HiMe: Hierarchical Embodied Memory for Long-Horizon Vision-Language-Action Control

Li Ji, Siyin Wang, Pengfang Qian, Xiaopeng Yu, Yihai Tian, Zhaoye Fei, Jingjing Gong, Xipeng Qiu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:51611-51631, 2026.

Abstract

Current Vision-Language-Action (VLA) models excel at robotic manipulation but often struggle with non-Markovian tasks requiring long-term memory and reasoning due to their reliance on immediate observations. Existing solutions face a frequency-competence paradox, where high-performance models are too slow for real-time control, while faster models lack sufficient reasoning capabilities. To resolve this architectural misalignment, we propose HiMe, a Hierarchical Embodied Memory framework that decouples embodied intelligence into a high-frequency Executor for execution, a Sentry for working memory, and a Planner for long-term strategy. We also introduce a dynamic knowledge system based on cross-modal semantic schemas and active management mechanisms, allowing robots to maintain memory plasticity through "Add, Update, and Delete" operations. This hierarchical design effectively balances the conflict between real-time execution and slow thinking planning, significantly improving success rates in long-horizon tasks. Experiments demonstrate that this approach not only outperforms flat memory baselines but also exhibits the novel ability to self-correct its internal knowledge based on human preferences.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-ji26k, title = {{H}i{M}e: Hierarchical Embodied Memory for Long-Horizon Vision-Language-Action Control}, author = {Ji, Li and Wang, Siyin and Qian, Pengfang and Yu, Xiaopeng and Tian, Yihai and Fei, Zhaoye and Gong, Jingjing and Qiu, Xipeng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {51611--51631}, 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/ji26k/ji26k.pdf}, url = {https://proceedings.mlr.press/v306/ji26k.html}, abstract = {Current Vision-Language-Action (VLA) models excel at robotic manipulation but often struggle with non-Markovian tasks requiring long-term memory and reasoning due to their reliance on immediate observations. Existing solutions face a frequency-competence paradox, where high-performance models are too slow for real-time control, while faster models lack sufficient reasoning capabilities. To resolve this architectural misalignment, we propose HiMe, a Hierarchical Embodied Memory framework that decouples embodied intelligence into a high-frequency Executor for execution, a Sentry for working memory, and a Planner for long-term strategy. We also introduce a dynamic knowledge system based on cross-modal semantic schemas and active management mechanisms, allowing robots to maintain memory plasticity through "Add, Update, and Delete" operations. This hierarchical design effectively balances the conflict between real-time execution and slow thinking planning, significantly improving success rates in long-horizon tasks. Experiments demonstrate that this approach not only outperforms flat memory baselines but also exhibits the novel ability to self-correct its internal knowledge based on human preferences.} }
Endnote
%0 Conference Paper %T HiMe: Hierarchical Embodied Memory for Long-Horizon Vision-Language-Action Control %A Li Ji %A Siyin Wang %A Pengfang Qian %A Xiaopeng Yu %A Yihai Tian %A Zhaoye Fei %A Jingjing Gong %A Xipeng Qiu %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-ji26k %I PMLR %P 51611--51631 %U https://proceedings.mlr.press/v306/ji26k.html %V 306 %X Current Vision-Language-Action (VLA) models excel at robotic manipulation but often struggle with non-Markovian tasks requiring long-term memory and reasoning due to their reliance on immediate observations. Existing solutions face a frequency-competence paradox, where high-performance models are too slow for real-time control, while faster models lack sufficient reasoning capabilities. To resolve this architectural misalignment, we propose HiMe, a Hierarchical Embodied Memory framework that decouples embodied intelligence into a high-frequency Executor for execution, a Sentry for working memory, and a Planner for long-term strategy. We also introduce a dynamic knowledge system based on cross-modal semantic schemas and active management mechanisms, allowing robots to maintain memory plasticity through "Add, Update, and Delete" operations. This hierarchical design effectively balances the conflict between real-time execution and slow thinking planning, significantly improving success rates in long-horizon tasks. Experiments demonstrate that this approach not only outperforms flat memory baselines but also exhibits the novel ability to self-correct its internal knowledge based on human preferences.
APA
Ji, L., Wang, S., Qian, P., Yu, X., Tian, Y., Fei, Z., Gong, J. & Qiu, X.. (2026). HiMe: Hierarchical Embodied Memory for Long-Horizon Vision-Language-Action Control. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:51611-51631 Available from https://proceedings.mlr.press/v306/ji26k.html.

Related Material