Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

The paper claims that the intrinsic difficulty of imitating a motion can be measured, independently of any policy, by the torque variation induced by small pose perturbations — larger torque swings flatten the reward landscape and make a mo

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A physics-based score (MDS) predicts how hard a motion is for a humanoid to imitate by measuring how much joint torques must change under small pose perturbations.

T0 review reviewed 2026-08-03 challenge →

load-bearing objection A genuinely new difficulty metric with a real contact-model gap and undisclosed fitting details; worth refereeing but needs hard revision. the 4 major comments →

arxiv 2512.07248 v2 pith:GOSXIPGJ submitted 2025-12-08 cs.GR

Distinguishing Imitation Error from Intrinsic Motion Learning Difficulty

Zhaorui Meng , Lu Yin , Xinrui Chen , Chengxu Zuo , Anjun Chen , Shihui Guo , Yipeng Qin This is my paper
classification cs.GR
keywords motion imitationreinforcement learningrigid-body dynamicsmotion difficultytorque variationhumanoid controlreward landscapecurriculum learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 argues that how hard it is to imitate a motion is not just a property of the learning algorithm — it is a measurable property of the motion itself. The proposed Motion Difficulty Score (MDS) measures the torque variation triggered by small pose perturbations: motions that demand wildly different torques for nearly identical poses give reinforcement learning a flat reward landscape, so they are intrinsically hard to imitate. The paper validates MDS on large-scale motion data with two state-of-the-art imitation policies, showing that MDS strongly correlates with imitation error. If this holds, evaluation of humanoid control can separate policy failures from motion-inherent challenges, and motion datasets can be curated and stratified by a physics-grounded difficulty label.

Core claim

The central claim is that imitation difficulty can be defined independently of any policy as the magnitude of torque variation induced within a bounded pose-error neighborhood, per Definition 4.1. High torque-to-pose variation collapses the reward landscape into a sharp spike surrounded by a plateau, so gradient-based reinforcement learning receives almost no directional signal. MDS operationalizes this via three complementary terms — spectral diversity (log-volume of the feasible torque space, derived through the coarea formula), variance diversity (per-joint variation of the Jacobian), and segment diversity (temporal uniformity of spectral diversity) — aggregated into a single scalar. Expe

What carries the argument

The central object is the Motion Difficulty Score (MDS), a policy-free scalar computed from the rigid-body inverse-dynamics map τ = M(q)q̈ + h(q,q̇) − f_ext. The paper perturbs each frame's state inside a small ball, maps the perturbed neighbourhood through the dynamics to a torque set, and characterizes that set by its volume (via singular values of the Jacobian), its joint-wise variance, and its temporal variability across four segments. The score is the weighted sum of these three diversities; the weights are empirically derived.

Load-bearing premise

The score is computed from torques obtained via Eq. 12, but for a floating-base humanoid the inverse-dynamics problem requires a ground-contact model that the paper never specifies; if mocap motions are not dynamically consistent, the torque magnitudes — and therefore the score — could depend on arbitrary modeling choices. Also, the aggregation weights in Eq. 9 are empirically derived with no validation split, so the reported correlations could be partly in-sample.

What would settle it

Compute MDS for the same motion set under two different contact models (e.g., full ground-reaction-force optimization versus simple residual-force compensation) and check whether the difficulty rankings, not just the scale, change; if the rankings scramble, the 'intrinsic' claim fails. Alternatively, hold out 20% of motion clips, fit the MDS aggregate weights on the rest, and test whether the correlation with imitation error survives on the held-out clips.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • MDS gives a quantitative answer to 'why did the policy fail?' — high error on low-MDS motions points at the policy, while high error on high-MDS motions reflects an intrinsic ceiling.
  • MID (Maximum Imitable Difficulty) locates the difficulty threshold beyond which a policy's error explodes, turning a scatter plot into a single robustness boundary.
  • DSJE (Difficulty-Stratified Joint Error) exposes difficulty-regime reversals invisible in aggregate error, such as a policy that beats another overall but loses on easy motions.
  • MDS-based curriculum learning improves final policy performance across all difficulty groups in the paper's experiments.
  • MDS flags corrupted or physically implausible motion sequences by assigning them unusually high difficulty, enabling automated mocap quality control.

Where Pith is reading between the lines

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

  • The same principle could drive reward shaping or exploration heuristics: agents could be biased toward low-MDS regions early and toward high-MDS regions once the plateau is navigable — an extension the paper's curriculum experiment only begins to explore.
  • Because MDS depends on the chosen mass distribution and contact handling, the same motion could receive different scores on different morphologies; the paper's retargeting experiment suggests the ordering is stable, but the score's invariance is not proven.
  • The reported correlations are computed on training data; a stronger test would freeze the aggregate weights and measure correlation on a held-out set of motion clips, which the paper does not report.
  • If MDS truly captures landscape flatness, it should also predict sample efficiency rather than only final error — a testable extension that would connect the metric to learning-curve dynamics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Motion Difficulty Score (MDS), a policy-independent metric that measures a motion's intrinsic imitation difficulty through the torque variation induced by small pose perturbations. MDS is computed as a weighted combination of three terms: Spectral Diversity, Variance Diversity, and Segment Diversity, and is used to build MD-AMASS, a difficulty-labeled repartitioning of AMASS. The authors validate MDS by showing correlations with imitation error of UHC and PHC+, and introduce two derived metrics: Maximum Imitable Difficulty (MID) and Difficulty-Stratified Joint Error (DSJE). They also report applications to curriculum learning and flawed-motion detection.

Significance. If validated, MDS would be a practically valuable tool: it offers a policy-free scalar for comparing motions across training conditions, enables difficulty-stratified evaluation, and could support curriculum learning and mocap curation. The paper provides a clear dynamical intuition, an explicit computational pipeline, and a large-scale empirical study (MD-AMASS, >30,000 clips). The correlation analysis, ablations, and cross-robot generalization study are useful steps. However, the central claim rests on several under-specified and potentially circular empirical choices; these need to be resolved before the metric can be trusted as a predictive, architecture-independent difficulty measure.

major comments (4)
  1. [§4.1, Eq. (12), Appendix A] The inverse-dynamics definition of torque is underdetermined for a floating-base humanoid. Eq. (12) defines τ = M(q)q̈ + h(q,q̇) − f_ext as a unique per-frame torque, but for a floating base the root degrees of freedom are not actuated and contact/ground-reaction forces are unknown. The paper neither specifies a contact model nor states how f_ext is computed. Since UHC applies residual root forces while PHC+ does not, the torque space being measured may not match the policy's action space. This makes MDS a function of an unstated modeling choice rather than an intrinsic motion property. The authors should specify the contact model, clarify which DoFs are considered actuated, and show MDS is robust to these choices.
  2. [§4.2.4, Eq. (9); §5.1] The aggregation weights w_i in Eq. (9) are described only as 'empirically derived', with no fitting procedure, no reported values, and no validation split. The claimed correlations in Table 1 may therefore be in-sample fits rather than predictions. The authors should state the weight-selection method, report the weights, and evaluate MDS on a held-out set of motions or with cross-validation. The perturbation radius ε in Eq. (14) is also never stated; its value directly affects all three diversity terms and should be reported and varied in sensitivity analysis.
  3. [§6.1, Eq. (10); Fig. 4] The MID definition in Eq. (10) selects the threshold c that maximizes the error gap on the very same policy data that MID is then used to characterize. This is circular: the 'onset of performance collapse' is a fitted maximum, not an independent boundary. The resulting MID values (UHC: 308.22, PHC+: 320.50) are unsurprisingly ordered and provide no statistical evidence of a real capability difference. The authors should evaluate MID on held-out motion clips or use a pre-defined threshold selection rule (e.g., cross-validated or based only on MDS, not on error).
  4. [§5.1, Fig. 4; Table 1] The validation is correlational and uses samples drawn from the policies' training sets, so it does not establish that MDS predicts generalization to unseen motions. The exclusion of 'extreme outliers' (MPJPE-G>250 and MDS>350, fewer than 100 clips) is post hoc and its effect on the reported correlations is not quantified. The authors should report correlations with and without exclusions, and ideally evaluate on a held-out subset of AMASS that was not used in policy training.
minor comments (5)
  1. [Throughout] Minor typos: 'Maximun' in Fig. 1, 'Temprol' in Fig. 2, 'mimicing' in §3, 'polices' in §4.2.4.
  2. [§4.3] The reference to AMASS appears as 'AMASS dataset []'—the citation placeholder needs to be filled.
  3. [§4.2 / Appendix A] Notation is inconsistent: Eq. (1) uses τ ∈ R^N, while Definition 4.1 and Appendix A use Y = R^{J×t}; the state space X = R^{N×3×t} is also hard to reconcile with per-frame states s_i ∈ R^{3N}. Please clarify dimensions.
  4. [§4.2.3] K=4 is chosen empirically with no sensitivity analysis. Since Segment Diversity depends on K, the authors should report how MDS and the correlations change for other K.
  5. [Appendix C.1] The curriculum learning results in Table 4 are presented in the appendix but not discussed in the main text; the authors should either move this result to the main text or clarify its status.

Circularity Check

2 steps flagged

Partial circularity: MDS validation uses empirically derived weights on the same training-set clips, and MID is defined as the argmax over the error data it is claimed to reveal.

specific steps
  1. fitted input called prediction [Sec. 4.2.4 (Eq. 9); Sec. 5.1 (Fig. 4 caption, Table 1)]
    "The final Motion Difficulty Score (MDS) is computed as the weighted sum of the three diversity terms, the weight assignments are empirically derived: MDS=Σ_{i=1}^3 w_i d_i (9). ... we visualize over 3,000 randomly sampled sequences ... with samples drawn from the policies’ training set."

    The only free parameters of MDS, the three weights w_i, are said to be 'empirically derived,' but the validation then correlates exactly this MDS with imitation error on the same clips (Fig. 4, Table 1) and calls the result 'predicts higher tracking error.' No held-out split, cross-validation, or weight values are reported. If the weights were chosen to improve agreement on these clips, the Pearson/Spearman/Kendall correlations are in-sample fits, not independent predictions: part of the 'explained' error is built into MDS by the weight selection. The ablation in Sec. 5.2 likewise selects the full aggregation because it 'best aligns with observed imitation outcomes' on that same data.

  2. self definitional [Sec. 6.1, Eq. 10]
    "To quantify this transition, we exhaustively evaluate all possible MDS split points ... The split that maximizes inter-group error disparity defines the critical difficulty level at which large errors begin to dominate. We term this threshold the Maximum Imitable Difficulty (MID) ... MID = arg max_{c∈C} [μ_high(c)−μ_low(c)] (10)"

    MID is defined as the threshold that maximizes the difference between the mean errors of the low- and high-difficulty partitions of the same data set. Any set of error values has such an argmax, so the 'onset of performance collapse' and the claim 'This quantitatively confirms PHC+'s superior imitation capacity, with UHC exhibiting earlier onset of outliers' (Sec. 6.1) follow from the definition rather than being discovered empirically. The 'Maximum Imitable Difficulty' is therefore a descriptive statistic of the error data used to construct it, not an independently measured capability boundary.

full rationale

The core MDS construction is not fully circular: MDS is computed from rigid-body dynamics and perturbation Jacobians independently of any policy output, and the three diversity terms have independent physical interpretations. The main circularity concerns are (1) the three aggregation weights in Eq. 9 are 'empirically derived' with no held-out split, while the paper's central validation is an in-sample correlation of this fitted MDS with imitation error on the policies' training-set clips, so part of the claimed predictive power may be a fit; and (2) MID is defined as the argmax over the same error data it is then used to 'confirm,' making the identified breakdown threshold tautological. No load-bearing self-citation chain is present: references to the authors' own prior work appear mainly in Appendix C for inertial-mocap datasets and are not needed to justify the MDS derivation. The floating-base contact-model gap raised in the skeptic analysis is a correctness/reproducibility risk rather than a circularity, because the paper does not define f_ext or the actuated root DoFs, but this does not by itself make the derivation equal to its inputs. Overall, partial circularity: the central metric retains independent content, but two of the paper's headline validation/application claims reduce in part to construction.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

MDS rests on three hand-built diversity terms whose aggregation weights are undisclosed, a perturbation radius that is never stated, and an inverse-dynamics model with unspecified contact handling. None of these choices are validated against independent data; they are the true price of the paper.

free parameters (4)
  • MDS weights w_1, w_2, w_3 = undisclosed
    Eq. 9; 'the weight assignments are empirically derived' — no values, no fitting procedure, no validation split. The reported correlations in Table 1 depend on these weights.
  • Segment count K = 4
    §4.2.3: 'Empirically, we set K=4 in our experiments' — chosen by hand, not justified.
  • Perturbation radius ε = not stated
    The neighborhood N(S) of perturbed states in §4.1/Appendix A is defined as a product of ε-balls; the value of ε is never given, yet the entire MDS value and all downstream thresholds depend on it.
  • Outlier exclusion rule = MPJPE-G>250 and MDS>350
    §5.1: fewer than 100 clips excluded post hoc to 'maintain visual clarity'; exclusion criteria chosen after seeing data, potentially inflating correlations.
axioms (3)
  • domain assumption The difficulty of imitation is fully determined by the local sensitivity of the inverse-dynamics torque map F to pose perturbations.
    §4.1: the paper equates motion difficulty with torque variation under perturbation; this ignores other factors such as exploration noise, policy parameterization, reward shaping, and contact-rich dynamics.
  • domain assumption A unique torque vector can be computed for each mocap frame via inverse dynamics without a fully specified contact/ground-reaction model.
    §3 Eq. 1 and §4.3: floating-base humanoid inverse dynamics is underdetermined without contact forces; RBDL is used but the contact handling is never described, and mocap motions are generally not dynamically consistent.
  • standard math The coarea-formula approximation (Eq. 18–22) is valid, i.e., each per-frame map F_i is a submersion at the reference state and ε is sufficiently small.
    Appendix A.2: the proof assumes surjectivity of dF_i and small ε; near kinematic singularities or with zero singular values the Gram determinant vanishes and the log-volume formula breaks down.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Distinguishing Imitation Error from Intrinsic Motion Learning Difficulty." pith.science (2026). https://pith.science/paper/GOSXIPGJ

@misc{pith2026251207248,
  author       = {Pith},
  title        = {Pith review of: Distinguishing Imitation Error from Intrinsic Motion Learning Difficulty},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GOSXIPGJ}},
  note         = {Machine review of arXiv:2512.07248}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Physics-based motion imitation is central to humanoid control, yet current evaluation metrics (e.g., MPJPE) only quantify imitation outcomes, not their underlying causes. This conflation obscures a critical diagnostic question: when imitation error occurs, does it stem from policy limitations or the intrinsic learning difficulty of the target motion? To resolve this ambiguity, we propose the Torque Variation Score (TVS), a physics-grounded metric that quantifies the inherent learning difficulty of a motion independently of any policy's performance. TVS measures the magnitude of torque variation required to correct small pose perturbations, directly capturing how dynamical properties shape the reinforcement learning landscape. We establish that high-TV motions induce flat reward landscapes and vanishing policy gradients, explaining persistent imitation failures. Extensive experiments with state-of-the-art methods (UHC, PHC+) confirm TVS strongly correlates with imitation error and enables principled error attribution: high error on low-TV motions indicates policy deficiency, while high error on high-TV motions reflects fundamental learning constraints. Beyond error diagnosis, TVS facilitates three practical applications: Maximum Imitable Difficulty (MID) for policy capability assessment, Difficulty-Stratified Joint Error (DSJE) for granular performance profiling, and Flawed Motion Detection for identifying segments with abnormally high learning difficulty to support mocap data curation and quality control. TVS provides a rigorous lens to distinguish policy-induced errors from motion-inherent challenges and enhances motion dataset reliability.

Figures

Figures reproduced from arXiv: 2512.07248 by Anjun Chen, Chengxu Zuo, Lu Yin, Shihui Guo, Xinrui Chen, Yipeng Qin, Zhaorui Meng.

Figure 1
Figure 1. Figure 1: Our Motion Difficulty Score (MDS) accurately quanti [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of MDS: For an easy motion (top-left), small pose perturbations induce small torque variance and hence low sensitivity to perturbation, the 1) smaller torque space volume, 2) larger variation across joints (larger volume on the waving hand joints than others) and 3) larger temporal variability makes MDS rates the motion as easier. In contract, for a a difficult motion (bottom-left), the same l… view at source ↗
Figure 3
Figure 3. Figure 3: Our Difficulty-aware AMASS Dataset MD-AMASS. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: We plot scatters of MDS versus imitation error on different polices for over 3000 motion clips (left: UHC; right: PHC+), with [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Imitation fidelity visibly degrades as MDS increases, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: More comprehensive scatter plots of MDS vs imitation error (top: UHC; bottom: PHC+), with over 10,000 motion samples [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative results of the Unitree G1-based TWIST [ [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. ThorArena: Benchmarking Humanoid Physical Interaction with Human Motion-Force Demonstrations

    cs.RO 2026-07 conditional novelty 6.0

    A force-aware humanoid benchmark pairs synchronized human motion-force data with simulation-based force replay to evaluate whole-body control policies under realistic physical disturbances.

Reference graph

Works this paper leans on

58 extracted references · 7 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Locomujoco: A comprehensive imitation learning bench- mark for locomotion.arXiv preprint arXiv:2311.02496,

    Firas Al-Hafez, Guoping Zhao, Jan Peters, and Davide Tateo. Locomujoco: A comprehensive imitation learning bench- mark for locomotion.arXiv preprint arXiv:2311.02496,

    Pith/arXiv arXiv

  2. [2]

    Noise-in, bias-out: Balanced and real-time mocap solving, 2023

    Georgios Albanis, Nikolaos Zioulis, Spyridon Thermos, Anargyros Chatzitofis, and Kostas Kolomvatsos. Noise-in, bias-out: Balanced and real-time mocap solving, 2023. 13

    2023

  3. [3]

    Digital dance ethnography: Organizing large dance collec- tions.J

    Andreas Aristidou, Ariel Shamir, and Yiorgos Chrysanthou. Digital dance ethnography: Organizing large dance collec- tions.J. Comput. Cult. Herit., 12(4), Nov. 2019. 3

    2019

  4. [4]

    Curriculum learning

    Yoshua Bengio, J ´erˆome Louradour, Ronan Collobert, and Ja- son Weston. Curriculum learning. InProceedings of the 26th Annual International Conference on Machine Learning, ICML ’09, page 41–48, New York, NY , USA, 2009. Associ- ation for Computing Machinery. 13

    2009

  5. [5]

    Drecon: data-driven responsive con- trol of physics-based characters.ACM Transactions On Graphics (TOG), 38(6):1–11, 2019

    Kevin Bergamin, Simon Clavet, Daniel Holden, and James Richard Forbes. Drecon: data-driven responsive con- trol of physics-based characters.ACM Transactions On Graphics (TOG), 38(6):1–11, 2019. 2

    2019

  6. [6]

    Humman: Multi-modal 4d human dataset for ver- satile sensing and modeling

    Zhongang Cai, Daxuan Ren, Ailing Zeng, Zhengyu Lin, Tao Yu, Wenjia Wang, Xiangyu Fan, Yang Gao, Yifan Yu, Liang Pan, et al. Humman: Multi-modal 4d human dataset for ver- satile sensing and modeling. InEuropean Conference on Computer Vision, pages 557–577. Springer, 2022. 3

    2022

  7. [7]

    CMU MoCap Dataset

    Carnegie Mellon University. CMU MoCap Dataset. 2, 3

  8. [8]

    Learning to sit: Synthesizing human-chair interactions via hierarchical control

    Yu-Wei Chao, Jimei Yang, Weifeng Chen, and Jia Deng. Learning to sit: Synthesizing human-chair interactions via hierarchical control. InProceedings of the AAAI Confer- ence on Artificial Intelligence, volume 35, pages 5887–5895,

  9. [9]

    Physics-based motion capture imitation with deep reinforcement learning

    Nuttapong Chentanez, Matthias M ¨uller, Miles Macklin, Vik- tor Makoviychuk, and Stefan Jeschke. Physics-based motion capture imitation with deep reinforcement learning. InPro- ceedings of the 11th ACM SIGGRAPH Conference on Mo- tion, Interaction and Games, pages 1–10, 2018. 2, 3

    2018

  10. [10]

    Haa500: Human-centric atomic action dataset with curated videos

    Jihoon Chung, Cheng-hsin Wuu, Hsuan-ru Yang, Yu-Wing Tai, and Chi-Keung Tang. Haa500: Human-centric atomic action dataset with curated videos. InProceedings of the IEEE/CVF international conference on computer vision, pages 13465–13474, 2021. 3

    2021

  11. [11]

    C·ase: Learning conditional adversarial skill embeddings for physics-based characters.arXiv preprint arXiv:2309.11351, 2023

    Zhiyang Dou, Xuelin Chen, Qingnan Fan, Taku Komura, and Wenping Wang. C·ase: Learning conditional adversarial skill embeddings for physics-based characters.arXiv preprint arXiv:2309.11351, 2023. 2

    Pith/arXiv arXiv 2023

  12. [12]

    Springer, 2008

    Roy Featherstone.Rigid body dynamics algorithms. Springer, 2008. 2, 3

    2008

  13. [13]

    Martin L. Felis. Rbdl: an efficient rigid-body dynam- ics library using recursive algorithms.Auton. Robots, 41(2):495–511, Feb. 2017. 6, 9

    2017

  14. [14]

    Super- track: Motion tracking for physically simulated characters using supervised learning.ACM Transactions on Graphics (TOG), 40(6):1–13, 2021

    Levi Fussell, Kevin Bergamin, and Daniel Holden. Super- track: Motion tracking for physically simulated characters using supervised learning.ACM Transactions on Graphics (TOG), 40(6):1–13, 2021. 2

    2021

  15. [15]

    Posetriplet: Co-evolving 3d human pose estimation, imita- tion, and hallucination under self-supervision

    Kehong Gong, Bingbing Li, Jianfeng Zhang, Tao Wang, Jing Huang, Michael Bi Mi, Jiashi Feng, and Xinchao Wang. Posetriplet: Co-evolving 3d human pose estimation, imita- tion, and hallucination under self-supervision. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11017–11027, 2022

    2022

  16. [16]

    Comic: Complementary task learn- ing & mimicry for reusable skills

    Leonard Hasenclever, Fabio Pardo, Raia Hadsell, Nicolas Heess, and Josh Merel. Comic: Complementary task learn- ing & mimicry for reusable skills. InInternational Confer- ence on Machine Learning, pages 4105–4115. PMLR, 2020. 2

    2020

  17. [17]

    Troje, and J ¨orn Munzert

    Fabian Helm, Nikolaus F. Troje, and J ¨orn Munzert. Mo- tion database of disguised and non-disguised team handball penalty throws by novice and expert performers.Data in Brief, 15:981–986, 2017. 3

    2017

  18. [18]

    Black, Otmar Hilliges, and Gerard Pons-Moll

    Yinghao Huang, Manuel Kaufmann, Emre Aksan, Michael J. Black, Otmar Hilliges, and Gerard Pons-Moll. Deep inertial poser learning to reconstruct human pose from sparseinertial measurements in real time.ACM Transactions on Graphics, (Proc. SIGGRAPH Asia), 37(6):185:1–185:15, Nov. 2018. 15

    2018

  19. [19]

    Human3.6m: Large scale datasets and predic- tive methods for 3d human sensing in natural environments

    Catalin Ionescu, Dragos Papava, Vlad Olaru, and Cristian Sminchisescu. Human3.6m: Large scale datasets and predic- tive methods for 3d human sensing in natural environments. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 36:1325–1339, 2014. 3

    2014

  20. [20]

    Phuma: Physically-grounded humanoid locomotion dataset.arXiv preprint arXiv:2510.26236, 2025

    Kyungmin Lee, Sibeen Kim, Minho Park, Hyunseung Kim, Dongyoon Hwang, Hojoon Lee, and Jaegul Choo. Phuma: Physically-grounded humanoid locomotion dataset.arXiv preprint arXiv:2510.26236, 2025. 3

    Pith/arXiv arXiv 2025

  21. [21]

    Finedance: A fine-grained choreography dataset for 3d full body dance generation

    Ronghui Li, Junfan Zhao, Yachao Zhang, Mingyang Su, Zeping Ren, Han Zhang, Yansong Tang, and Xiu Li. Finedance: A fine-grained choreography dataset for 3d full body dance generation. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages 10234–10243, October 2023. 3

    2023

  22. [22]

    Motion-x: A large-scale 3d expressive whole-body human motion dataset.Advances in Neural Information Processing Sys- tems, 36:25268–25280, 2023

    Jing Lin, Ailing Zeng, Shunlin Lu, Yuanhao Cai, Ruimao Zhang, Haoqian Wang, and Lei Zhang. Motion-x: A large-scale 3d expressive whole-body human motion dataset.Advances in Neural Information Processing Sys- tems, 36:25268–25280, 2023. 3

    2023

  23. [23]

    Smpl: A skinned multi- person linear model

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multi- person linear model. InSeminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 851–866. 2023. 3

    2023

  24. [24]

    Perpetual humanoid control for real-time simulated avatars

    Zhengyi Luo, Jinkun Cao, Kris Kitani, Weipeng Xu, et al. Perpetual humanoid control for real-time simulated avatars. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 10895–10904, 2023. 1, 2, 3, 14

    2023

  25. [25]

    Universal hu- manoid motion representations for physics-based control

    Zhengyi Luo, Jinkun Cao, Josh Merel, Alexander Winkler, Jing Huang, Kris Kitani, and Weipeng Xu. Universal hu- manoid motion representations for physics-based control. arXiv preprint arXiv:2310.04582, 2023. 1

    Pith/arXiv arXiv 2023

  26. [26]

    Dynamics-regulated kinematic policy for egocentric pose es- timation.Advances in Neural Information Processing Sys- tems, 34:25019–25032, 2021

    Zhengyi Luo, Ryo Hachiuma, Ye Yuan, and Kris Kitani. Dynamics-regulated kinematic policy for egocentric pose es- timation.Advances in Neural Information Processing Sys- tems, 34:25019–25032, 2021. 1, 2, 14

    2021

  27. [27]

    Amass: Archive of motion capture as surface shapes

    Naureen Mahmood, Nima Ghorbani, Nikolaus F Troje, Ger- ard Pons-Moll, and Michael J Black. Amass: Archive of motion capture as surface shapes. InProceedings of the IEEE/CVF international conference on computer vision, pages 5442–5451, 2019. 2, 3

    2019

  28. [28]

    Universal humanoid robot pose learning from internet human videos

    Jiageng Mao, Siheng Zhao, Siqi Song, Chuye Hong, Tian- heng Shi, Junjie Ye, Mingtong Zhang, Haoran Geng, Jitendra Malik, Vitor Guizilini, et al. Universal humanoid robot pose learning from internet human videos. In2025 IEEE-RAS 24th International Conference on Humanoid Robots (Hu- manoids), pages 1–8. IEEE, 2025. 3

    2025

  29. [29]

    Catch & carry: reusable neural controllers for vision-guided whole-body tasks.ACM Trans- actions on Graphics (TOG), 39(4):39–1, 2020

    Josh Merel, Saran Tunyasuvunakool, Arun Ahuja, Yuval Tassa, Leonard Hasenclever, Vu Pham, Tom Erez, Greg Wayne, and Nicolas Heess. Catch & carry: reusable neural controllers for vision-guided whole-body tasks.ACM Trans- actions on Graphics (TOG), 39(4):39–1, 2020. 2, 3

    2020

  30. [30]

    Stablemotion: Training motion cleanup models with un- paired corrupted data, 2025

    Yuxuan Mu, Hung Yu Ling, Yi Shi, Ismael Baira Ojeda, Pengcheng Xi, Chang Shu, Fabio Zinno, and Xue Bin Peng. Stablemotion: Training motion cleanup models with un- paired corrupted data, 2025. 13

    2025

  31. [31]

    Learning predict-and-simulate policies from un- organized human motion data.ACM Transactions on Graph- ics (TOG), 38(6):1–11, 2019

    Soohwan Park, Hoseok Ryu, Seyoung Lee, Sunmin Lee, and Jehee Lee. Learning predict-and-simulate policies from un- organized human motion data.ACM Transactions on Graph- ics (TOG), 38(6):1–11, 2019. 3

    2019

  32. [32]

    Expressive body capture: 3d hands, face, and body from a single image

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed AA Osman, Dimitrios Tzionas, and Michael J Black. Expressive body capture: 3d hands, face, and body from a single image. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10975–10985, 2019. 3

    2019

  33. [33]

    Deepmimic: Example-guided deep reinforce- ment learning of physics-based character skills.ACM Trans- actions On Graphics (TOG), 37(4):1–14, 2018

    Xue Bin Peng, Pieter Abbeel, Sergey Levine, and Michiel Van de Panne. Deepmimic: Example-guided deep reinforce- ment learning of physics-based character skills.ACM Trans- actions On Graphics (TOG), 37(4):1–14, 2018. 2, 3

    2018

  34. [34]

    Deeploco: Dynamic locomotion skills using hierarchical deep reinforcement learning.Acm transactions on graphics (TOG), 36(4):1–13, 2017

    Xue Bin Peng, Glen Berseth, KangKang Yin, and Michiel Van De Panne. Deeploco: Dynamic locomotion skills using hierarchical deep reinforcement learning.Acm transactions on graphics (TOG), 36(4):1–13, 2017. 2, 3

    2017

  35. [35]

    Mcp: Learning composable hierarchical control with multiplicative compositional policies.Advances in neural information processing systems, 32, 2019

    Xue Bin Peng, Michael Chang, Grace Zhang, Pieter Abbeel, and Sergey Levine. Mcp: Learning composable hierarchical control with multiplicative compositional policies.Advances in neural information processing systems, 32, 2019

    2019

  36. [36]

    Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters.ACM Transactions On Graphics (TOG), 41(4):1–17, 2022

    Xue Bin Peng, Yunrong Guo, Lina Halper, Sergey Levine, and Sanja Fidler. Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters.ACM Transactions On Graphics (TOG), 41(4):1–17, 2022. 2

    2022

  37. [37]

    Sfv: Reinforcement learning of physical skills from videos.ACM Transactions On Graphics (TOG), 37(6):1–14, 2018

    Xue Bin Peng, Angjoo Kanazawa, Jitendra Malik, Pieter Abbeel, and Sergey Levine. Sfv: Reinforcement learning of physical skills from videos.ACM Transactions On Graphics (TOG), 37(6):1–14, 2018. 3

    2018

  38. [38]

    Amp: Adversarial motion priors for styl- ized physics-based character control.ACM Transactions 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 styl- ized physics-based character control.ACM Transactions on Graphics (TOG), 40(4):1–20, 2021. 2, 3

    2021

  39. [39]

    Magshield: Towards better robustness in sparse inertial motion capture under magnetic disturbances

    Yunzhe Shao, Xinyu Yi, Lu Yin, Shihui Guo, Junhai Yong, and Feng Xu. Magshield: Towards better robustness in sparse inertial motion capture under magnetic disturbances. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 29021–29030, October

  40. [40]

    Physcap: Physically plausible monocular 3d motion capture in real time.ACM Transactions on Graphics, 39(6), dec 2020

    Soshi Shimada, Vladislav Golyanik, Weipeng Xu, and Chris- tian Theobalt. Physcap: Physically plausible monocular 3d motion capture in real time.ACM Transactions on Graphics, 39(6), dec 2020. 3

    2020

  41. [41]

    Curriculum learning: A survey.Int

    Petru Soviany, Radu Tudor Ionescu, Paolo Rota, and Nicu Sebe. Curriculum learning: A survey.Int. J. Comput. Vision, 130(6):1526–1565, June 2022. 14

    2022

  42. [42]

    Aist dance video database: Multi-genre, multi-dancer, and multi-camera database for dance informa- tion processing

    Shuhei Tsuchida, Satoru Fukayama, Masahiro Hamasaki, and Masataka Goto. Aist dance video database: Multi-genre, multi-dancer, and multi-camera database for dance informa- tion processing. InISMIR, volume 1, page 6, 2019. 3

    2019

  43. [43]

    Unitree G1 humanoid robot.https: //www.unitree.com/cn/g1, 2024

    Unitree Robotics. Unitree G1 humanoid robot.https: //www.unitree.com/cn/g1, 2024. Accessed: 2025- 11-17. 14

    2024

  44. [44]

    Mocapact: A multi-task dataset for simulated humanoid con- trol.Advances in Neural Information Processing Systems, 35:35418–35431, 2022

    Nolan Wagener, Andrey Kolobov, Felipe Vieira Frujeri, Ricky Loynd, Ching-An Cheng, and Matthew Hausknecht. Mocapact: A multi-task dataset for simulated humanoid con- trol.Advances in Neural Information Processing Systems, 35:35418–35431, 2022. 2

    2022

  45. [45]

    Unicon: Universal neural controller for physics-based character motion.arXiv preprint arXiv:2011.15119, 2020

    Tingwu Wang, Yunrong Guo, Maria Shugrina, and Sanja Fi- dler. Unicon: Universal neural controller for physics-based character motion.arXiv preprint arXiv:2011.15119, 2020. 2

    Pith/arXiv arXiv 2011

  46. [46]

    Quest- sim: Human motion tracking from sparse sensors with simu- lated avatars

    Alexander Winkler, Jungdam Won, and Yuting Ye. Quest- sim: Human motion tracking from sparse sensors with simu- lated avatars. InSIGGRAPH Asia 2022 Conference Papers, pages 1–8, 2022. 2

    2022

  47. [47]

    A scalable approach to control diverse behaviors for physi- cally simulated characters.ACM Transactions on Graphics (TOG), 39(4):33–1, 2020

    Jungdam Won, Deepak Gopinath, and Jessica Hodgins. A scalable approach to control diverse behaviors for physi- cally simulated characters.ACM Transactions on Graphics (TOG), 39(4):33–1, 2020. 2, 3

    2020

  48. [48]

    Controlvae: Model-based learning of generative controllers for physics-based characters.ACM Trans

    Heyuan Yao, Zhenhua Song, Baoquan Chen, and Libin Liu. Controlvae: Model-based learning of generative controllers for physics-based characters.ACM Trans. Graph., 41(6),

  49. [49]

    Moconvq: Unified physics- based motion control via scalable discrete representations

    Heyuan Yao, Zhenhua Song, Yuyang Zhou, Tenglong Ao, Baoquan Chen, and Libin Liu. Moconvq: Unified physics- based motion control via scalable discrete representations. ACM Trans. Graph., 43(4), July 2024. 2

    2024

  50. [50]

    Shape-aware inertial poser: Motion tracking for humans with diverse shapes using sparse inertial sensors,

    Lu Yin, Ziying Shi, Yinghao Wu, Xinyu Yi, Feng Xu, and Shihui Guo. Shape-aware inertial poser: Motion tracking for humans with diverse shapes using sparse inertial sensors,

  51. [51]

    Ego-pose estimation and forecast- ing as real-time pd control

    Ye Yuan and Kris Kitani. Ego-pose estimation and forecast- ing as real-time pd control. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 10082– 10092, 2019. 2

    2019

  52. [52]

    Residual force control for ag- ile human behavior imitation and extended motion synthe- sis.Advances in Neural Information Processing Systems, 33:21763–21774, 2020

    Ye Yuan and Kris Kitani. Residual force control for ag- ile human behavior imitation and extended motion synthe- sis.Advances in Neural Information Processing Systems, 33:21763–21774, 2020. 2, 3

    2020

  53. [53]

    Karen Liu

    Yanjie Ze, Zixuan Chen, Jo ˜ao Pedro Ara ´ujo, Zi ang Cao, Xue Bin Peng, Jiajun Wu, and C. Karen Liu. Twist: Teleoperated whole-body imitation system.arXiv preprint arXiv:2505.02833, 2025. 14

    Pith/arXiv arXiv 2025

  54. [54]

    Ego- body: Human body shape and motion of interacting peo- ple from head-mounted devices

    Siwei Zhang, Qianli Ma, Yan Zhang, Zhiyin Qian, Taein Kwon, Marc Pollefeys, Federica Bogo, and Siyu Tang. Ego- body: Human body shape and motion of interacting peo- ple from head-mounted devices. InEuropean conference on computer vision, pages 180–200. Springer, 2022. 3

    2022

  55. [55]

    A plug-and-play physical motion restoration approach for in-the-wild high- difficulty motions

    Youliang Zhang, Ronghui Li, Yachao Zhang, Liang Pan, Jingbo Wang, Yebin Liu, and Xiu Li. A plug-and-play physical motion restoration approach for in-the-wild high- difficulty motions. InProceedings of the IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages 13281–13292, October 2025. 3

    2025

  56. [56]

    Motion-x++: A large-scale multi- modal 3d whole-body human motion dataset.arXiv preprint arXiv:2501.05098, 2025

    Yuhong Zhang, Jing Lin, Ailing Zeng, Guanlin Wu, Shun- lin Lu, Yurong Fu, Yuanhao Cai, Ruimao Zhang, Haoqian Wang, and Lei Zhang. Motion-x++: A large-scale multi- modal 3d whole-body human motion dataset.arXiv preprint arXiv:2501.05098, 2025. 3

    Pith/arXiv arXiv 2025

  57. [57]

    Human motion tracking con- trol with strict contact force constraints for floating-base hu- manoid robots

    Yu Zheng and Katsu Yamane. Human motion tracking con- trol with strict contact force constraints for floating-base hu- manoid robots. In2013 13th IEEE-RAS International Con- ference on Humanoid Robots (Humanoids), pages 34–41,

  58. [58]

    Transformer imu calibrator: Dynamic on-body imu calibration for inertial motion capture.ACM Trans

    Chengxu Zuo, Jiawei Huang, Xiao Jiang, Yuan Yao, Xian- gren Shi, Rui Cao, Xinyu Yi, Feng Xu, Shihui Guo, and Yipeng Qin. Transformer imu calibrator: Dynamic on-body imu calibration for inertial motion capture.ACM Trans. Graph., 44(4), July 2025. 13, 15 A. Spectral Diversity Details Following the definitions from the main paper, we complete the detailed pr...

    2025

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.