pith. sign in

arxiv: 2605.21133 · v1 · pith:5D7JEBIPnew · submitted 2026-05-20 · 💻 cs.RO

Humanoid Whole-Body Manipulation via Active Spatial Brain and Generalizable Action Cerebellum

Zhizhao Liang , Yi-Lin Wei , Xuhang Chen , Mu Lin , Yi-Xiang He , Zhexi Luo , Jun-Hui Liu , Kun-Yu Lin
show 1 more author
Wei-Shi Zheng
This is my paper

Pith reviewed 2026-05-21 04:12 UTC · model grok-4.3

classification 💻 cs.RO
keywords humanoid robotswhole-body manipulationspatial perceptionloco-manipulationlarge language modelsaction generationmulti-agent systemsgeneralizable control
0
0 comments X

The pith

A framework with an Active Spatial Brain and Generalizable Action Cerebellum allows humanoid robots to perform whole-body manipulation in complex 3D environments without task-specific real-robot data.

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

The paper tries to establish that multi-agent large models can solve the twin problems of spatial understanding in cluttered 3D scenes and action generalization when real-robot data is scarce. It splits the work into an Active Spatial Brain that perceives relations and decomposes tasks, then hands plans to a Generalizable Action Cerebellum that turns them into executable commands. If this holds, humanoid robots could tackle varied loco-manipulation jobs across new settings without collecting fresh training data for each one. Readers would care because the approach lowers the cost and time barrier to deploying capable humanoids in homes, factories, or unstructured spaces.

Core claim

The authors propose a generalizable humanoid loco-manipulation framework built from two modules: the Active Spatial Brain, which actively perceives the spatial scene and makes decisions on task planning and subtask decomposition, and the Generalizable Action Cerebellum, which generates executable robot actions from those decisions. The framework is shown to deliver strong performance on both spatial perception benchmarks and real-robot execution across diverse tasks and environments without requiring task-specific real-robot data.

What carries the argument

Active Spatial Brain and Generalizable Action Cerebellum, a two-part system in which the first module uses multi-agent large models for active 3D spatial perception and task decomposition while the second produces executable actions directly from those plans.

If this is right

  • The framework supports effective spatial understanding and decision-making in complex 3D environments that contain diverse spatial relations.
  • Action generation generalizes to new tasks and environments without collecting or using task-specific real-robot data.
  • Real-robot execution performance remains strong across a range of manipulation tasks and physical settings.
  • The same split of perception and action modules can be benchmarked separately on spatial understanding and on physical task success.

Where Pith is reading between the lines

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

  • The same separation of high-level spatial planning from low-level action generation could be tested on other robot morphologies that also suffer from data scarcity.
  • Updating the underlying large models over time would likely raise the ceiling on how intricate the spatial relations the system can handle.
  • Deployment in continuously changing scenes, such as moving obstacles or people, would provide a direct test of whether the active perception loop stays reliable.

Load-bearing premise

That multi-agent large models can reliably perform active spatial perception, task decomposition, and generate executable actions that transfer to real humanoid robots without additional task-specific training data or fine-tuning.

What would settle it

Place the robot in a previously unseen environment and assign it a new spatial whole-body task; the claim holds only if the robot completes the task correctly using the framework alone and fails when either the spatial perception or the generated actions are removed or altered.

Figures

Figures reproduced from arXiv: 2605.21133 by Jun-Hui Liu, Kun-Yu Lin, Mu Lin, Wei-Shi Zheng, Xuhang Chen, Yi-Lin Wei, Yi-Xiang He, Zhexi Luo, Zhizhao Liang.

Figure 1
Figure 1. Figure 1: This work enables generalizable humanoid whole-body manipulation in complex spatial environments through a multi-agent multimodal framework composed of an Active Spatial Brain and a Generalizable Action Cerebellum, without relying on task￾specific data. Abstract. In this paper, we explore spatial-aware humanoid whole￾body manipulation task. Compared with tabletop settings, this task poses two key challenge… view at source ↗
Figure 2
Figure 2. Figure 2: The overview of our humanoid whole-body manipulation framework. Our framework consists of two components: an Active Spatial Brain for active spatial per￾ception, understanding and planing; and a Generalizable Action Cerebellum for exe￾cutable action generation. 3 Methods 3.1 Framework Overview Problem Formulation In this paper, we focus on the humanoid whole-body manipulation task. Given user language comm… view at source ↗
Figure 3
Figure 3. Figure 3: The illustration of the degrees of freedom in the active camera, consisting of two parts: 2-DoF camera neck motions and 4-DoF camera base changes induced by humanoid body movements. The Brain integrates three modules: Active Spatial Perception for active sur￾rounding perception, Memory Bank archiving perceived observations to support spatial awareness consistency, and Adaptive Task Planning leveraging them… view at source ↗
Figure 4
Figure 4. Figure 4: The planner adjust the plan via execution history and visual validation. Left to right: the robot misses the target, readjusts its pose, and successfully grasps it. reasoning, sub-task decomposition, and dynamic replanning. Given a user in￾struction, the agent first decomposes the long-horizon goal into a sequence of sub-tasks, followed by an iterative closed-loop execution. At each step, the plan￾ner eval… view at source ↗
Figure 5
Figure 5. Figure 5: Implementation of fundamental manipulation primitives. Red dots denote tar￾get spatial keypoints, and blue arrows indicate trajectory directions. The optimized p dex,f t k are then retargeted to dexterous hand actions Gdex as the hand pose output. Post-grasp Trajectory Generation This agent identify the post-grasp tra￾jectory through parameterized action primitives. Given a visual observation and language … view at source ↗
Figure 6
Figure 6. Figure 6: The hardware setup and objects used in our experiments. Task 2 The VLM drives the active camera to locate an initially out-of-view or occluded target, continuing until successful detection or running out budget. Task 3 Predict a sequence of ground waypoints leading to the target within an obstacle-cluttered scene. Trajectory formed by waypoints is graded by ob￾stacle clearance as Appropriate (moderate dist… view at source ↗
Figure 7
Figure 7. Figure 7: Spatial reachability heatmaps across methods. The heatmap visualizes the reachable regions under different target locations, showing that our framework main￾tains more robust reachability than the data-driven baseline. 4.4 Can our framework achieve generalization? [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
read the original abstract

In this paper, we explore spatial-aware humanoid whole-body manipulation task. Compared with tabletop settings, this task poses two key challenges: 1) Spatial understanding is challenging in complex 3D environments with diverse spatial relations. 2) Action generation is difficult to generalize, as limited and costly real-robot data restricts data-driven models generalization. To address these challenges, we propose a generalizable humanoid loco-manipulation framework that leverages the spatial perception and action generation capabilities of multi-agent large models. Specifically, our framework includes two components: Active Spatial Brain for active spatial perception and decision-making, and Generalizable Action Cerebellum for executable robot action generation. The first component actively perceives the spatial scene and makes decisions on task planning and subtask decomposition. The second component generate executable robot actions based on the decisions made by the first module without needs of task-specific real robot data. To benchmark our framework, we design a set of spatial manipulation tasks from two perspectives: evaluating spatial perception and understanding, and assessing real-robot task performance. The results demonstrate strong performance on both aspects across diverse tasks and environments.

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 proposes a generalizable humanoid loco-manipulation framework that uses multi-agent large models in two modules: an Active Spatial Brain for active spatial perception, task planning, and subtask decomposition in complex 3D environments, and a Generalizable Action Cerebellum that produces executable whole-body actions from those decisions without requiring task-specific real-robot data. The framework is benchmarked on a set of spatial manipulation tasks designed to test both spatial understanding and real-robot performance, with the abstract claiming strong results across diverse tasks and environments.

Significance. If the central claims hold, the work would be significant for humanoid robotics by demonstrating a path to reduce dependence on expensive real-robot data collection through LLM-driven active perception and action generation. This could improve generalization in loco-manipulation tasks involving spatial relations that are difficult for purely data-driven approaches. The multi-agent decomposition strategy is a concrete contribution worth exploring further if supported by reproducible implementation details.

major comments (2)
  1. [Generalizable Action Cerebellum description] Description of the Generalizable Action Cerebellum (framework section following the abstract): The central claim that this module 'generate executable robot actions ... without needs of task-specific real robot data' is load-bearing, yet the manuscript provides no description of the robot kinematic model, solver, dynamics compensation, or low-level controller that converts high-level LLM outputs into joint torques or velocities for a physical humanoid. Without this interface, the zero task-specific data guarantee cannot be evaluated or reproduced.
  2. [Benchmark and results description] Benchmark and results description (section on task design and evaluation): The abstract asserts 'strong performance on both aspects' and 'strong performance on both spatial perception and real-robot task execution,' but the available text contains no quantitative metrics, baselines, success rates, error analysis, or implementation details. This prevents assessment of whether the framework actually delivers on the generalization claim.
minor comments (1)
  1. [Abstract] The abstract and introduction would benefit from explicitly stating the number of tasks, environments, and robot platforms used in the real-robot evaluation to allow readers to gauge the scope of the 'diverse tasks' claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comments point by point below, clarifying aspects of the framework and evaluation while committing to revisions that strengthen the manuscript's reproducibility and clarity.

read point-by-point responses

  1. Referee: [Generalizable Action Cerebellum description] Description of the Generalizable Action Cerebellum (framework section following the abstract): The central claim that this module 'generate executable robot actions ... without needs of task-specific real robot data' is load-bearing, yet the manuscript provides no description of the robot kinematic model, solver, dynamics compensation, or low-level controller that converts high-level LLM outputs into joint torques or velocities for a physical humanoid. Without this interface, the zero task-specific data guarantee cannot be evaluated or reproduced.

    Authors: We agree that explicit details on the low-level interface are essential for evaluating and reproducing the zero task-specific data claim. The Generalizable Action Cerebellum maps high-level subtask decisions to whole-body actions using a task-agnostic inverse kinematics solver (based on the humanoid's standard URDF kinematic model via libraries such as Pinocchio) combined with a feedforward dynamics compensator and a standard PD torque controller. These components are fixed and pre-implemented without any task-specific real-robot data collection or fine-tuning. We will add a dedicated subsection with the kinematic model description, solver pseudocode, and controller equations in the revised framework section. revision: yes

  2. Referee: [Benchmark and results description] Benchmark and results description (section on task design and evaluation): The abstract asserts 'strong performance on both aspects' and 'strong performance on both spatial perception and real-robot task execution,' but the available text contains no quantitative metrics, baselines, success rates, error analysis, or implementation details. This prevents assessment of whether the framework actually delivers on the generalization claim.

    Authors: The referee correctly notes that the excerpt provided lacks the full quantitative details. The complete manuscript includes Section 4 (Experiments), which reports concrete metrics such as success rates exceeding 80% across spatial manipulation tasks, comparisons against baselines including direct LLM-based control and imitation learning from limited data, and error breakdowns for perception versus execution failures. We will revise the manuscript to reference these results more explicitly from the abstract and introduction, and include a summary table of key metrics for immediate accessibility. revision: partial

Circularity Check

0 steps flagged

No circularity: framework description relies on external LLM capabilities without internal reductions

full rationale

The paper presents a conceptual framework consisting of an Active Spatial Brain for perception/decision-making and a Generalizable Action Cerebellum for action generation, both leveraging multi-agent large models. The provided text contains no mathematical equations, fitted parameters, derivations, or self-citations that reduce any claim to its own inputs by construction. The assertion of executable actions 'without needs of task-specific real robot data' is stated as a property of the second component but is not derived from or equivalent to any internal fit or self-referential definition. This is a standard framework paper whose central claims rest on the independent capabilities of pre-existing large models rather than any circular chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract introduces no mathematical axioms, free parameters, or new physical entities; the two named modules are architectural choices rather than postulated objects with independent evidence.

pith-pipeline@v0.9.0 · 5753 in / 1138 out tokens · 26130 ms · 2026-05-21T04:12:17.653235+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

56 extracted references · 56 canonical work pages · 6 internal anchors

  1. [1]

    arXiv preprint arXiv:2502.13013 (2025)

    Ben, Q., Jia, F., Zeng, J., Dong, J., Lin, D., Pang, J.: Homie: Hu- manoid loco-manipulation with isomorphic exoskeleton cockpit. arXiv preprint arXiv:2502.13013 (2025)

    work page arXiv 2025

  2. [2]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

    Bjorck, J., Castañeda, F., Cherniadev, N., Da, X., Ding, R., Fan, L., Fang, Y., Fox, D., Hu, F., Huang, S., et al.: Gr00t n1: An open foundation model for generalist humanoid robots. arXiv preprint arXiv:2503.14734 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    In: Advances in Neural Information Processing Systems (2020)

    Brown, T.B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Nee- lakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D.M., Wu, J., Win- ter, C., Hesse, C., Chen, M., Sigler, E., Litwin, M., Gray, S., Chess, B., Clark, J., Berner, C., McCandlish, S., Radford...

    work page 2020

  4. [4]

    AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systems

    Bu, Q., Cai, J., Chen, L., Cui, X., Ding, Y., Feng, S., Gao, S., He, X., Hu, X., Huang, X., et al.: Agibot world colosseo: A large-scale manipulation platform for scalable and intelligent embodied systems. arXiv preprint arXiv:2503.06669 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy

    Contributors, I.M.: Internvla-m1: A spatially guided vision-language-action frame- work for generalist robot policy. arXiv preprint arXiv:2510.13778 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    Conference on Robot Learning (2025)

    Dai, Y., Lee, J., et al: Aimbot: A simple auxiliary visual cue to enhance spatial awareness of visuomotor policies. Conference on Robot Learning (2025)

    work page 2025

  7. [7]

    Humanoid-vla: Towards universal humanoid control with visual integration.arXiv preprint arXiv:2502.14795,

    Ding, P., Ma, J., Tong, X., Zou, B., Luo, X., Fan, Y., Wang, T., Lu, H., Mo, P., Liu, J., et al.: Humanoid-vla: Towards universal humanoid control with visual integration. arXiv preprint arXiv:2502.14795 (2025)

    work page arXiv 2025

  8. [8]

    A Survey on Code Generation with LLM-based Agents

    Dong, Y., Jiang, X., Qian, J., Wang, T., Zhang, K., Jin, Z., Li, G.: A survey on code generation with llm-based agents. arXiv preprint arXiv:2508.00083 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    Robotics: Science and Systems (2024)

    Fang, K., Liu, F., Abbeel, P., Levine, S.: Moka: Open-world robotic manipulation through mark-based visual prompting. Robotics: Science and Systems (2024)

    work page 2024

  10. [10]

    Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation

    Fu, Z., Zhao, T.Z., Finn, C.: Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation. arXiv preprint arXiv:2401.02117 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  11. [11]

    In: Conference on Robot Learning (2024)

    He, T., Luo, Z., He, X., Xiao, W., Zhang, C., Zhang, W., Kitani, K.M., Liu, C., Shi, G.: Omnih2o: Universal and dexterous human-to-humanoid whole-body teleoperation and learning. In: Conference on Robot Learning (2024)

    work page 2024

  12. [12]

    In: IEEE International Confer- ence on Intelligent Robots and Systems (2024)

    He, T., Luo, Z., Xiao, W., Zhang, C., Kitani, K., Liu, C., Shi, G.: Learning human- to-humanoid real-time whole-body teleoperation. In: IEEE International Confer- ence on Intelligent Robots and Systems (2024)

    work page 2024

  13. [13]

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

    He, T., Wang, Z., Xue, H., Ben, Q., Luo, Z., Xiao, W., Yuan, Y., Da, X., Cas- tañeda, F., Sastry, S., et al.: Viral: Visual sim-to-real at scale for humanoid loco- manipulation. arXiv preprint arXiv:2511.15200 (2025)

    work page arXiv 2025

  14. [14]

    Huang, H., Lin, F., Hu, Y., Wang, S., Gao, Y.: Copa: General robotic manipulation throughspatialconstraintsofpartswithfoundationmodels.In:IEEEInternational Conference on Intelligent Robots and Systems (2024)

    work page 2024

  15. [15]

    In: Conference on Robot Learning (2024)

    Huang, W., Wang, C., Li, Y., Zhang, R., Fei-Fei, L.: Rekep: Spatio-temporal rea- soning of relational keypoint constraints for robotic manipulation. In: Conference on Robot Learning (2024)

    work page 2024

  16. [16]

    In: Conference on Robot Learning (2023)

    Huang, W., Wang, C., Zhang, R., Li, Y., Wu, J., Fei-Fei, L.: Voxposer: Composable 3d value maps for robotic manipulation with language models. In: Conference on Robot Learning (2023)

    work page 2023

  17. [17]

    In: Conference on Robot Learning (2022)

    Huang, W., Xia, F., Xiao, T., Chan, H., Liang, J., Florence, P., Zeng, A., Tompson, J., Mordatch, I., Chebotar, Y., Sermanet, P., Jackson, T., Brown, N., Luu, L., Levine, S., Hausman, K., Ichter, B.: Inner monologue: Embodied reasoning through planning with language models. In: Conference on Robot Learning (2022)

    work page 2022

  18. [18]

    Hyper3D: Hyper3d: Ai-powered 3d model generator (2024),https://hyper3d.ai/

    work page 2024

  19. [19]

    In: Conference on Robot Learning (2022)

    Ichter, B., Brohan, A., Chebotar, Y., Finn, C., Hausman, K., Herzog, A., Ho, D., Ibarz, J., Irpan, A., Jang, E., Julian, R., Kalashnikov, D., Levine, S., Lu, Y., Parada, C., Rao, K., Sermanet, P., Toshev, A., Vanhoucke, V., Xia, F., Xiao, T., Xu, P., Yan, M., Brown, N., Ahn, M., Cortes, O., Sievers, N., Tan, C., Xu, S., Reyes, D., Rettinghouse, J., Quiamb...

    work page 2022

  20. [20]

    IEEE Robotics and Automation Letters (2023)

    Jeon, S., Jung, M., Choi, S., Kim, B., Hwangbo, J.: Learning whole-body manip- ulation for quadrupedal robot. IEEE Robotics and Automation Letters (2023)

    work page 2023

  21. [21]

    arXiv preprint arXiv:2512.11047 , year=

    Jiang, H., Chen, J., Bu, Q., Chen, L., Shi, M., Zhang, Y., Li, D., Suo, C., Wang, C., Peng, Z., et al.: Wholebodyvla: Towards unified latent vla for whole-body loco- manipulation control. arXiv preprint arXiv:2512.11047 (2025)

    work page arXiv 2025

  22. [22]

    In: Advances in Neural Information Processing Systems (2022)

    Kojima, T., Gu, S.S., Reid, M., Matsuo, Y., Iwasawa, Y.: Large language models are zero-shot reasoners. In: Advances in Neural Information Processing Systems (2022)

    work page 2022

  23. [23]

    arXiv preprint arXiv:2505.03738 (2025)

    Li, J., Cheng, X., Huang, T., Yang, S., Qiu, R.Z., Wang, X.: Amo: Adaptive motion optimization for hyper-dexterous humanoid whole-body control. arXiv preprint arXiv:2505.03738 (2025)

    work page arXiv 2025

  24. [24]

    In: Conference on Robot Learning (2024)

    Li, J., Zhu, Y., Xie, Y., Jiang, Z., Seo, M., Pavlakos, G., Zhu, Y.: Okami: Teaching humanoid robots manipulation skills through single video imitation. In: Conference on Robot Learning (2024)

    work page 2024

  25. [25]

    In: Conference on Robot Learning (2025)

    Li, Y., Lin, Y., Cui, J., Liu, T., Liang, W., Zhu, Y., Huang, S.: Clone: Closed- loop whole-body humanoid teleoperation for long-horizon tasks. In: Conference on Robot Learning (2025)

    work page 2025

  26. [26]

    In: IEEE International Conference on Robotics and Automation (2023)

    Liang, J., Huang, W., Xia, F., Xu, P., Hausman, K., Ichter, B., Florence, P., Zeng, A.: Code as policies: Language model programs for embodied control. In: IEEE International Conference on Robotics and Automation (2023)

    work page 2023

  27. [27]

    arXiv preprint arXiv:2509.11839 (2025)

    Liu, J., Ding, P., Zhou, Q., Wu, Y., Huang, D., Peng, Z., Xiao, W., Zhang, W., Yang, L., Lu, C., et al.: Trajbooster: Boosting humanoid whole-body manipulation via trajectory-centric learning. arXiv preprint arXiv:2509.11839 (2025)

    work page arXiv 2025

  28. [28]

    arXiv preprint arXiv:2508.15874 (2025)

    Liu, Y., Liu, Y., Meng, Y., Zhang, J., Zhou, Y., Li, Y., Jiang, J., Ji, K., Ge, S., Wang, Z., et al.: Spatial policy: Guiding visuomotor robotic manipulation with spatial-aware modeling and reasoning. arXiv preprint arXiv:2508.15874 (2025)

    work page arXiv 2025

  29. [29]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Liu, Y., Long, X., Yang, Z., Liu, Y., Habermann, M., Theobalt, C., Ma, Y., Wang, W.: Easyhoi: Unleashing the power of large models for reconstructing hand-object interactions in the wild. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 7037–7047 (2025)

    work page 2025

  30. [30]

    arXiv preprint arXiv:2601.08325 (2026)

    Liu, Z., Gu, Y., Wang, Y., Xue, X., Fu, Y.: Activevla: Injecting active percep- tion into vision-language-action models for precise 3d robotic manipulation. arXiv preprint arXiv:2601.08325 (2026)

    work page arXiv 2026

  31. [31]

    In: Proceedings of the Conference on Empirical Methods in Natural Language Processing (2022)

    Madaan, A., Zhou, S., Alon, U., Yang, Y., Neubig, G.: Language models of code are few-shot commonsense learners. In: Proceedings of the Conference on Empirical Methods in Natural Language Processing (2022)

    work page 2022

  32. [32]

    arXiv preprint arXiv:2602.06643 (2026)

    Nai, R., Zheng, B., Zhao, J., Zhu, H., Dai, S., Chen, Z., Hu, Y., Hu, Y., Zhang, T., Wen, C., et al.: Humanoid manipulation interface: Humanoid whole-body manip- ulation from robot-free demonstrations. arXiv preprint arXiv:2602.06643 (2026)

    work page arXiv 2026

  33. [33]

    arXiv preprint arXiv:2506.16475 (2025)

    Niu, Y., Zhang, Y., Yu, M., Lin, C., Li, C., Wang, Y., Yang, Y., Yu, W., Zhang, T., Li, Z., et al.: Human2locoman: Learning versatile quadrupedal manipulation with human pretraining. arXiv preprint arXiv:2506.16475 (2025)

    work page arXiv 2025

  34. [34]

    In: Proceedings of the Computer Vision and Pattern Recognition Con- ference (2025)

    Pan, M., Zhang, J., Wu, T., Zhao, Y., Gao, W., Dong, H.: Omnimanip: Towards general robotic manipulation via object-centric interaction primitives as spatial constraints. In: Proceedings of the Computer Vision and Pattern Recognition Con- ference (2025)

    work page 2025

  35. [35]

    In: Advances in Neural Information Processing Systems (2025)

    Qi, Z., Zhang, W., Ding, Y., Dong, R., Yu, X., Li, J., Xu, L., Li, B., He, X., Fan, G., et al.: Sofar: Language-grounded orientation bridges spatial reasoning and object manipulation. In: Advances in Neural Information Processing Systems (2025)

    work page 2025

  36. [36]

    In: Conference on Robot Learning (2025)

    Qiu, R.Z., Yang, S., Cheng, X., Chawla, C., Li, J., He, T., Yan, G., Yoon, D.J., Hoque, R., Paulsen, L., et al.: Humanoid policy˜ human policy. In: Conference on Robot Learning (2025)

    work page 2025

  37. [37]

    arXiv preprint arXiv:2511.09555 (2025)

    Shi, H., Xie, B., Liu, Y., Yue, Y., Wang, T., Fan, H., Zhang, X., Huang, G.: Spatialactor: Exploring disentangled spatial representations for robust robotic ma- nipulation. arXiv preprint arXiv:2511.09555 (2025)

    work page arXiv 2025

  38. [38]

    In: IEEE International Conference on Robotics and Automation (2023)

    Singh, I., Blukis, V., Mousavian, A., Goyal, A., Xu, D., Tremblay, J., Fox, D., Thomason, J., Garg, A.: Progprompt: Generating situated robot task plans us- ing large language models. In: IEEE International Conference on Robotics and Automation (2023)

    work page 2023

  39. [39]

    arXiv preprint arXiv:2507.18262 (2025)

    Su,C.,Shang,W.,Qian,C.,Zhang,F.,Cong,S.:Resemact:Advancingfine-grained robotic manipulation via semantic structuring and affordance refinement. arXiv preprint arXiv:2507.18262 (2025)

    work page arXiv 2025

  40. [40]

    arXiv preprint arXiv:2507.06905 (2025)

    Sun, W., Feng, L., Cao, B., Liu, Y., Jin, Y., Xie, Z.: Ulc: A unified and fine-grained controller for humanoid loco-manipulation. arXiv preprint arXiv:2507.06905 (2025)

    work page arXiv 2025

  41. [41]

    In: 2012 IEEE/RSJ international conference on intelligent robots and systems

    Todorov, E., Erez, T., Tassa, Y.: Mujoco: A physics engine for model-based control. In: 2012 IEEE/RSJ international conference on intelligent robots and systems. pp. 5026–5033. IEEE (2012)

    work page 2012

  42. [42]

    In: IEEE International Conference on Robotics and Automation (2025)

    Wang, J., Rajabov, J., Xu, C., Zheng, Y., Wang, H.: Quadwbg: Generalizable quadrupedal whole-body grasping. In: IEEE International Conference on Robotics and Automation (2025)

    work page 2025

  43. [43]

    In: Robotics: Science and Systems (2026)

    Wei, S., Jing, H., Li, B., Zhao, Z., Mao, J., Ni, Z., He, S., Liu, J., Liu, X., Kang, K., Zang, S., Yuan, W., Pavone, M., Huang, D., Wang, Y.:ψ0: An open foundation model towards universal humanoid loco-manipulation. In: Robotics: Science and Systems (2026)

    work page 2026

  44. [44]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference (2025)

    Wu, Z., Zhou, Y., Xu, X., Wang, Z., Yan, H.: Momanipvla: Transferring vision- language-action models for general mobile manipulation. In: Proceedings of the Computer Vision and Pattern Recognition Conference (2025)

    work page 2025

  45. [45]

    Leverb: Humanoid whole-body control with latent vision-language instruction.arXiv preprint arXiv:2506.13751, 2025

    Xue, H., Huang, X., Niu, D., Liao, Q., Kragerud, T., Gravdahl, J.T., Peng, X.B., Shi, G., Darrell, T., Sreenath, K., et al.: Leverb: Humanoid whole-body control with latent vision-language instruction. arXiv preprint arXiv:2506.13751 (2025)

    work page arXiv 2025

  46. [46]

    arXiv preprint arXiv:2509.26633 (2025)

    Yang, L., Huang, X., Wu, Z., Kanazawa, A., Abbeel, P., Sferrazza, C., Liu, C.K., Duan, R., Shi, G.: Omniretarget: Interaction-preserving data generation for humanoid whole-body loco-manipulation and scene interaction. arXiv preprint arXiv:2509.26633 (2025)

    work page arXiv 2025

  47. [47]

    EgoVLA: Learning Vision-Language-Action Models from Egocentric Human Videos

    Yang, R., Yu, Q., Wu, Y., Yan, R., Li, B., Cheng, A.C., Zou, X., Fang, Y., Cheng, X., Qiu, R.Z., et al.: Egovla: Learning vision-language-action models from egocen- tric human videos. arXiv preprint arXiv:2507.12440 (2025)

    work page internal anchor Pith review arXiv 2025

  48. [48]

    In: Annual Conference on Robot Learning (2023)

    Yenamandra, S., Ramachandran, A., Yadav, K., Wang, A.S., Khanna, M., Gervet, T., Yang, T.Y., Jain, V., Clegg, A., Turner, J.M., et al.: Homerobot: Open- vocabulary mobile manipulation. In: Annual Conference on Robot Learning (2023)

    work page 2023

  49. [49]

    Being-0: A humanoid robotic agent with vision-language models and modular skills.arXiv preprint arXiv:2503.12533, 2025

    Yuan, H., Bai, Y., Fu, Y., Zhou, B., Feng, Y., Xu, X., Zhan, Y., Karlsson, B.F., Lu, Z.: Being-0: A humanoid robotic agent with vision-language models and modular skills. arXiv preprint arXiv:2503.12533 (2025)

    work page arXiv 2025

  50. [50]

    arXiv preprint arXiv:2406.10721 (2024)

    Yuan, W., Duan, J., Blukis, V., Pumacay, W., Krishna, R., Murali, A., Mousavian, A., Fox, D.: Robopoint: A vision-language model for spatial affordance prediction for robotics. arXiv preprint arXiv:2406.10721 (2024)

    work page arXiv 2024

  51. [51]

    In: Conference on Robot Learning (2025)

    Ze, Y., Chen, Z., Araujo, J.P., Cao, Z.a., Peng, X.B., Wu, J., Liu, K.: Twist: Tele- operated whole-body imitation system. In: Conference on Robot Learning (2025)

    work page 2025

  52. [52]

    In: IEEE International Conference on Intelligent Robots and Systems (2025)

    Ze, Y., Chen, Z., Wang, W., Chen, T., He, X., Yuan, Y., Peng, X.B., Wu, J.: Gener- alizable humanoid manipulation with 3d diffusion policies. In: IEEE International Conference on Intelligent Robots and Systems (2025)

    work page 2025

  53. [53]

    arXiv preprint arXiv:2505.06776 (2025)

    Zhang, Y., Yuan, Y., Gurunath, P., Gupta, I., Omidshafiei, S., Agha-mohammadi, A.a., Vazquez-Chanlatte, M., Pedersen, L., He, T., Shi, G.: Falcon: Learning force- adaptive humanoid loco-manipulation. arXiv preprint arXiv:2505.06776 (2025)

    work page arXiv 2025

  54. [54]

    IEEE Robotics and Automation Letters (2025)

    Zhang, Z., Chen, C., Xue, H., Wang, J., Liang, S., Liu, Y., Zhang, Z., Wang, H., Yi, L.: Unleashing humanoid reaching potential via real-world-ready skill space. IEEE Robotics and Automation Letters (2025)

    work page 2025

  55. [55]

    Resmimic: From general motion tracking to humanoid whole-body loco-manipulation via residual learning,

    Zhao, S., Ze, Y., Wang, Y., Liu, C.K., Abbeel, P., Shi, G., Duan, R.: Resmimic: From general motion tracking to humanoid whole-body loco-manipulation via residual learning. arXiv preprint arXiv:2510.05070 (2025)

    work page arXiv 2025

  56. [56]

    arXiv preprint arXiv:2602.15060 (2026)

    Zhu, T., Cai, G., Zhaohui, Y., Ren, G., Xie, H., Wang, Z., Wu, J., Wang, J., Yang, X., Mu, Y., et al.: Clot: Closed-loop global motion tracking for whole-body humanoid teleoperation. arXiv preprint arXiv:2602.15060 (2026)

    work page arXiv 2026