Knowledge Diversion for Efficient Morphology Control and Policy Transfer

Fu Feng, Ruixiao Shi, Yucheng Xie, Jianlu Shen, Jing Wang, Xin Geng
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:30342-30358, 2026.

Abstract

Universal morphology control aims to learn a universal policy that generalizes across heterogeneous robot morphologies, with Transformer-based controllers emerging as a dominant choice. However, such architectures incur substantial computational costs, resulting in high deployment overhead, and existing methods exhibit limited cross-task generalization, necessitating training from scratch for each new task. To this end, we propose DivMorph, a modular training paradigm that leverages knowledge diversion to learn decomposable controllers. DivMorph factorizes randomly initialized Transformer weights into basic knowledge units via SVD and employs dynamic soft gating, conditioned on task and morphology embeddings, to adaptively modulate these units into universal learngenes and morphology- and task-specific tailors during training, thereby achieving knowledge disentanglement. By selectively activating relevant components, DivMorph adaptively recomposes the controller, enabling efficient policy deployment and effective policy transfer to novel tasks. Extensive experiments demonstrate that DivMorph achieves state-of-the-art performance, improving sample efficiency for cross-task transfer by 3.3$\times$ and reducing model size for single-agent deployment by 16.7$\times$.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-feng26n, title = {Knowledge Diversion for Efficient Morphology Control and Policy Transfer}, author = {Feng, Fu and Shi, Ruixiao and Xie, Yucheng and Shen, Jianlu and Wang, Jing and Geng, Xin}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {30342--30358}, 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/feng26n/feng26n.pdf}, url = {https://proceedings.mlr.press/v306/feng26n.html}, abstract = {Universal morphology control aims to learn a universal policy that generalizes across heterogeneous robot morphologies, with Transformer-based controllers emerging as a dominant choice. However, such architectures incur substantial computational costs, resulting in high deployment overhead, and existing methods exhibit limited cross-task generalization, necessitating training from scratch for each new task. To this end, we propose DivMorph, a modular training paradigm that leverages knowledge diversion to learn decomposable controllers. DivMorph factorizes randomly initialized Transformer weights into basic knowledge units via SVD and employs dynamic soft gating, conditioned on task and morphology embeddings, to adaptively modulate these units into universal learngenes and morphology- and task-specific tailors during training, thereby achieving knowledge disentanglement. By selectively activating relevant components, DivMorph adaptively recomposes the controller, enabling efficient policy deployment and effective policy transfer to novel tasks. Extensive experiments demonstrate that DivMorph achieves state-of-the-art performance, improving sample efficiency for cross-task transfer by 3.3$\times$ and reducing model size for single-agent deployment by 16.7$\times$.} }
Endnote
%0 Conference Paper %T Knowledge Diversion for Efficient Morphology Control and Policy Transfer %A Fu Feng %A Ruixiao Shi %A Yucheng Xie %A Jianlu Shen %A Jing Wang %A Xin Geng %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-feng26n %I PMLR %P 30342--30358 %U https://proceedings.mlr.press/v306/feng26n.html %V 306 %X Universal morphology control aims to learn a universal policy that generalizes across heterogeneous robot morphologies, with Transformer-based controllers emerging as a dominant choice. However, such architectures incur substantial computational costs, resulting in high deployment overhead, and existing methods exhibit limited cross-task generalization, necessitating training from scratch for each new task. To this end, we propose DivMorph, a modular training paradigm that leverages knowledge diversion to learn decomposable controllers. DivMorph factorizes randomly initialized Transformer weights into basic knowledge units via SVD and employs dynamic soft gating, conditioned on task and morphology embeddings, to adaptively modulate these units into universal learngenes and morphology- and task-specific tailors during training, thereby achieving knowledge disentanglement. By selectively activating relevant components, DivMorph adaptively recomposes the controller, enabling efficient policy deployment and effective policy transfer to novel tasks. Extensive experiments demonstrate that DivMorph achieves state-of-the-art performance, improving sample efficiency for cross-task transfer by 3.3$\times$ and reducing model size for single-agent deployment by 16.7$\times$.
APA
Feng, F., Shi, R., Xie, Y., Shen, J., Wang, J. & Geng, X.. (2026). Knowledge Diversion for Efficient Morphology Control and Policy Transfer. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:30342-30358 Available from https://proceedings.mlr.press/v306/feng26n.html.

Related Material