviors. Motivated by this structure, we propose Causal Spectral Policy (CSP), which models action generation as a causal coarse-to-fine process: coarse motion is predicted from observation and language, and fine corrections are generated conditionally on the realized trajectory. Across simulation and real-world evaluations, CSP consistently outperforms strong baselines on precision-sensitive manipulation tasks. Additionally, we propose human-inspired teleoperation noise injection as a data augmentation method under which our approach demonstrates strong robustness to noisy demonstrations
APA
Cao, S., Wang, L., Byrnes, W., Chen, Y., Du, Y. & Garg, A.. (2026). Hierarchical Policy Learning via Spectral Decomposition. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:11518-11535 Available from https://proceedings.mlr.press/v306/cao26u.html.

Related Material