Pith. sign in

REVIEW 3 major objections 4 minor 88 references

GBC: Generalized Behavior-Cloning Framework for Whole-Body Humanoid Imitation

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read One pipeline retargets human motion capture to any humanoid, then learns whole-body imitation from it.

desk verdict Solid retargeting pipeline, but the 'any humanoid' claim hits a real injectivity problem on lower-DoF robots, and the key imitation tables are missing. read the letter →

arxiv 2508.09960 v1 pith:RX4LOY3L submitted 2025-08-13 cs.RO cs.AIcs.LG

Yifei Yao , Chengyuan Luo , Jiaheng Du , Wentao He , Jun-Guo Lu This is my paper
classification cs.ROcs.AIcs.LG
keywords humanoidroboticsimitationlearningbehaviorcloningmotionretargetingdifferentiableinversekinematicstransformerpolicyreinforcementwhole-bodycontrol
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a single, robot-agnostic pipeline can convert ordinary human motion-capture data into training data and control policies for whole-body humanoid imitation, without hand-tuning per robot. The claim is that a differentiable inverse-kinematics network can learn the mapping from human pose to robot joint configuration once and for all, and that a Transformer-based imitation-learning algorithm (DAgger-MMPPO) can then learn robust whole-body policies. If true, any new humanoid—no matter its degrees of freedom or joint layout—can be given human-like motor behavior by supplying a configuration file and an existing motion-capture dataset, removing a major bottleneck in humanoid control. The authors validate the framework on four distinct humanoid robots and report good imitation fidelity, robustness, and transfer to unseen motions.

What carries the argument

The differentiable IK network (the 'Pose Transformer') is the load-bearing object: a lightweight Transformer encoder trained with a robot-specific differentiable forward-kinematics library to map AMASS-style human poses (angle-axis, from SMPL+H) to robot joint angles, supervised by end-effector position error plus joint-limit, action-disturbance (Lipschitz), and symmetry losses. This converts any motion-capture frame into a physically feasible robot configuration in real time. On the learning side, DAgger-MMPPO—PPO augmented by DAgger-style expert intervention and a two-stage curriculum—uses the MMTransformer backbone to fuse robot state observations and reference motion observations as two

What would settle it

Retarget an AMASS motion with pronounced torso flexion onto a humanoid that has a torso pitch joint and onto one that does not, using the same trained pipeline, then compare joint-trajectory continuity and end-effector error; if the under-actuated robot shows large frame-to-frame discontinuities or much higher error, the learned mapping is not actually capturing the full space of valid motions.

Watch

Extended reading notes

Core claim

The central claim is that whole-body humanoid imitation can be made universal by treating motion retargeting as a learned regression problem, not a per-robot engineering task. On the demonstration data the human-pose-to-robot-joint mapping is injective, so a lightweight Transformer encoder can be trained to reproduce it by matching robot end-effector positions to the human reference, with joint-limit, Lipschitz, and symmetry losses. The resulting dataset feeds a two-stage PPO policy whose MMTransformer backbone treats robot observations and reference motion as separate modalities, letting one policy imitate demonstrated motions or follow velocity commands. Experiments across four distinct hu

Load-bearing premise

The retargeting step assumes every human pose in the demonstration data has exactly one robot joint configuration that a regression network can learn; robots with fewer joints than the human model collapse distinct human poses into the same robot state, making this premise false and the training targets ambiguous.

Editorial extensions

If this is right

  • New humanoid models need only a configuration file and a joint-mapping table; the rest of the data pipeline and learning algorithm transfer unchanged.
  • Large human motion-capture archives such as AMASS become directly usable as robot training data, removing a major data bottleneck for humanoid imitation.
  • A single policy can switch between reference-based imitation and reference-free command following, using the same learned weights.
  • The framework is released as an open-source extension, so other research groups can reproduce the full workflow and extend it to their own robots.
  • Sim-to-sim transfer results suggest the trained policies are deployable in realistic physics simulations, a step toward eventual real-hardware deployment.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The injectivity assumption is likely to degrade gracefully rather than catastrophically: for under-actuated robots the retargeting error should concentrate in the missing body segments, so a targeted extension would quantify retargeting accuracy as a function of kinematic degree-of-freedom mismatch and consider explicit pose-ambiguity resolution for robots with fewer joints than the human model.
  • Because the framework consumes AMASS-format data regardless of capture modality, it could in principle be coupled with text-to-motion generators to synthesize training data for novel motions without any new motion-capture recordings.
  • The MMTransformer's modality-grouping idea is a generic architectural pattern, so the same grouped-embedding approach could transfer to other reference-based control problems, such as manipulation imitation from human video.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes GBC, a generalized behavior-cloning framework for whole-body humanoid imitation. It consists of three components: a differentiable-IK retargeting pipeline that maps AMASS-style SMPL+H MoCap data to different humanoid robots; a DAgger-MMPPO algorithm with an MMTransformer backbone that trains imitation policies able to switch between reference-tracking and reference-free velocity-command control; and an Isaac Lab-based open-source platform. The authors validate shape fitting and retargeting on four humanoids (Unitree G1, H1-2, Fourier GR1, Turin V3) and report comparisons of MMTransformer against MLP baselines. As supplied, the manuscript omits the core algorithm section and several referenced tables/figures, so the central performance and generalization claims cannot be fully verified from the provided text.

Significance. If the missing results are as strong as claimed, GBC would be a genuinely useful contribution: a unified pipeline from heterogeneous human MoCap to heterogeneous humanoid policies, with open-source release, evaluation on four robots, and use of an out-of-distribution test set (HumanAct12). The retargeting component is the best-supported part: shape-fitting errors in Table III are under 2 cm, and the loss curves in Fig. 10 show no obvious overfitting. However, the significance of the DAgger-MMPPO/MMTransformer contribution and the 'truly generalized controller' claim currently rest on evidence that is not present in the supplied manuscript.

major comments (3)
  1. [§V.B and omitted tables (Tables IV, VI, VII)] The supplied manuscript does not contain Table IV, although §V.B states that the final retargeting accuracy is detailed there, and the central DAgger-MMPPO and generalization comparisons (Tables VI/VII, together with the Section IV algorithm description) are absent. These are the evidence for the paper's main claims; without them, the imitation performance, ablations, and transfer results cannot be assessed. Please include all referenced tables, figures, and the full algorithm definition in the revised version.
  2. [§III-A, Eq. (2) and footnote 1] The claim that the map f:P→A is bijective/injective on the support of the demonstration data is not justified. For H1-2 and Turin V3, Jhn < Jhm and the robots lack a torso pitch joint, so many human poses differing in torso orientation collapse to the same robot joint vector; the paper's own §V.B reports larger retargeting error on torso motions for these robots. If the downstream policy is conditioned on the reference robot state, such collapses can pair identical observations with different future reference states, making the imitation objective ambiguous. Please either prove/condition the injectivity, model the retargeting as a stochastic/multivalued map, or provide evidence that the collapse is rare and does not degrade the downstream policy.
  3. [§III-A and Table II] The text claims that the shape fitting and retargeting 'requires no robot-specific modelling,' but Table II requires a user-provided joint-correspondence table with per-joint weights, and Table II also notes that for Turin V3 an additional fixed joint and virtual link had to be added to improve foot regression. This manual configuration is precisely the per-robot engineering the paper says it avoids. Please soften the claim or quantify the amount of per-robot adaptation required.
minor comments (4)
  1. [§III-A, Eq. (4)] The action-disturbance loss uses 'sup P = inf P = 0/2π' in a way that is not rigorous for vector-valued spaces, and the Lipschitz regularization is heuristic. Please state the exact sampling distribution and clarify whether the loss is dimensionless.
  2. [§V, Action Similarity Score] Clarify the 'Action Similarity Score': if it is the output of an AMP discriminator trained on the same demonstrations used for evaluation, it is not an independent measure of human-likeness and should be labeled as a relative comparison between architectures only.
  3. [Throughout] Typos and terminology: 'data argumentation' should be 'data augmentation'; the main text says 'bijective' while the footnote says 'injective on the support'; please make this consistent.
  4. [§V, Table V] The RL/IL comparisons appear to be based on a single fixed seed. Please report mean and standard deviation over multiple seeds, or state why a single seed is sufficient for the conclusions drawn.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: retargeting is validated on out-of-distribution data and imitation claims rest on held-out tracking, not on fitted constants or self-citations.

full rationale

The GBC paper's derivation chain is self-contained rather than circular. The differentiable IK network f is trained to minimize forward-kinematics position error (Eq. 2) and its validation (Section V.B, Table IV) is performed on the out-of-distribution HumanAct12 dataset, so the retargeting accuracy claim is an external check, not a relabeled training loss. The downstream DAgger-MMPPO policies are evaluated on held-out AMASS test motions and newly converted motions, giving the generalization claim independent content. The paper's injectivity assumption (Section III-A) is an explicit modeling premise rather than a conclusion derived from its own outputs; the fact that it may fail for lower-DoF robots (H1-2, Turin V3 lacking torso pitch) is a correctness risk that would create ambiguous training targets, but it does not make the derivation circular. The only self-citations are to the project code repository (ref 35) and a prior paper (ref 80), neither of which is load-bearing for the mathematical derivation. The AMP discriminator's 'Action Similarity Score' is a learned metric trained on demonstrations and could be partially self-referential if used as both a training reward and evaluation score; however, the paper's central claims do not reduce to this metric, and external position-error and held-out tracking metrics carry the argument. No fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The retargeting contribution rests on hand-set loss weights and per-robot calibration; the learning pipeline depends on an untested injectivity assumption and a learned discriminator metric. The visible text supplies no independent evidence for the invented virtual link beyond the regression improvement.

free parameters (4)
  • Retargeting loss weights = lambda_dist=5000, lambda_limit=1000, lambda_disturb=100 (annealed to 200), lambda_sym=1000
    Hand-set in Section V-B; no sensitivity analysis shown in the provided text.
  • Shape calibration parameters (alpha, beta, delta) = Optimized per robot via Eq. (1)
    Fitted so the SMPL+H mesh matches the robot URDF in reference poses; this is legitimate calibration but adds per-morphology degrees of freedom.
  • MMTransformer hyperparameters = d_model, layers, heads (values garbled in the provided text, Table V)
    Architecture sizes chosen without visible ablation in the excerpt.
  • Data augmentation and curriculum schedules = Not detailed in the provided text
    The framework claims curriculum learning, physics-based assistance, and augmented randomization; their schedules are not visible in the reviewed fragment.
assumptions (4)
  • domain assumption The human-to-robot retargeting map is injective on the support of the demonstration data
    Section III-A states the map is bijective (footnoted to injective) despite the multi-solution nature of redundant IK; robots lacking DoFs (H1-2, Turin V3 without torso pitch) let distinct human poses map to the same robot state, making the regression target ambiguous.
  • domain assumption SMPL+H and robot URDF forward kinematics are accurate differentiable surrogates
    Equations (1) and (2) optimize through FKhm and FKhn; errors in these models propagate directly into the retargeted dataset.
  • ad hoc to paper Gaussian perturbation enforces sufficiently small Lipschitz constant for temporal smoothness
    Eq. (4) assumes bounded pose spaces (sup P = 2 pi) and a specific noise scale; the link to physical feasibility is asserted, not proven.
  • domain assumption The AMP discriminator score is a valid imitation-quality metric
    The 'Action Similarity Score' used in Section V-C comes from a learned discriminator; if it is trained on the same demonstrations as the policy, the metric is partially circular.
invented entities (1)
  • Virtual foot link for Turin V3
    purpose: Provide a fixed joint and virtual link to regress foot pitch angles (Table II)
    An internal kinematic bookkeeping device added to improve regression; no physical counterpart.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GBC: Generalized Behavior-Cloning Framework for Whole-Body Humanoid Imitation." pith.science (2026). https://pith.science/paper/RX4LOY3L

@misc{pith2026250809960,
  author       = {Pith},
  title        = {Pith review of: GBC: Generalized Behavior-Cloning Framework for Whole-Body Humanoid Imitation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RX4LOY3L}},
  note         = {Machine review of arXiv:2508.09960}
}
read the original abstract

The creation of human-like humanoid robots is hindered by a fundamental fragmentation: data processing and learning algorithms are rarely universal across different robot morphologies. This paper introduces the Generalized Behavior Cloning (GBC) framework, a comprehensive and unified solution designed to solve this end-to-end challenge. GBC establishes a complete pathway from human motion to robot action through three synergistic innovations. First, an adaptive data pipeline leverages a differentiable IK network to automatically retarget any human MoCap data to any humanoid. Building on this foundation, our novel DAgger-MMPPO algorithm with its MMTransformer architecture learns robust, high-fidelity imitation policies. To complete the ecosystem, the entire framework is delivered as an efficient, open-source platform based on Isaac Lab, empowering the community to deploy the full workflow via simple configuration scripts. We validate the power and generality of GBC by training policies on multiple heterogeneous humanoids, demonstrating excellent performance and transfer to novel motions. This work establishes the first practical and unified pathway for creating truly generalized humanoid controllers.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

88 extracted references · 41 canonical work pages

  1. [1]

    A comprehensive survey on humanoid robot development,

    S. Saeedvand, M. Jafari, H. S. Aghdasi, and J. Baltes, “A comprehensive survey on humanoid robot development,” The Knowledge Engineering Review, p. e20, 2019

    2019

  2. [2]

    A survey: dynamics of humanoid robots,

    T. Sugihara and M. Morisawa, “A survey: dynamics of humanoid robots,” Advanced Robotics, vol. 34, no. 21-22, pp. 1338–1352, 2020

    2020

  3. [3]

    Fast online trajectory optimization for the bipedal robot cassie

    T. Apgar, P. Clary, K. Green, A. Fern, and J. W. Hurst, “Fast online trajectory optimization for the bipedal robot cassie.” in Robotics: Science and Systems, vol. 101. Pittsburgh, Pennsylvania, USA, 2018, p. 14

    2018

  4. [4]

    Model predictive control: Theory and practice—a survey,

    C. E. Garcia, D. M. Prett, and M. Morari, “Model predictive control: Theory and practice—a survey,”Automatica, vol. 25, no. 3, pp. 335–348, 1989

    1989

  5. [5]

    Optimization based full body control for the atlas robot,

    S. Feng, E. Whitman, X. Xinjilefu, and C. G. Atkeson, “Optimization based full body control for the atlas robot,” in 2014 IEEE-RAS Interna- tional Conference on Humanoid Robots . IEEE, 2014, pp. 120–127

    2014

  6. [6]

    Squat motion of a humanoid robot using three-particle model predictive control and whole-body control,

    H. Chen, X. Zhang, and M. Zhao, “Squat motion of a humanoid robot using three-particle model predictive control and whole-body control,” Sensors, vol. 25, no. 2, p. 435, 2025

    2025

  7. [7]

    Optimization-based locomotion planning, estimation, and control design for the atlas humanoid robot,

    S. Kuindersma, R. Deits, M. Fallon, A. Valenzuela, H. Dai, F. Permenter, T. Koolen, P. Marion, and R. Tedrake, “Optimization-based locomotion planning, estimation, and control design for the atlas humanoid robot,” Autonomous robots, vol. 40, pp. 429–455, 2016

    2016

  8. [8]

    Walking control of humanoid robots based on improved footstep planner and whole-body coordination controller,

    X. Wang, W. Guo, S. Yin, S. Zhang, F. Zha, M. Li, P. Wang, X. Li, and L. Sun, “Walking control of humanoid robots based on improved footstep planner and whole-body coordination controller,” Frontiers in Neurorobotics, vol. 19, p. 1538979, 2025

    2025

Show all 88 references
  1. [9]

    Deep learning in robotics: Survey on model structures and training strategies,

    A. I. K ´aroly, P. Galambos, J. Kuti, and I. J. Rudas, “Deep learning in robotics: Survey on model structures and training strategies,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 51, no. 1, pp. 266–279, 2020

    2020

  2. [10]

    Humanoid-gym: Reinforcement learning for humanoid robot with zero-shot sim2real transfer,

    X. Gu, Y .-J. Wang, and J. Chen, “Humanoid-gym: Reinforcement learning for humanoid robot with zero-shot sim2real transfer,” arXiv preprint arXiv:2404.05695, 2024

    2024 arXiv

  3. [11]

    Whole-body control of humanoid robots,

    F. L. Moro and L. Sentis, “Whole-body control of humanoid robots,” Humanoid robotics: a reference , pp. 1161–1183, 2019

    2019

  4. [12]

    Imitation learning: A survey of learning methods,

    A. Hussein, M. M. Gaber, E. Elyan, and C. Jayne, “Imitation learning: A survey of learning methods,” ACM Computing Surveys (CSUR), vol. 50, no. 2, pp. 1–35, 2017

    2017

  5. [13]

    Teleoperation of humanoid robots: A survey,

    K. Darvish, L. Penco, J. Ramos, R. Cisneros, J. Pratt, E. Yoshida, S. Ivaldi, and D. Pucci, “Teleoperation of humanoid robots: A survey,” IEEE Transactions on Robotics , vol. 39, no. 3, pp. 1706–1727, 2023

    2023

  6. [14]

    Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation,

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

    2024 arXiv

  7. [15]

    Humanplus: Humanoid shadowing and imitation from humans,

    Z. Fu, Q. Zhao, Q. Wu, G. Wetzstein, and C. Finn, “Humanplus: Humanoid shadowing and imitation from humans,” in Conference on Robot Learning (CoRL) , 2024

    2024

  8. [16]

    Learn- ing human-to-humanoid real-time whole-body teleoperation,

    T. He, Z. Luo, W. Xiao, C. Zhang, K. Kitani, C. Liu, and G. Shi, “Learn- ing human-to-humanoid real-time whole-body teleoperation,” 2024

    2024

  9. [17]

    Amp: Adversarial motion priors for stylized physics-based character control,

    X. B. Peng, Z. Ma, P. Abbeel, S. Levine, and A. Kanazawa, “Amp: Adversarial motion priors for stylized physics-based character control,” ACM Transactions on Graphics (ToG) , vol. 40, no. 4, pp. 1–20, 2021

    2021

  10. [18]

    Constrained behavior cloning for robotic learning,

    W. Liang, J. Xie, Z. Wang, J. Tan, and X. Ma, “Constrained behavior cloning for robotic learning,” arXiv preprint arXiv:2408.10568 , 2024

    2024 arXiv

  11. [19]

    Omnih2o: Universal and dexterous human- to-humanoid whole-body teleoperation and learning,

    T. He, Z. Luo, X. He, W. Xiao, C. Zhang, W. Zhang, K. Kitani, C. Liu, and G. Shi, “Omnih2o: Universal and dexterous human- to-humanoid whole-body teleoperation and learning,” arXiv preprint arXiv:2406.08858, 2024

    2024 arXiv

  12. [20]

    Openvla: An open- source vision-language-action model,

    M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, R. Rafailov, E. Foster, G. Lam, P. Sanketi et al. , “Openvla: An open- source vision-language-action model,” arXiv preprint arXiv:2406.09246, 2024

    2024 arXiv

  13. [21]

    Humanvla: Towards vision-language directed object rearrangement by physical humanoid,

    X. Xu, Y . Zhang, Y .-L. Li, L. Han, and C. Lu, “Humanvla: Towards vision-language directed object rearrangement by physical humanoid,” arXiv preprint arXiv:2406.19972 , 2024

    2024 arXiv

  14. [22]

    A comprehensive review of humanoid robots,

    Q. Sheng, Z. Zhou, J. Li, X. Mi, P. Xiang, Z. Chen, H. Xu, S. Jia, X. Wu, Y . Cui et al. , “A comprehensive review of humanoid robots,” SmartBot, vol. 1, no. 1, p. e12008, 2025

    2025

  15. [23]

    Interaction mesh based motion adaptation for biped humanoid robots,

    S. Nakaoka and T. Komura, “Interaction mesh based motion adaptation for biped humanoid robots,” in 2012 12th IEEE-RAS International Conference on Humanoid Robots (Humanoids 2012) , 2012, pp. 625– 631

    2012

  16. [24]

    Whole-body geometric retargeting for humanoid robots,

    K. Darvish, Y . Tirupachuri, G. Romualdi, L. Rapetti, D. Ferigo, F. J. A. Chavez, and D. Pucci, “Whole-body geometric retargeting for humanoid robots,” in 2019 IEEE-RAS 19th International Conference on Humanoid Robots (Humanoids). IEEE, 2019, pp. 679–686

    2019

  17. [25]

    Human motion retargeting to a full-scale humanoid robot using a monocular camera and human pose estimation,

    S. Baek, A. Kim, J.-Y . Choi, E. Ha, and J.-W. Kim, “Human motion retargeting to a full-scale humanoid robot using a monocular camera and human pose estimation,” International Journal of Control, Automation and Systems, vol. 22, no. 9, pp. 2860–2870, 2024

    2024

  18. [26]

    Self-supervised motion retar- geting with safety guarantee,

    S. Choi, M. J. Song, H. Ahn, and J. Kim, “Self-supervised motion retar- geting with safety guarantee,” in 2021 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2021, pp. 8097–8103

    2021

  19. [27]

    Imitationnet: Unsupervised human- to-robot motion retargeting via shared latent space,

    Y . Yan, E. V . Mascaro, and D. Lee, “Imitationnet: Unsupervised human- to-robot motion retargeting via shared latent space,” in 2023 IEEE- RAS 22nd International Conference on Humanoid Robots (Humanoids) . IEEE, 2023, pp. 1–8

    2023

  20. [28]

    Unsupervised neural motion retargeting for humanoid teleoperation,

    S. Yagi, M. Tada, E. Uchibe, S. Kanoga, T. Matsubara, and J. Morimoto, “Unsupervised neural motion retargeting for humanoid teleoperation,” arXiv preprint arXiv:2406.00727 , 2024

    2024 arXiv

  21. [29]

    AMASS: Archive of motion capture as surface shapes,

    N. Mahmood, N. Ghorbani, N. F. Troje, G. Pons-Moll, and M. J. Black, “AMASS: Archive of motion capture as surface shapes,” inInternational Conference on Computer Vision , Oct. 2019, pp. 5442–5451

    2019

  22. [30]

    Motion-x: A large-scale 3d expressive whole-body human motion dataset,

    J. Lin, A. Zeng, S. Lu, Y . Cai, R. Zhang, H. Wang, and L. Zhang, “Motion-x: A large-scale 3d expressive whole-body human motion dataset,” Advances in Neural Information Processing Systems , vol. 36, pp. 25 268–25 280, 2023

    2023

  23. [31]

    A reduction of imitation learning and structured prediction to no-regret online learning,

    S. Ross, G. Gordon, and D. Bagnell, “A reduction of imitation learning and structured prediction to no-regret online learning,” in Proceedings of the fourteenth international conference on artificial intelligence and statistics. JMLR Workshop and Conference Proceedings, 2011, ...

    2011

  24. [32]

    Generative adversarial imitation learning,

    J. Ho and S. Ermon, “Generative adversarial imitation learning,” Ad- vances in neural information processing systems , vol. 29, 2016

    2016

  25. [33]

    Twist: Teleoperated whole-body imitation system,

    Y . Ze, Z. Chen, J. P. Ara ˜Aˇsjo, Z.-a. Cao, X. B. Peng, J. Wu, and C. K. Liu, “Twist: Teleoperated whole-body imitation system,” arXiv preprint arXiv:2505.02833, 2025

    2025 arXiv

  26. [34]

    Curriculum learning,

    Y . Bengio, J. Louradour, R. Collobert, and J. Weston, “Curriculum learning,” in Proceedings of the 26th annual international conference on machine learning , 2009, pp. 41–48

    2009

  27. [35]

    GBC: Generalized Behavior-Cloning Frame- work - Official Code and Documentation,

    Y . Yao and C. Luo, “GBC: Generalized Behavior-Cloning Frame- work - Official Code and Documentation,” https://github.com/ sjtu-mvasl-robotics/GBC, 2025, documentation available at: https:// sjtu-mvasl-robotics.github.io/gbc/

    2025

  28. [36]

    Motion retargeting for humanoid robots based on identification to preserve and reproduce human motion features,

    K. Ayusawa, M. Morisawa, and E. Yoshida, “Motion retargeting for humanoid robots based on identification to preserve and reproduce human motion features,” in 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2015, pp. 2774–2779

    2015

  29. [37]

    Real-time whole-body imitation by humanoid robots and task-oriented teleoperation using an analytical mapping method and quantitative evaluation,

    Z. Zhang, Y . Niu, Z. Yan, and S. Lin, “Real-time whole-body imitation by humanoid robots and task-oriented teleoperation using an analytical mapping method and quantitative evaluation,” Applied Sciences, vol. 8, no. 10, p. 2005, 2018

    2005

  30. [38]

    Robust real-time whole- body motion retargeting from human to humanoid,

    L. Penco, B. Cl ´ement, V . Modugno, E. M. Hoffman, G. Nava, D. Pucci, N. G. Tsagarakis, J.-B. Mouret, and S. Ivaldi, “Robust real-time whole- body motion retargeting from human to humanoid,” in 2018 IEEE- RAS 18th International Conference on Humanoid Robots (Humanoids) . IEEE...

    2018

  31. [39]

    Un- paired motion style transfer from video to animation,

    K. Aberman, Y . Weng, D. Lischinski, D. Cohen-Or, and B. Chen, “Un- paired motion style transfer from video to animation,”ACM Transactions On Graphics (TOG) , vol. 39, no. 4, pp. 64–1, 2020

    2020

  32. [40]

    Kino- dynamically consistent motion retargeting for humanoids,

    G. Bin Hammam, P. M. Wensing, B. Dariush, and D. E. Orin, “Kino- dynamically consistent motion retargeting for humanoids,” International Journal of Humanoid Robotics , vol. 12, no. 04, p. 1550017, 2015

    2015

  33. [41]

    Multi-contact motion retargeting from human to humanoid robot,

    A. Di Fava, K. Bouyarmane, K. Chappellet, E. Ruffaldi, and A. Kheddar, “Multi-contact motion retargeting from human to humanoid robot,” in 2016 IEEE-RAS 16th international conference on humanoid robots (humanoids). IEEE, 2016, pp. 1081–1086

    2016

  34. [42]

    Dynamic movement primitive based motion retargeting for dual-arm sign language motions,

    Y . Liang, W. Li, Y . Wang, R. Xiong, Y . Mao, and J. Zhang, “Dynamic movement primitive based motion retargeting for dual-arm sign language motions,” in 2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 8195–8201

    2021

  35. [43]

    Nonparametric motion retargeting for humanoid robots on shared latent space

    S. Choi, M. K. Pan, and J. Kim, “Nonparametric motion retargeting for humanoid robots on shared latent space.” in Robotics: science and systems, 2020. 21

    2020

  36. [44]

    Human-robot motion retargeting via neural latent optimization,

    H. Zhang, W. Li, Y . Liang, Z. Chen, Y . Cui, Y . Wang, and R. Xiong, “Human-robot motion retargeting via neural latent optimization,” CoRR, 2021

    2021

  37. [45]

    Homie: Humanoid loco-manipulation with isomorphic exoskeleton cockpit,

    Q. Ben, F. Jia, J. Zeng, J. Dong, D. Lin, and J. Pang, “Homie: Humanoid loco-manipulation with isomorphic exoskeleton cockpit,” arXiv preprint arXiv:2502.13013, 2025

    2025 arXiv

  38. [46]

    Generating diverse and natural 3d human motions from text,

    C. Guo, S. Zou, X. Zuo, S. Wang, W. Ji, X. Li, and L. Cheng, “Generating diverse and natural 3d human motions from text,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 5152–5161

    2022

  39. [47]

    Learning smooth humanoid locomotion through lipschitz-constrained policies,

    Z. Chen, X. He, Y .-J. Wang, Q. Liao, Y . Ze, Z. Li, S. S. Sastry, J. Wu, K. Sreenath, S. Gupta et al. , “Learning smooth humanoid locomotion through lipschitz-constrained policies,” arXiv preprint arXiv:2410.11825, 2024

    2024 arXiv

  40. [48]

    Real-world humanoid locomotion with reinforcement learning,

    I. Radosavovic, T. Xiao, B. Zhang, T. Darrell, J. Malik, and K. Sreenath, “Real-world humanoid locomotion with reinforcement learning,” Sci- ence Robotics, vol. 9, no. 89, p. eadi9579, 2024

    2024

  41. [49]

    Wococo: Learning whole- body humanoid control with sequential contacts,

    C. Zhang, W. Xiao, T. He, and G. Shi, “Wococo: Learning whole- body humanoid control with sequential contacts,” arXiv preprint arXiv:2406.06005, 2024

    2024 arXiv

  42. [50]

    Ex- pressive whole-body control for humanoid robots,

    X. Cheng, Y . Ji, J. Chen, R. Yang, G. Yang, and X. Wang, “Ex- pressive whole-body control for humanoid robots,” arXiv preprint arXiv:2402.16796, 2024

    2024 arXiv

  43. [51]

    Behavior transformers: Cloning � modes with one stone,

    N. M. Shafiullah, Z. Cui, A. A. Altanzaya, and L. Pinto, “Behavior transformers: Cloning � modes with one stone,” Advances in neural information processing systems , vol. 35, pp. 22 955–22 968, 2022

    2022

  44. [52]

    Behavior generation with latent actions,

    S. Lee, Y . Wang, H. Etukuru, H. J. Kim, N. M. M. Shafiullah, and L. Pinto, “Behavior generation with latent actions,” arXiv preprint arXiv:2403.03181, 2024

    2024 arXiv

  45. [53]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

    2017

  46. [54]

    Deepmimic: Example-guided deep reinforcement learning of physics-based character skills,

    X. B. Peng, P. Abbeel, S. Levine, and M. van de Panne, “Deepmimic: Example-guided deep reinforcement learning of physics-based character skills,” ACM Trans. Graph. , vol. 37, no. 4, pp. 143:1–143:14, Jul

  47. [55]

    Prox- imal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Prox- imal policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

  48. [56]

    Domain randomization for transferring deep neural networks from simulation to the real world,

    J. Tobin, R. Fong, A. Ray, J. Schneider, W. Zaremba, and P. Abbeel, “Domain randomization for transferring deep neural networks from simulation to the real world,” in 2017 IEEE/RSJ international conference on intelligent robots and systems (IROS) . IEEE, 2017, pp. 23–30

    2017

  49. [57]

    Sim-to-real transfer of robotic control with dynamics randomization,

    X. B. Peng, M. Andrychowicz, W. Zaremba, and P. Abbeel, “Sim-to-real transfer of robotic control with dynamics randomization,” in 2018 IEEE international conference on robotics and automation (ICRA) . IEEE, 2018, pp. 3803–3810

    2018

  50. [58]

    Using simulation and domain adaptation to improve efficiency of deep robotic grasping,

    K. Bousmalis, A. Irpan, P. Wohlhart, Y . Bai, M. Kelcey, M. Kalakrish- nan, L. Downs, J. Ibarz, P. Pastor, K. Konolige et al., “Using simulation and domain adaptation to improve efficiency of deep robotic grasping,” in 2018 IEEE international conference on robotics and automat...

    2018

  51. [59]

    How to pick the domain randomization parameters for sim-to-real transfer of reinforcement learning policies?

    Q. Vuong, S. Vikram, H. Su, S. Gao, and H. I. Christensen, “How to pick the domain randomization parameters for sim-to-real transfer of reinforcement learning policies?” arXiv preprint arXiv:1903.11774 , 2019

    1903 arXiv

  52. [60]

    Self- paced curriculum learning,

    L. Jiang, D. Meng, Q. Zhao, S. Shan, and A. Hauptmann, “Self- paced curriculum learning,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 29, no. 1, 2015

    2015

  53. [61]

    Reverse curriculum generation for reinforcement learning,

    C. Florensa, D. Held, M. Wulfmeier, M. Zhang, and P. Abbeel, “Reverse curriculum generation for reinforcement learning,” in Conference on robot learning. PMLR, 2017, pp. 482–495

    2017

  54. [62]

    Learning symmetric and low-energy locomotion,

    W. Yu, G. Turk, and C. K. Liu, “Learning symmetric and low-energy locomotion,” ACM Transactions on Graphics (TOG) , vol. 37, no. 4, pp. 1–12, 2018

    2018

  55. [63]

    On learning symmetric locomotion,

    F. Abdolhosseini, H. Y . Ling, Z. Xie, X. B. Peng, and M. Van de Panne, “On learning symmetric locomotion,” in Proceedings of the 12th ACM SIGGRAPH Conference on Motion, Interaction and Games , 2019, pp. 1–10

    2019

  56. [64]

    LAFAN1 Retargeting Dataset,

    H. Lv, “LAFAN1 Retargeting Dataset,” https://huggingface.co/datasets/ lvhaidong/LAFAN1 Retargeting Dataset, 2025

    2025

  57. [65]

    Robust motion in-betweening,

    F. G. Harvey, M. Yurick, D. Nowrouzezahrai, and C. Pal, “Robust motion in-betweening,” vol. 39, no. 4, 2020

    2020

  58. [66]

    SMPL: A skinned multi-person linear model,

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black, “SMPL: A skinned multi-person linear model,” ACM Trans. Graphics (Proc. SIGGRAPH Asia) , vol. 34, no. 6, pp. 248:1–248:16, Oct. 2015

    2015

  59. [67]

    Embodied hands: Modeling and capturing hands and bodies together,

    J. Romero, D. Tzionas, and M. J. Black, “Embodied hands: Modeling and capturing hands and bodies together,” ACM Transactions on Graph- ics, (Proc. SIGGRAPH Asia) , vol. 36, no. 6, Nov. 2017

    2017

  60. [68]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

    2010 arXiv

  61. [69]

    Llama: Open and efficient foundation language models,

    H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozi `ere, N. Goyal, E. Hambro, F. Azhar et al. , “Llama: Open and efficient foundation language models,” arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  62. [70]

    Recurrent model-free rl can be a strong baseline for many pomdps,

    T. Ni, B. Eysenbach, and R. Salakhutdinov, “Recurrent model-free rl can be a strong baseline for many pomdps,” arXiv preprint arXiv:2110.05038, 2021

    2021 arXiv

  63. [71]

    Decision transformer: Reinforcement learning via sequence modeling,

    L. Chen, K. Lu, A. Rajeswaran, K. Lee, A. Grover, M. Laskin, P. Abbeel, A. Srinivas, and I. Mordatch, “Decision transformer: Reinforcement learning via sequence modeling,” Advances in neural information pro- cessing systems, vol. 34, pp. 15 084–15 097, 2021

    2021

  64. [72]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision . Springer, 2020, pp. 213– 229

    2020

  65. [73]

    Vilt: Vision-and-language transformer without convolution or region supervision,

    W. Kim, B. Son, and I. Kim, “Vilt: Vision-and-language transformer without convolution or region supervision,” in International conference on machine learning . PMLR, 2021, pp. 5583–5594

    2021

  66. [74]

    Bert: Pre-training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” in Pro- ceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technolo...

    2019

  67. [75]

    Root mean square layer normalization,

    B. Zhang and R. Sennrich, “Root mean square layer normalization,” Advances in Neural Information Processing Systems , vol. 32, 2019

    2019

  68. [76]

    Distilling the knowledge in a neural network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” arXiv preprint arXiv:1503.02531 , 2015

    2015 arXiv

  69. [77]

    Policy distillation,

    A. A. Rusu, S. G. Colmenarejo, C. Gulcehre, G. Desjardins, J. Kirk- patrick, R. Pascanu, V . Mnih, K. Kavukcuoglu, and R. Hadsell, “Policy distillation,” arXiv preprint arXiv:1511.06295 , 2015

    2015 arXiv

  70. [78]

    Distilling task-specific knowledge from bert into simple neural networks,

    R. Tang, Y . Lu, L. Liu, L. Mou, O. Vechtomova, and J. Lin, “Distilling task-specific knowledge from bert into simple neural networks,” arXiv preprint arXiv:1903.12136, 2019

    1903 arXiv

  71. [79]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chen et al., “Lora: Low-rank adaptation of large language models.” ICLR, vol. 1, no. 2, p. 3, 2022

    2022

  72. [80]

    Anybipe: An end-to-end framework for training and deploying bipedal robots guided by large language models,

    Y . Yao, W. He, C. Gu, J. Du, F. Tan, Z. Zhu, and J. Lu, “Anybipe: An end-to-end framework for training and deploying bipedal robots guided by large language models,” 2024. [Online]. Available: https://arxiv.org/abs/2409.08904

    2024 arXiv

  73. [81]

    Better-than-demonstrator imi- tation learning via automatically-ranked demonstrations,

    D. S. Brown, W. Goo, and S. Niekum, “Better-than-demonstrator imi- tation learning via automatically-ranked demonstrations,” in Conference on robot learning . PMLR, 2020, pp. 330–359

    2020

  74. [82]

    Isaac Sim

    NVIDIA, “Isaac Sim.” [Online]. Available: https://github.com/isaac-sim/ IsaacSim

  75. [83]

    Mujoco: A physics engine for model- based control,

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

    2012

  76. [84]

    Action2motion: Conditioned generation of 3d human motions,

    C. Guo, X. Zuo, S. Wang, S. Zou, Q. Sun, and A. Deng, “Action2motion: Conditioned generation of 3d human motions,” in Proceedings of the 28th ACM International Conference on Multimedia , 2020, pp. 2420– 2428

    2020

  77. [85]

    Wham: Reconstructing world-grounded humans with accurate 3d motion,

    S. Shin, J. Kim, E. Halilaj, and M. J. Black, “Wham: Reconstructing world-grounded humans with accurate 3d motion,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 2070–2080

    2024

  78. [86]

    Towards unifying human likeness: Evaluating metrics for human-like motion retargeting on bimanual manipulation tasks,

    A. Meixner, M. Carl, F. Krebs, N. Jaquier, and T. Asfour, “Towards unifying human likeness: Evaluating metrics for human-like motion retargeting on bimanual manipulation tasks,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 13 015–13 022

    2024

  79. [87]

    Humanoid- bench: Simulated humanoid benchmark for whole-body locomotion and manipulation,

    C. Sferrazza, D.-M. Huang, X. Lin, Y . Lee, and P. Abbeel, “Humanoid- bench: Simulated humanoid benchmark for whole-body locomotion and manipulation,” arXiv preprint arXiv:2403.10506 , 2024

    2024 arXiv

  80. [2018]

    Available: http://doi.acm.org/10.1145/3197517.3201311

    [Online]. Available: http://doi.acm.org/10.1145/3197517.3201311

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.