pith. sign in

arxiv: 2606.26201 · v1 · pith:RBA2MO7Mnew · submitted 2026-06-24 · 💻 cs.RO

OmniContact: Chaining Meta-Skills via Contact Flow for Generalizable Humanoid Loco-Manipulation

Runyi Yu , Xiaoyi Lin , Ji Ma , Yinhuai Wang , Koukou Luo , Jiahao Ji , Huayi Wang , Wenjia Wang
show 6 more authors
Runhan Zhang Ping Tan Ting Wu Ruoli Dai Qifeng Chen Lei Han
This is my paper

Pith reviewed 2026-06-26 02:00 UTC · model grok-4.3

classification 💻 cs.RO
keywords humanoid loco-manipulationcontact flowmeta-skillsskill chainingautonomous recoveryhierarchical frameworkloco-manipulation dataset
0
0 comments X

The pith

Contact flow representation lets humanoid robots chain meta-skills for long-horizon loco-manipulation with high success.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper seeks to address the dual challenge of executing meta-skills robustly and chaining them in closed loop with recovery for humanoid loco-manipulation tasks. It introduces contact flow as a compact shared interface of key body trajectories and binary contact signals to connect low-level skill learning with high-level sequence synthesis. A sympathetic reader would care because prior methods either provide precise but hard-to-plan interaction details or compact but uninterpretable embeddings that hinder reliable composition over extended horizons. If the approach holds, robots gain the ability to perform tasks like carrying and stacking boxes while recovering from failures and incorporating language-based task breakdowns.

Core claim

OmniContact centers on contact flow, a compact representation consisting of key body trajectories and time-series binary contact signals. This shared interface supports a low-level policy called CF-Track that learns a unified library of loco-manipulation skills and a high-level module called CF-Gen that heuristically synthesizes future contact-flow sequences. Together with the collected OmniContact MoCap-based dataset, the framework enables robust execution, autonomous failure recovery, and flexible composition of meta-skills, demonstrated by 98.7 percent success on Carry Box and 76.5 percent on Push-Stack Boxes while outperforming baselines.

What carries the argument

Contact flow (CF), the compact representation of key body trajectories and time-series binary contact signals that acts as the shared interface between low-level skill execution and high-level sequence composition.

If this is right

  • The low-level policy learns a unified library of loco-manipulation skills from the contact flow interface.
  • The high-level module can synthesize sequences that include autonomous recovery from failures.
  • The framework integrates directly with vision-language models for semantic task decomposition into meta-skills.
  • Complex behaviors become possible, such as arranging scattered boxes into specified shapes like a heart.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Contact flow might transfer to non-humanoid robots if their bodies can produce analogous trajectory and contact signals.
  • Binary contact signals could prove sufficient for bridging perception and planning in other contact-rich manipulation domains.
  • The dataset collection method suggests a scalable way to gather training data for similar hierarchical skill systems.
  • Extending the approach to fully dynamic scenes with moving obstacles would test whether the representation remains stable.

Load-bearing premise

Contact flow serves as a sufficient shared interface that preserves enough information for both robust low-level execution and reliable high-level composition with autonomous recovery.

What would settle it

A long-horizon task where contact flow sequences lose critical object interaction details, causing the high-level module to produce compositions with success rates no higher than prior baselines.

Figures

Figures reproduced from arXiv: 2606.26201 by Huayi Wang, Jiahao Ji, Ji Ma, Koukou Luo, Lei Han, Ping Tan, Qifeng Chen, Runhan Zhang, Runyi Yu, Ruoli Dai, Ting Wu, Wenjia Wang, Xiaoyi Lin, Yinhuai Wang.

Figure 1
Figure 1. Figure 1: OmniContact for Generalizable Loco-Manipulation. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of OmniContact. Given task goals and object states, CF-Gen heuristically synthesizes kinematic contact-flow segments, and CF-Track executes these segments through a robust low-level policy. CF-Gen LocoMotion Recovery 𝑡0 𝑡1 𝑡2 0 1 time time 0 1 𝑡0 𝑡1 𝑡2 𝑡3 𝑡4 𝑡5 𝑡6 𝑡0 𝑡1 𝑡2 0 1 time Carry Push Kick 𝑡0 𝑡1 𝑡2 𝑡3 𝑡4 𝟎 (no contact) 𝟏 (contact) Contact States CF-Track CF-Gen 𝑡0 𝑡1 𝑡2 𝑡3 𝑡0 𝑡1 𝑡2 𝑡3 time… view at source ↗
Figure 4
Figure 4. Figure 4: Scaling with HOI data size. cannot generalize to randomized test states. These failures highlight that relying solely on body kinemat￾ics or narrow trajectory memorization is insufficient for robust HOI. (2) Long-horizon composability. OmniContact successfully solves multi-stage tasks, whereas all baselines completely fail (0%) due to frag￾ile long-horizon execution (Stack Boxes) or missing skill transitio… view at source ↗
Figure 5
Figure 5. Figure 5: VLM integration examples. Given prompt: “Arrange scattered boxes into a heart shape.” diverse data distributions, highlighting its promising potential as a robust, universal foundation for HOI tracking. 5. Conclusion and Discussion We introduced OmniContact, a hierarchical frame￾work that leverages contact flow to bridge the gap be￾tween high-level task reasoning and low-level whole￾body execution. By unif… view at source ↗
Figure 6
Figure 6. Figure 6: Skill coverage of the OmniContact dataset. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative rollouts of representative loco-manipulation skills. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Progress visualizations from VLM-guided planning rollouts. Each row shows five frames sampled from a demonstration video in temporal order. The first two examples require language-grounded object selection and goal assignment, while the last two require concept￾driven spatial decomposition into object-level target poses. The “Noitom” task additionally requires matching each box to the target location with … view at source ↗
Figure 9
Figure 9. Figure 9: VLM-related failure cases. Representative failures include placing a box intended for the “R” target onto the “O” target with mismatched colors, pushing the basket into the cylinder and displacing it instead of first moving the basket near the cylinder for pickup, and a low-level execution failure where excessive robot rotation causes a fall. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
read the original abstract

Learning long-horizon humanoid loco-manipulation poses a dual challenge: it requires not only the robust execution of meta-skills but also their seamless, closed-loop chaining equipped with autonomous recovery. Existing approaches remain limited: explicit humanoid-object interaction representations offer precision but are notoriously difficult for high-level planning, whereas implicit skill embeddings are compact but lack the interpretability required for reliable composition. We propose \ours, a hierarchical framework centered on \textbf{contact flow (CF)}, a compact representation consisting of key body trajectories and time-series binary contact signals. Leveraging this shared interface, our low-level policy \textbf{CF-Track} learns a unified library of loco-manipulation skills, while our high-level module \textbf{CF-Gen} heuristically synthesizes future contact-flow sequences. To support this setting, we additionally collect the OmniContact dataset, a MoCap-based HOI corpus for humanoid loco-manipulation (Appendix~\ref{sec:dataset}). Together, they enable robust execution, autonomous failure recovery, and flexible composition of meta-skills for long-horizon tasks. Experiments show that OmniContact achieves \(98.7\%\) success on \textit{Carry Box} and \(76.5\%\) on \textit{Push-Stack Boxes}, outperforming prior baselines by average margins of \(40.9\%\) in meta-skill and \(66.5\%\) in skill chaining. Besides, our framework naturally integrates with VLMs for semantic task decomposition, enabling complex, semantically grounded loco-manipulation behaviors, such as arranging scattered boxes into a heart shape.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces OmniContact, a hierarchical framework for humanoid loco-manipulation that centers on contact flow (CF)—a representation of key body trajectories plus binary contact time-series—as a shared interface. CF-Track learns a library of meta-skills for low-level execution while CF-Gen heuristically synthesizes CF sequences for high-level chaining with autonomous recovery; a new MoCap-based OmniContact dataset supports training. Experiments report 98.7% success on Carry Box and 76.5% on Push-Stack Boxes, with average gains of 40.9% (meta-skill) and 66.5% (chaining) over baselines, plus natural VLM integration for semantic decomposition.

Significance. If the reported performance holds under rigorous evaluation and the binary-contact representation is shown to suffice for recovery, the work would provide a concrete, interpretable bridge between low-level control and compositional planning that improves on both explicit HOI models and opaque skill embeddings. The release of the OmniContact dataset constitutes a clear positive contribution to the community.

major comments (2)
  1. [§5] §5 (Experiments): The central empirical claims rest on specific success rates (98.7%, 76.5%) and improvement margins (40.9%, 66.5%), yet the section supplies no information on trial counts, variance or error bars, baseline implementations or hyperparameters, statistical tests, or failure-mode analysis. Without these, the data cannot be assessed as support for the claim that CF is the enabling factor.
  2. [§3.1] §3.1 (Contact Flow definition): CF is defined using binary contact signals that discard force magnitude, friction coefficients, and continuous velocities. The manuscript provides no ablation replacing binary contacts with richer signals nor any analysis showing that the omitted dynamics are unnecessary for autonomous recovery on Push-Stack Boxes; this leaves the sufficiency of the interface for closed-loop chaining unverified.
minor comments (1)
  1. [§4.2] The description of CF-Gen heuristics could include a pseudocode listing or explicit decision rules to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive feedback. The two major comments highlight important aspects of experimental reporting and the design choices in the contact flow representation. We address each point below and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§5] §5 (Experiments): The central empirical claims rest on specific success rates (98.7%, 76.5%) and improvement margins (40.9%, 66.5%), yet the section supplies no information on trial counts, variance or error bars, baseline implementations or hyperparameters, statistical tests, or failure-mode analysis. Without these, the data cannot be assessed as support for the claim that CF is the enabling factor.

    Authors: We agree that the current presentation of results in §5 lacks sufficient detail for rigorous evaluation. In the revised manuscript we will expand the experimental section to report the exact number of trials per task (100 independent rollouts), standard deviations across trials, full baseline implementation details and hyperparameter settings, results of statistical significance tests, and a categorized failure-mode analysis. These additions will make the contribution of contact flow clearer and allow direct assessment of the reported margins. revision: yes

  2. Referee: [§3.1] §3.1 (Contact Flow definition): CF is defined using binary contact signals that discard force magnitude, friction coefficients, and continuous velocities. The manuscript provides no ablation replacing binary contacts with richer signals nor any analysis showing that the omitted dynamics are unnecessary for autonomous recovery on Push-Stack Boxes; this leaves the sufficiency of the interface for closed-loop chaining unverified.

    Authors: Binary contact signals were selected to maintain compactness and interpretability for high-level chaining. The 76.5% success rate on Push-Stack Boxes, which includes autonomous recovery, offers task-level evidence that the representation is sufficient for the evaluated scenarios. To directly address the concern we will add a dedicated paragraph in §3.1 explaining the design rationale and will include, where data permits, a brief comparison of binary versus richer contact signals in the revision. revision: partial

Circularity Check

0 steps flagged

No circularity: experimental claims rest on reported outcomes, not self-referential definitions or fits

full rationale

The paper presents a hierarchical framework using contact flow as a shared interface between low-level CF-Track policies and high-level CF-Gen synthesis, supported by a new MoCap dataset. All load-bearing claims (98.7% Carry Box success, 76.5% Push-Stack success, 40.9% and 66.5% margins) are stated as direct experimental measurements rather than derived quantities. No equations, parameter fits, uniqueness theorems, or self-citations appear in the provided text that would reduce any prediction to an input by construction. The representation choice and dataset collection are presented as design decisions validated externally by task performance, with no self-definitional loops or renamed empirical patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Abstract-only review yields no explicit free parameters or standard axioms; the primary addition is the contact flow representation itself.

invented entities (1)
  • contact flow (CF) no independent evidence
    purpose: Compact shared representation of key body trajectories and time-series binary contact signals for skill learning and chaining
    Introduced in the abstract as the central interface enabling the hierarchical framework

pith-pipeline@v0.9.1-grok · 5861 in / 1041 out tokens · 28509 ms · 2026-06-26T02:00:17.874749+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references · 1 canonical work pages

  1. [1]

    Homie: Humanoid loco- manipulation with isomorphic exoskeleton cockpit.arXiv preprint arXiv:2502.13013, 2025

    Qingwei Ben, Feiyu Jia, Jia Zeng, Junting Dong, Dahua Lin, and Jiangmiao Pang. Homie: Humanoid loco- manipulation with isomorphic exoskeleton cockpit.arXiv preprint arXiv:2502.13013, 2025. 2, 3

    arXiv 2025

  2. [2]

    Expressive whole- body control for humanoid robots.arXiv preprint arXiv:2402.16796, 2024

    Xuxin Cheng, Yandong Ji, Junming Chen, Ruihan Yang, Ge Yang, and Xiaolong Wang. Expressive whole- body control for humanoid robots.arXiv preprint arXiv:2402.16796, 2024. 3

    arXiv 2024

  3. [3]

    Task and motion planning for humanoid loco-manipulation

    Michal Ciebielski, Victor Dhédin, and Majid Khadiv. Task and motion planning for humanoid loco-manipulation. In 2025 IEEE-RAS 24th International Conference on Hu- manoid Robots (Humanoids), pages 1179–1186. IEEE,

    2025

  4. [4]

    Learning humanoid end-effector control for open- vocabulary visual loco-manipulation.arXiv preprint arXiv:2602.16705, 2026

    Runpei Dong, Ziyan Li, Xialin He, and Saurabh Gupta. Learning humanoid end-effector control for open- vocabulary visual loco-manipulation.arXiv preprint arXiv:2602.16705, 2026. 2

    Pith/arXiv arXiv 2026

  5. [5]

    Humanplus: Humanoid shadowing and imitation from humans.arXiv preprint arXiv:2406.10454,

    Zipeng Fu, Qingqing Zhao, Qi Wu, Gordon Wetzstein, and Chelsea Finn. Humanplus: Humanoid shadowing and imitation from humans.arXiv preprint arXiv:2406.10454,

    arXiv

  6. [6]

    Omnih2o: Universal and dexterous human-to- humanoid whole-body teleoperation and learning.arXiv preprint arXiv:2406.08858, 2024

    Tairan He, Zhengyi Luo, Xialin He, Wenli Xiao, Chong Zhang, Weinan Zhang, Kris Kitani, Changliu Liu, and Guanya Shi. Omnih2o: Universal and dexterous human-to- humanoid whole-body teleoperation and learning.arXiv preprint arXiv:2406.08858, 2024. 2, 3

    arXiv 2024

  7. [7]

    Asap: Aligning simulation and real-world physics for learning agile humanoid whole- body skills.arXiv preprint arXiv:2502.01143, 2025

    Tairan He, Jiawei Gao, Wenli Xiao, Yuanhang Zhang, Zi Wang, Jiashun Wang, Zhengyi Luo, Guanqi He, Nikhil Sobanbab, Chaoyi Pan, et al. Asap: Aligning simulation and real-world physics for learning agile humanoid whole- body skills.arXiv preprint arXiv:2502.01143, 2025. 3

    arXiv 2025

  8. [8]

    Viral: Visual sim-to-real at scale for humanoid loco-manipulation.arXiv preprint arXiv:2511.15200, 2025

    Tairan He, Zi Wang, Haoru Xue, Qingwei Ben, Zhengyi Luo, Wenli Xiao, Ye Yuan, Xingye Da, Fernando Cas- tañeda, Shankar Sastry, et al. Viral: Visual sim-to-real at scale for humanoid loco-manipulation.arXiv preprint arXiv:2511.15200, 2025. 2, 3

    arXiv 2025

  9. [9]

    Learning getting-up policies for real-world hu- manoid robots.ArXiv, abs/2502.12152, 2025

    Xialin He, Runpei Dong, Zixuan Chen, and Saurabh Gupta. Learning getting-up policies for real-world hu- manoid robots.ArXiv, abs/2502.12152, 2025. 3

    arXiv 2025

  10. [10]

    Ultra: Unified multi- modal control for autonomous humanoid whole-body loco- manipulation.arXiv preprint arXiv:2603.03279, 2026

    Xialin He, Sirui Xu, Xinyao Li, Runpei Dong, Liuyu Bian, Yu-Xiong Wang, and Liang-Yan Gui. Ultra: Unified multi- modal control for autonomous humanoid whole-body loco- manipulation.arXiv preprint arXiv:2603.03279, 2026. 3

    arXiv 2026

  11. [11]

    Wholebodyvla: Towards unified latent vla for whole-body loco-manipulation control.arXiv preprint arXiv:2512.11047, 2025

    Haoran Jiang, Jin Chen, Qingwen Bu, Li Chen, Modi Shi, Yanjie Zhang, Delong Li, Chuanzhe Suo, Chuang Wang, Zhihui Peng, et al. Wholebodyvla: Towards unified latent vla for whole-body loco-manipulation control.arXiv preprint arXiv:2512.11047, 2025. 4

    arXiv 2025

  12. [12]

    Switch: Learn- ing agile skills switching for humanoid robots.arXiv preprint arXiv:2604.14834, 2026

    Yuen-Fui Lau, Qihan Zhao, Yinhuai Wang, Runyi Yu, Hok Wai Tsui, Qifeng Chen, and Ping Tan. Switch: Learn- ing agile skills switching for humanoid robots.arXiv preprint arXiv:2604.14834, 2026. 3

    Pith/arXiv arXiv 2026

  13. [13]

    Haic: Humanoid agile object in- teraction control via dynamics-aware world model.arXiv preprint arXiv:2602.11758, 2026

    Dongting Li, Xingyu Chen, Qianyang Wu, Bo Chen, Sikai Wu, Hanyu Wu, Guoyao Zhang, Liang Li, Mingliang Zhou, Diyun Xiang, et al. Haic: Humanoid agile object in- teraction control via dynamics-aware world model.arXiv preprint arXiv:2602.11758, 2026. 2

    Pith/arXiv arXiv 2026

  14. [14]

    Amo: Adaptive motion optimization for hyper-dexterous humanoid whole-body control.arXiv preprint arXiv:2505.03738, 2025

    Jialong Li, Xuxin Cheng, Tianshu Huang, Shiqi Yang, Ri- Zhao Qiu, and Xiaolong Wang. Amo: Adaptive motion optimization for hyper-dexterous humanoid whole-body control.arXiv preprint arXiv:2505.03738, 2025. 3

    arXiv 2025

  15. [15]

    Yitang Li, Zhengyi Luo, Tonghe Zhang, Cunxi Dai, Anssi Kanervisto, Andrea Tirinzoni, Haoyang Weng, Kris Ki- tani, Mateusz Guzek, Ahmed Touati, et al. Bfm-zero: A 9 OmniContact : Chaining Meta-Skills via Contact Flow promptable behavioral foundation model for humanoid control using unsupervised reinforcement learning.arXiv preprint arXiv:2511.04131, 2025

    arXiv 2025

  16. [16]

    Hold my beer: Learning gentle humanoid locomotion and end-effector stabilization control

    Yitang Li, Yuanhang Zhang, Wenli Xiao, Chaoyi Pan, Haoyang Weng, Guanqi He, Tairan He, and Guanya Shi. Hold my beer: Learning gentle humanoid locomotion and end-effector stabilization control. InRSS 2025 Workshop on Whole-body Control and Bimanual Manipulation: Ap- plications in Humanoids and Beyond, 2025

    2025

  17. [17]

    Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion.arXiv preprint arXiv:2508.08241, 2025

    Qiayuan Liao, Takara E Truong, Xiaoyu Huang, Yu- man Gao, Guy Tevet, Koushil Sreenath, and C Karen Liu. Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion.arXiv preprint arXiv:2508.08241, 2025. 3

    Pith/arXiv arXiv 2025

  18. [18]

    Lessmimic: Long-horizon hu- manoid interaction with unified distance field representa- tions.arXiv preprint arXiv:2602.21723, 2026

    Yutang Lin, Jieming Cui, Yixuan Li, Baoxiong Jia, Yixin Zhu, and Siyuan Huang. Lessmimic: Long-horizon hu- manoid interaction with unified distance field representa- tions.arXiv preprint arXiv:2602.21723, 2026. 2, 3, 7, 15, 17, 21

    arXiv 2026

  19. [19]

    Opt2skill: Imitating dynamically- feasible whole-body trajectories for versatile humanoid loco-manipulation.IEEE Robotics and Automation Let- ters, 2025

    Fukang Liu, Zhaoyuan Gu, Yilin Cai, Ziyi Zhou, Hyun- young Jung, Jaehwi Jang, Shijie Zhao, Sehoon Ha, Yue Chen, Danfei Xu, et al. Opt2skill: Imitating dynamically- feasible whole-body trajectories for versatile humanoid loco-manipulation.IEEE Robotics and Automation Let- ters, 2025. 2

    2025

  20. [20]

    Ego-vision world model for humanoid contact planning.arXiv preprint arXiv:2510.11682, 2025

    Hang Liu, Yuman Gao, Sangli Teng, Yufeng Chi, Yakun Sophia Shao, Zhongyu Li, Maani Ghaffari, and Koushil Sreenath. Ego-vision world model for humanoid contact planning.arXiv preprint arXiv:2510.11682, 2025. 4

    arXiv 2025

  21. [21]

    Sonic: Supersizing motion tracking for natural humanoid whole-body control.arXiv preprint arXiv:2511.07820, 2025

    Zhengyi Luo, Ye Yuan, Tingwu Wang, Chenran Li, Sirui Chen, Fernando Castaneda, Zi-Ang Cao, Jiefeng Li, David Minor, Qingwei Ben, et al. Sonic: Supersizing motion tracking for natural humanoid whole-body control.arXiv preprint arXiv:2511.07820, 2025. 2, 3, 7, 15, 16, 21

    Pith/arXiv arXiv 2025

  22. [22]

    Carlson, Ji Yuan Feng, Animesh Garg, Renato Gasoto, Lionel Gulich, Yijie Guo, M

    Mayank Mittal, Pascal Roth, James Tigue, Antoine Richard, Octi Zhang, Peter Du, Antonio Serrano-Muñoz, Xinjie Yao, René Zurbrügg, Nikita Rudin, Lukasz Wawrzyniak, Milad Rakhsha, Alain Denzler, Eric Heiden, Ales Borovicka, Ossama Ahmed, Iretiayo Akinola, Abrar Anwar, Mark T. Carlson, Ji Yuan Feng, Animesh Garg, Renato Gasoto, Lionel Gulich, Yijie Guo, M. G...

    Pith/arXiv arXiv 2025

  23. [23]

    Amp: Adversarial motion priors for stylized physics-based character control.ACM Transac- tions on Graphics (ToG), 40(4):1–20, 2021

    Xue Bin Peng, Ze Ma, Pieter Abbeel, Sergey Levine, and Angjoo Kanazawa. Amp: Adversarial motion priors for stylized physics-based character control.ACM Transac- tions on Graphics (ToG), 40(4):1–20, 2021. 3

    2021

  24. [24]

    Humanoid locomotion as next token prediction.Advances in neural information processing systems, 37:79307–79324, 2024

    Ilija Radosavovic, Bike Zhang, Baifeng Shi, Jathushan Rajasegaran, Sarthak Kamat, Trevor Darrell, Koushil Sreenath, and Jitendra Malik. Humanoid locomotion as next token prediction.Advances in neural information processing systems, 37:79307–79324, 2024. 3

    2024

  25. [25]

    Humanoid goalkeeper: Learning from position conditioned task-motion constraints.arXiv preprint arXiv:2510.18002, 2025

    Junli Ren, Junfeng Long, Tao Huang, Huayi Wang, Zirui Wang, Feiyu Jia, Wentao Zhang, Jingbo Wang, Ping Luo, and Jiangmiao Pang. Humanoid goalkeeper: Learning from position conditioned task-motion constraints.arXiv preprint arXiv:2510.18002, 2025. 3

    arXiv 2025

  26. [26]

    Hierarchical vision-language planning for multi-step humanoid manipulation.arXiv preprint arXiv:2506.22827, 2025

    André Schakkal, Ben Zandonati, Zhutian Yang, and Navid Azizan. Hierarchical vision-language planning for multi-step humanoid manipulation.arXiv preprint arXiv:2506.22827, 2025. 4

    arXiv 2025

  27. [27]

    Langwbc: Language-directed humanoid whole-body control via end-to-end learning

    Yiyang Shao, Xiaoyu Huang, Bike Zhang, Qiayuan Liao, Yuman Gao, Yufeng Chi, Zhongyu Li, Sophia Shao, and Koushil Sreenath. Langwbc: Language-directed humanoid whole-body control via end-to-end learning. ArXiv, abs/2504.21738, 2025. 3

    arXiv 2025

  28. [28]

    Egohumanoid: Unlocking in-the-wild loco-manipulation with robot-free egocentric demonstration.arXiv preprint arXiv:2602.10106, 2026

    Modi Shi, Shijia Peng, Jin Chen, Haoran Jiang, Yinghui Li, Di Huang, Ping Luo, Hongyang Li, and Li Chen. Egohumanoid: Unlocking in-the-wild loco-manipulation with robot-free egocentric demonstration.arXiv preprint arXiv:2602.10106, 2026. 2, 3

    Pith/arXiv arXiv 2026

  29. [29]

    Hitter: A humanoid table tennis robot via hierarchical planning and learning.arXiv preprint arXiv:2508.21043, 2025

    Zhi Su, Bike Zhang, Nima Rahmanian, Yuman Gao, Qiayuan Liao, Caitlin Regan, Koushil Sreenath, and S Shankar Sastry. Hitter: A humanoid table tennis robot via hierarchical planning and learning.arXiv preprint arXiv:2508.21043, 2025. 3

    arXiv 2025

  30. [30]

    Ulc: A unified and fine- grained controller for humanoid loco-manipulation.arXiv preprint arXiv:2507.06905, 2025

    Wandong Sun, Luying Feng, Baoshi Cao, Yang Liu, Yaochu Jin, and Zongwu Xie. Ulc: A unified and fine- grained controller for humanoid loco-manipulation.arXiv preprint arXiv:2507.06905, 2025. 2

    arXiv 2025

  31. [31]

    Physically consistent humanoid loco- manipulation using latent diffusion models

    Ilyass Taouil, Haizhou Zhao, Angela Dai, and Ma- jid Khadiv. Physically consistent humanoid loco- manipulation using latent diffusion models. In2025 IEEE- RAS 24th International Conference on Humanoid Robots (Humanoids), pages 1–8. IEEE, 2025. 4

    2025

  32. [32]

    Maskedmimic: Unified physics-based character control through masked motion inpainting.ACM Transactions On Graphics (TOG), 43(6):1–21, 2024

    Chen Tessler, Yunrong Guo, Ofir Nabati, Gal Chechik, and Xue Bin Peng. Maskedmimic: Unified physics-based character control through masked motion inpainting.ACM Transactions On Graphics (TOG), 43(6):1–21, 2024. 3, 4

    2024

  33. [33]

    Todorov, T

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mu- joco: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026–5033. IEEE, 2012. doi: 10.1109/IROS.2012.6386109. 15

    work page doi:10.1109/iros.2012.6386109 2012

  34. [34]

    Beamdojo: Learning agile humanoid locomotion on sparse footholds

    Huayi Wang, Zirui Wang, Junli Ren, Qingwei Ben, Tao Huang, Weinan Zhang, and Jiangmiao Pang. Beamdojo: Learning agile humanoid locomotion on sparse footholds. ArXiv, abs/2502.10363, 2025. 3

    arXiv 2025

  35. [35]

    Physhsi: Towards a real-world generalizable and natural humanoid-scene interaction system.arXiv preprint arXiv:2510.11072, 2025

    Huayi Wang, Wentao Zhang, Runyi Yu, Tao Huang, Junli Ren, Feiyu Jia, Zirui Wang, Xiaojie Niu, Xiao Chen, Jiahe Chen, et al. Physhsi: Towards a real-world generalizable and natural humanoid-scene interaction system.arXiv preprint arXiv:2510.11072, 2025. 2, 3, 7, 15, 17, 21

    arXiv 2025

  36. [36]

    Physhoi: Physics-based imita- tion of dynamic human-object interaction.arXiv preprint arXiv:2312.04393, 2023

    Yinhuai Wang, Jing Lin, Ailing Zeng, Zhengyi Luo, Jian Zhang, and Lei Zhang. Physhoi: Physics-based imita- tion of dynamic human-object interaction.arXiv preprint arXiv:2312.04393, 2023. 2

    arXiv 2023

  37. [37]

    Skillmimic: Learning basketball inter- action skills from demonstrations

    Yinhuai Wang, Qihan Zhao, Runyi Yu, Hok Wai Tsui, Ailing Zeng, Jing Lin, Zhengyi Luo, Jiwen Yu, Xiu Li, Qifeng Chen, et al. Skillmimic: Learning basketball inter- action skills from demonstrations. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17540–17549, 2025. 2

    2025

  38. [38]

    Humanx: Toward agile and gener- alizable humanoid interaction skills from human videos

    Yinhuai Wang, Qihan Zhao, Yuen Fui Lau, Runyi Yu, Hok Wai Tsui, Qifeng Chen, Jingbo Wang, Jiangmiao Pang, and Ping Tan. Humanx: Toward agile and gener- alizable humanoid interaction skills from human videos. arXiv preprint arXiv:2602.02473, 2026. 2, 3

    arXiv 2026

  39. [39]

    Hdmi: Learning interactive humanoid whole-body control from human videos.arXiv preprint arXiv:2509.16757, 2025

    Haoyang Weng, Yitang Li, Nikhil Sobanbabu, Zihan Wang, Zhengyi Luo, Tairan He, Deva Ramanan, and 10 OmniContact : Chaining Meta-Skills via Contact Flow Guanya Shi. Hdmi: Learning interactive humanoid whole-body control from human videos.arXiv preprint arXiv:2509.16757, 2025. 2, 3, 7, 15, 16, 21

    arXiv 2025

  40. [40]

    Sugar: A scalable human-video-driven generalizable humanoid loco-manipulation learning framework.arXiv preprint arXiv:2605.20373, 2026

    Tianshu Wu, Xiangqi Kong, Yue Chen, Qize Yu, Hang Ye, Jia Li, Yizhou Wang, and Hao Dong. Sugar: A scalable human-video-driven generalizable humanoid loco-manipulation learning framework.arXiv preprint arXiv:2605.20373, 2026. 2

    Pith/arXiv arXiv 2026

  41. [41]

    Uniphys: Unified planner and controller with dif- fusion for flexible physics-based character control

    Yan Wu, Korrawe Karunratanakul, Zhengyi Luo, and Siyu Tang. Uniphys: Unified planner and controller with dif- fusion for flexible physics-based character control. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 13214–13224, 2025. 3, 4

    2025

  42. [42]

    Parc: Physics-based augmentation with reinforcement learning for character controllers

    Michael Xu, Yi Shi, KangKang Yin, and Xue Bin Peng. Parc: Physics-based augmentation with reinforcement learning for character controllers. InProceedings of the Special Interest Group on Computer Graphics and Inter- active Techniques Conference Conference Papers, pages 1–11, 2025. 3, 4

    2025

  43. [43]

    Opening the sim-to-real door for humanoid pixel-to-action policy transfer.arXiv preprint arXiv:2512.01061, 2025

    Haoru Xue, Tairan He, Zi Wang, Qingwei Ben, Wenli Xiao, Zhengyi Luo, Xingye Da, Fernando Castañeda, Guanya Shi, Shankar Sastry, et al. Opening the sim-to-real door for humanoid pixel-to-action policy transfer.arXiv preprint arXiv:2512.01061, 2025. 2, 3

    arXiv 2025

  44. [44]

    Leverb: Humanoid whole-body control with la- tent vision-language instruction,(2025).URL https://arxiv

    Haoru Xue, Xiaoyu Huang, Dantong Niu, Qiayuan Liao, Thomas Kragerud, Jan Tommy Gravdahl, Xue Bin Peng, Guanya Shi, Trevor Darrell, Koushil Sreenath, et al. Leverb: Humanoid whole-body control with la- tent vision-language instruction,(2025).URL https://arxiv. org/abs/2506.13751, 3(10), 2025. 4

    arXiv 2025

  45. [45]

    A unified and general humanoid whole-body controller for fine-grained locomotion.ArXiv, abs/2502.03206, 2025

    Yufei Xue, Wentao Dong, Minghuan Liu, Weinan Zhang, and Jiangmiao Pang. A unified and general humanoid whole-body controller for fine-grained locomotion.ArXiv, abs/2502.03206, 2025. 3

    arXiv 2025

  46. [46]

    Omniretarget: Interaction- preserving data generation for humanoid whole-body loco-manipulation and scene interaction.arXiv preprint arXiv:2509.26633, 2025

    Lujie Yang, Xiaoyu Huang, Zhen Wu, Angjoo Kanazawa, Pieter Abbeel, Carmelo Sferrazza, C Karen Liu, Rocky Duan, and Guanya Shi. Omniretarget: Interaction- preserving data generation for humanoid whole-body loco-manipulation and scene interaction.arXiv preprint arXiv:2509.26633, 2025. 2, 3

    Pith/arXiv arXiv 2025

  47. [47]

    Visualmimic: Visual humanoid loco- manipulation via motion tracking and generation.arXiv preprint arXiv:2509.20322, 2025

    Shaofeng Yin, Yanjie Ze, Hong-Xing Yu, C Karen Liu, and Jiajun Wu. Visualmimic: Visual humanoid loco- manipulation via motion tracking and generation.arXiv preprint arXiv:2509.20322, 2025. 2, 3

    arXiv 2025

  48. [48]

    Skillmimic- v2: Learning robust and generalizable interaction skills from sparse and noisy demonstrations

    Runyi Yu, Yinhuai Wang, Qihan Zhao, Hok Wai Tsui, Jingbo Wang, Ping Tan, and Qifeng Chen. Skillmimic- v2: Learning robust and generalizable interaction skills from sparse and noisy demonstrations. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers, pages 1–11, 2025. 2

    2025

  49. [49]

    Twist: Tele- operated whole-body imitation system.arXiv preprint arXiv:2505.02833, 2025

    Yanjie Ze, Zixuan Chen, Joao Pedro Araújo, Zi-ang Cao, Xue Bin Peng, Jiajun Wu, and C Karen Liu. Twist: Tele- operated whole-body imitation system.arXiv preprint arXiv:2505.02833, 2025. 2, 3

    arXiv 2025

  50. [50]

    Wococo: Learning whole-body humanoid control with sequential contacts

    Chong Zhang, Wenli Xiao, Tairan He, and Guanya Shi. Wococo: Learning whole-body humanoid control with sequential contacts. InConference on Robot Learning, pages 455–472. PMLR, 2025. 2, 3

    2025

  51. [51]

    Falcon: Learning force- adaptive humanoid loco-manipulation.arXiv preprint arXiv:2505.06776, 2025

    Yuanhang Zhang, Yifu Yuan, Prajwal Gurunath, Ishita Gupta, Shayegan Omidshafiei, Ali-akbar Agha- mohammadi, Marcell Vazquez-Chanlatte, Liam Peder- sen, Tairan He, and Guanya Shi. Falcon: Learning force- adaptive humanoid loco-manipulation.arXiv preprint arXiv:2505.06776, 2025. 3

    arXiv 2025

  52. [52]

    Learning athletic humanoid tennis skills from imperfect human motion data.arXiv preprint arXiv:2603.12686, 2026

    Zhikai Zhang, Haofei Lu, Yunrui Lian, Ziqing Chen, Yun Liu, Chenghuai Lin, Han Xue, Zicheng Zeng, Zekun Qi, Shaolin Zheng, et al. Learning athletic humanoid tennis skills from imperfect human motion data.arXiv preprint arXiv:2603.12686, 2026. 3

    arXiv 2026

  53. [53]

    Resmimic: From general motion tracking to humanoid whole-body loco-manipulation via residual learning.arXiv preprint arXiv:2510.05070, 2025

    Siheng Zhao, Yanjie Ze, Yue Wang, C Karen Liu, Pieter Abbeel, Guanya Shi, and Rocky Duan. Resmimic: From general motion tracking to humanoid whole-body loco-manipulation via residual learning.arXiv preprint arXiv:2510.05070, 2025. 3

    arXiv 2025

  54. [54]

    Humanoid parkour learning.arXiv preprint arXiv:2406.10759, 2024

    Ziwen Zhuang, Shenzhe Yao, and Hang Zhao. Humanoid parkour learning.arXiv preprint arXiv:2406.10759, 2024. 3 11 OmniContact : Chaining Meta-Skills via Contact Flow Appendix A. Dataset We introduce theOmniContact dataset, a compre- hensive human-object interaction (HOI) corpus tai- lored specifically for humanoid loco-manipulation. It captures object-const...

    arXiv 2024

  55. [55]

    Walking stability: balance and gait quality

  56. [56]

    Box contact: whether the hands/body contact the box in a plausible carrying pose

  57. [57]

    Box stability: whether the box moves smoothly without obvious sliding, bouncing, penetration, or falling

  58. [58]

    Motion smoothness: absence of sudden jitter, joint twitching, or velocity discontinuities

  59. [59]

    Output a table in the following format: Video ID | Success Valid | Naturalness Score | Main Reason F

    Task-level naturalness: whether the robot moves the box near the target in a reasonable way. Output a table in the following format: Video ID | Success Valid | Naturalness Score | Main Reason F. Compatibility with VLMs The compact and structured representation of contact flow provides a natural interface for high-level seman- tic planners, such as vision-...

  60. [60]

    A top-down image of the scene with movable objects

  61. [61]

    A natural-language task instruction

  62. [62]

    task_type

    Available meta-skills: pick-place, push , kick, and spatial rearrangement. Your job: - Identify the task-relevant objects from the image. - Convert the instruction into object- level subgoals. - For each subgoal, choose a meta-skill and specify the target pose or target region. - Do not output humanoid joint motions, contact timings, or low-level controls...