Pith. sign in

REVIEW 3 major objections 5 minor 29 cited by

GMT: General Motion Tracking for Humanoid Whole-Body Control

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read GMT claims a single policy can track diverse whole-body human motions on a real humanoid robot, with lower simulation tracking error than a re-implemented ExBody2 baseline.

desk verdict A solid, incremental contribution to humanoid control with a genuinely useful motion-input design, but the SOTA claim needs more baselines and curation details. read the letter →

arxiv 2506.14770 v2 pith:HZB74YG2 submitted 2025-06-17 cs.RO

Zixuan Chen , Mazeyu Ji , Xuxin Cheng , Xuanbin Peng , Xue Bin Peng , Xiaolong Wang This is my paper · ORCID
classification cs.RO
keywords HumanoidLocomotionLearning-basedControlMotionImitationWhole-bodyMixture-of-ExpertsAdaptiveSamplingTeacher-studentTraining
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

The paper claims that a single policy can be trained to make a real humanoid robot track a broad spectrum of whole-body human motions, from walking and stretching to kicks, dancing, and kung-fu. It argues that two obstacles stand in the way: large motion datasets are skewed toward easy, common movements, and a single neural network does not have enough expressive power to cover a wide motion manifold. GMT addresses the first with Adaptive Sampling, which re-weights training clips by how well the current policy completes them, and the second with a Mixture-of-Experts policy head that lets different experts specialize in different motion regions. In simulation, the resulting policy reports lower tracking errors than a re-implemented ExBody2 baseline, and the same policy is deployed on a Unitree G1 humanoid to reproduce a range of dynamic skills. If correct, this moves toward a general whole-body controller that a high-level planner could sequence into tasks.

What carries the argument

The load-bearing machinery is a two-stage teacher-student training loop whose teacher combines a Motion Mixture-of-Experts head with an Adaptive Sampling curriculum. The MoE is a set of expert networks plus a gating network that reads the same observations and outputs a probability distribution; the final action is the probability-weighted combination of the experts' sampled actions, giving the model a way to specialize across the motion manifold. Adaptive Sampling sets each clip's sampling probability from its completion count and current tracking error, so easy motions are down-weighted and hard motions dominate training. Around this core, GMT also relies on a convolutional encoder that compresses about two seconds of future motion frames into a latent vector fed to the policy alongside the immediate next frame. The whole system is trained first as a privileged PPO policy and then distilled into a deployable student policy by DAgger.

What would settle it

Retrieve the clips discarded by the second-stage filter, add them back into the training set, and re-train GMT; if a substantial fraction of those clips still cannot be tracked by the final policy, the curated 'general' claim is not supported.

Watch

Extended reading notes

Core claim

On its own terms, GMT's discovery is that joint curation of the data distribution and the model architecture unlocks a single general motion-tracking policy for a physical humanoid. The training set is first filtered by rules and then by a preliminary policy's completion rates, leaving 8,925 clips, or 33.12 hours, from AMASS and LAFAN1. The teacher policy is trained with PPO under domain randomization, using a Mixture-of-Experts network whose gating network weights the outputs of specialized expert policies, while Adaptive Sampling raises the sampling probability of motions the current policy still fails. Motion targets are represented with local key-body positions aligned to the robot's heading, and the policy sees both the immediate next frame and roughly two seconds of future frames compressed by a convolutional encoder. The paper reports that this single policy outperforms its re-implemented ExBody2 baseline on AMASS-test and LAFAN1 tracking metrics and reproduces the motions on a real Unitree G1, including dynamic skills.

Load-bearing premise

The claim of generality rests on the assumption that the second-stage data filter, which drops motions a preliminary policy fails to complete, is not removing motions that a better final policy could have learned and that would change the reported performance.

Editorial extensions

If this is right

  • A single deployed policy can replace a bank of specialist policies for walking, kicking, dancing, and other whole-body skills on a 23-DoF humanoid.
  • Because the policy tracks text-generated motions from a motion diffusion model in simulation, it can act as the low-level executor for high-level planners that generate or select motions on the fly.
  • Most of the performance gain from Adaptive Sampling and the MoE appears on the hardest motion categories, so the framework's advantage grows with dataset diversity.
  • Using local key-body positions and a multi-frame future-motion window is a transferable design for other motion-imitation pipelines.

Reading between the lines

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

  • A testable consequence the paper does not pursue: if the single-policy claim holds, tracking error on a held-out set of skills should keep dropping as the curated dataset grows; a plateau would point to the MoE's capacity or the sampling schedule as the bottleneck.
  • An extension would apply the same two mechanisms to contact-rich skills such as getting up or rolling, which the paper lists as unsupported; this would probe whether the limitation is the data filter or the architecture.
  • A practical benchmark would report per-category completion rates on a fixed motion taxonomy, so future general controllers can be compared by coverage rather than aggregate error.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces GMT, a two-stage teacher-student framework for training a single unified whole-body motion-tracking policy for a humanoid robot (Unitree G1). The two main technical contributions are an Adaptive Sampling strategy that reweights training motions by tracking difficulty, and a Motion Mixture-of-Experts (MoE) architecture for the policy network. The policy is trained on a filtered combination of AMASS and LAFAN1 (8,925 clips, 33.12 hours) and evaluated in simulation against a re-implemented ExBody2 baseline, with ablations for each component and for the motion-input representation. Real-world deployment is shown qualitatively on a variety of motor skills, and an additional MuJoCo experiment demonstrates tracking of motion-diffusion-model-generated motions. The central empirical claim is that a single policy can track a broad spectrum of whole-body motions more accurately than the compared baseline, with each component contributing to improved simulation metrics.

Significance. If the claims hold, GMT would be a useful step toward general-purpose whole-body controllers for humanoid robots, and the Adaptive Sampling and MoE design choices are plausible and well motivated by the observed category imbalance in AMASS. The ablation study is systematic and the simulation metrics are standard; the real-world deployment videos are compelling evidence of basic feasibility. The paper also deserves credit for explicitly stating limitations (no contact-rich skills, no challenging terrains) and for evaluating on an external motion-generation pipeline (MDM), which is a good generalization test. However, the significance is currently bounded by two issues: the 'state-of-the-art' claim rests on a single re-implemented baseline, and the dataset-curation procedure is not documented quantitatively, so the breadth of the 'general' claim cannot be fully verified from the manuscript as written.

major comments (3)
  1. [Section 3.3] The second stage of dataset curation is described only qualitatively: a preliminary policy is trained on the rule-filtered data and 'based on the completion rates achieved by this policy, we further filter out failed motions.' No completion-rate threshold, no fraction of removed clips, and no analysis of which motion categories or complexity levels were discarded are reported. Since the final policy is more capable than the preliminary policy (as the ablations in Table 2(a) suggest), this filtering may preferentially remove exactly the hard motions that a stronger policy could learn, and the 'general' claim is therefore unverified. Please report the threshold, the removal statistics, and a breakdown of removed motions by category and difficulty, and consider evaluating on an unfiltered or independently filtered test set.
  2. [Section 4.2 and Table 2] The abstract and conclusion claim 'state-of-the-art performance,' but the only quantitative baseline is a re-implemented ExBody2. Other recent whole-body controllers mentioned in the related work (HumanPlus, OmniH2O, VMP, ASAP) are not compared numerically, and the real-world evidence is qualitative. The comparison would support a claim of 'improvement over the re-implemented ExBody2 baseline' but not a general state-of-the-art claim. Please either add quantitative comparisons to at least one or two additional recent methods or revise the claim to match the evidence.
  3. [Section 4.1 and Table 2] The simulation evaluation is reported on 'AMASS test set' and 'LAFAN1,' but the text does not state whether these evaluation sets were filtered by the same two-stage curation pipeline described in Section 3.3. If the evaluation sets are subsets produced by the same preliminary-policy filter, the reported numbers may not reflect performance on the full diversity of human motion, and the comparison with the baseline would be measured on a potentially biased subset. The manuscript should clarify the relationship between the training filter and the evaluation sets, and ideally report results on an unfiltered test set as well.
minor comments (5)
  1. [Table 2 and Section 4.1] The text says all baseline comparisons and ablations use privileged policies, but Table 2 also lists a 'Student Policy' row; please clarify whether the student row is part of the main comparison and how it relates to the stated evaluation protocol.
  2. [Equation (1)] The variable E_max in the sampling level definition is not clearly defined; please specify whether it is a per-motion maximum key-body error, a threshold, or a normalization constant, and give its value or estimation procedure.
  3. [Figure 5] The 'top percentile tracking errors' are not defined; please state how the top percentiles are computed (e.g., worst episodes, worst timesteps, or worst motion clips) and whether the errors are aggregated over all motions or per category.
  4. [Section 4.4] The real-world deployment section provides no quantitative metrics; adding even basic measured tracking errors or success rates for a few representative motions would considerably strengthen the real-world claim.
  5. [Section 3.4] The motion-input ablation 'GMT-L2' is said to degrade because the immediate next frame is omitted, but the comparison also varies the total input window; please make explicit whether the window length is the only difference between GMT-L2 and GMT-L2-M, or whether the encoding architecture also differs.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all reported results are empirical, comparisons share the same filtered training data, and no fitted parameter is renamed as a prediction.

full rationale

The paper's derivation chain is empirical throughout: a privileged teacher policy is trained by PPO with hand-specified rewards (Table 3) and a student policy is trained by DAgger against the teacher; the Adaptive Sampling rule of Eq. (1) uses online tracking errors only as a training curriculum, which is not a prediction of the evaluation metrics on held-out data. The two-stage dataset curation of Sec. 3.3 uses a preliminary policy's completion rates to prune infeasible or overly hard clips; although the paper does not report pruning thresholds or removed fractions, the evaluation is against the AMASS test set and LAFAN1, and all compared policies are trained on the same filtered set, so the reported advantage over the re-implemented ExBody2 baseline is not forced by construction. The only self-citations (to [5] and [7]) are methodological, not load-bearing: the two-stage teacher-student scheme and curation procedure are described in the paper itself, and the ExBody2 baseline is re-implemented rather than taken on faith. Limitations are acknowledged (contact-rich skills and terrain), which bounds the 'general' claim but does not create circularity. No equation or fitted parameter is renamed as a prediction; therefore the score is 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard assumptions about data coverage, simulator fidelity, and RL optimization; no new physical entities are introduced. The main hand-tuned elements are the adaptive sampling constants and domain randomization ranges.

free parameters (4)
  • Adaptive Sampling constants = c0=10, decay=0.99, min=1, E range [0.25, 0.6], key-body error normalizer 0.15, exponent 5
    Hand-picked in Section 3.1, Eq. (1); no sensitivity study. They control the curriculum and are not derived from data.
  • Motion input future window length = 2 seconds (GMT-L2-M)
    Selected via ablations in Table 2(b); this is a hyperparameter, not a fitted constant, but it affects performance.
  • Domain randomization ranges = e.g., friction [0.1,2.0], motor strength [0.8,1.2], push velocity [0,1] m/s
    Chosen from prior practice; appendix Table 4, not swept in this paper.
  • MoE architectural hyperparameters = not reported
    Number of experts, expert hidden sizes, and gating details are absent, so replication requires guessing.
assumptions (4)
  • domain assumption AMASS and LAFAN1 cover a sufficiently broad and feasible motion manifold for general whole-body tracking after curation.
    Section 3.3; if the curated 8925 clips miss important motions, the 'general' claim weakens.
  • domain assumption The IsaacGym simulation with the given domain randomization and armature inertia model is a faithful proxy for the Unitree G1 on the tested skills.
    Appendix 6.2; real-world success depends on this.
  • standard math PPO and DAgger solve the posed optimization problems without special tuning beyond the described settings.
    Appendix 6.3; the paper uses standard implementations.
  • domain assumption Mixture-of-experts does not introduce optimization instability in this setting.
    Section 3.2; ablations suggest it helps, but the mechanism is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GMT: General Motion Tracking for Humanoid Whole-Body Control." pith.science (2026). https://pith.science/paper/HZB74YG2

@misc{pith2026250614770,
  author       = {Pith},
  title        = {Pith review of: GMT: General Motion Tracking for Humanoid Whole-Body Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HZB74YG2}},
  note         = {Machine review of arXiv:2506.14770}
}
read the original abstract

The ability to track general whole-body motions in the real world is a useful way to build general-purpose humanoid robots. However, achieving this can be challenging due to the temporal and kinematic diversity of the motions, the policy's capability, and the difficulty of coordination of the upper and lower bodies. To address these issues, we propose GMT, a general and scalable motion-tracking framework that trains a single unified policy to enable humanoid robots to track diverse motions in the real world. GMT is built upon two core components: an Adaptive Sampling strategy and a Motion Mixture-of-Experts (MoE) architecture. The Adaptive Sampling automatically balances easy and difficult motions during training. The MoE ensures better specialization of different regions of the motion manifold. We show through extensive experiments in both simulation and the real world the effectiveness of GMT, achieving state-of-the-art performance across a broad spectrum of motions using a unified general policy. Videos and additional information can be found at https://gmt-humanoid.github.io.

Figures

Figures reproduced from arXiv: 2506.14770 by the authors.

Figure 1
Figure 1. We deploy the general unified motion tracking policy on a medium-sized humanoid robot. GMT can perform a wide range of motion skills with good stability and generalizability, including (a) stretching, (b) kicking-ball, (c) dancing, (d) high kicking, (e) kungfu, and (f) other dynamic skills such as boxing, running, side stepping, and squatting. Abstract: The ability to track general whole-body motions in the real wor… view at source ↗
Figure 2
Figure 2. Distribution of motion categories in the AMASS dataset. The figure shows the proportion of the total motion duration corresponding to each category. In this paper, we propose GMT, a general and effec￾tive framework that trains a single unified, high-quality motion tracking policy for real-world humanoid robots from large mocap datasets. Central to GMT are two key innovations: a novel Adaptive Sampling strategy, desi… view at source ↗
Figure 3
Figure 3. An overview of GMT. Here gt denotes the motion target frame, ot denotes proprioceptive observation, and et denotes privileged information. tasks [30, 2, 31, 32, 33]. However, due to the partial observability in the real world, developing such whole-body controllers for real robots [10, 34, 9, 8, 7, 11, 35, 36, 37] can be challenging. For devel￾oping a unified general whole-body motion tracking controller, some works… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Plot of the output of gating network with respect to time on a motion clip composed of a [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Top percentile tracking errors on the whole AMASS dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The performance of policies with and without Adaptive Sampling on one segment ex [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Motion tracking on MDM-generated motions. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 29 Pith papers

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

  1. HumanTracker: Towards Comprehensive and Human-Aligned Motion Tracking Benchmark

    cs.RO 2026-08 conditional novelty 6.0 of 10

    HumanTracker introduces a 153-hour categorized humanoid tracking benchmark and a preference-trained metric, HumanScore, that agrees with human judgments better than kinematic error metrics.

  2. $\omega$-0: A Latent Predictive World Action Model for Concurrent Humanoid Loco-Manipulation

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A single whole-body model with latent future prediction outperforms prior robot policies on 11 real-world humanoid household loco-manipulation tasks.

  3. Learning Context-Aware Motion Priors for Humanoid Control

    cs.RO 2026-08 conditional novelty 6.0 of 10

    CMP learns to reweight reference motions by task context, improving task return and sample efficiency for AMP- and SMP-based humanoid control in simulation.

  4. PFM-HR: Pose Flow Matching for Humanoid Robots

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A flow-matching pose prior, kept frozen, provides a Jacobian-based directional score that modulates tracking rewards and improves RL-based humanoid motion tracking, especially for dynamic acrobatic skills.

  5. Teleopit: A Full-Embodiment Humanoid Teleoperation System

    cs.RO 2026-08 conditional novelty 6.0 of 10

    Teleopit combines VR body, hand, and head tracking with a learned whole-body tracker and a cross-hand retargeter, and teleop-collected demos train ACT and GR00T policies to around 90 to 95 percent success on a humanoi...

  6. Bicycle Acrobatics with Reinforcement Learning

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A reinforcement-learning pipeline trained on a custom bicycle robot, then orchestrated by a state machine, performs repeated acrobatic stunts including jumps, flips, wheelies, and kip-ups in hardware.

  7. LooperMuscle: Fast and Stable Learning of Humanoid Whole-Body Tracking via Structured Mixture-of-Experts

    cs.RO 2026-08 conditional novelty 6.0 of 10

    LooperMuscle trains a humanoid whole-body tracking policy in about 45 minutes, cutting body error by 34% versus FastSAC and reaching 72% of PPO's reward, versus PPO's 6 hours.

  8. A Change of Frame Makes Balance Observable: Distillation-Free Humanoid Single-Leg Stance

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A support-relative dynamic capture-point observation, reconstructible without base linear velocity, lets a humanoid policy hold clean single-leg balance at 86/90 in simulation and deploy on a Unitree G1 without distillation.

  9. Extreme-RGMT: Continual Learning of Highly Dynamic Skills for Robust Generalist Humanoid Control

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A two-stage continual-learning framework lets a generalist humanoid tracking policy acquire highly dynamic acrobatic skills while preserving its general-purpose motion capabilities.

  10. What Matters in Humanoid General Motion Tracking? An Empirical Study

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A controlled ablation of humanoid motion-tracking pipelines shows that explicit reference joint velocities and a short observation history improve tracking, while residual actions and teacher-student training yield on...

  11. Scaling Behavior Foundation Model for Humanoid Robots

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A scaling recipe for humanoid behavior foundation models—global-frame motion tracking, on-policy data quantity plus reference diversity, and a transformer with hyperspherical latents—cuts global tracking error by roug...

  12. EgoHTR: Egocentric 4D Demonstrations of Human Terrain Traversal

    cs.RO 2026-07 conditional novelty 6.0 of 10

    EgoHTR is a 55-sequence, 150k-frame egocentric 4D human-terrain dataset with a reconstruction pipeline, MoCap-validated benchmark, and perceptive locomotion policies deployed on a Unitree G1.

  13. ContactMimic: Humanoid Object Interaction via Contact Control

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A humanoid tracking policy is trained with contact-following rewards and trajectory augmentation to decouple physical contact from keypoint geometry, enabling runtime contact control.

  14. Behavior Foundations for Quadruped Robots: ABot-C0 Technical Report

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A multi-source 16,074-clip quadruped motion library plus a flow-matching generalist tracker shows empirical data scaling and zero-shot unseen tracking, integrated with all-terrain locomotion and real-robot deployment.

  15. WristMimic: Full-Body Humanoid Control with Wrist-Guided Manipulation

    cs.RO 2026-07 conditional novelty 6.0 of 10

    WristMimic achieves comparable or superior object manipulation retargeting by supervising wrist kinematics while letting finger behavior emerge from object and contact dynamics.

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

    cs.RO 2026-07 conditional novelty 6.0 of 10

    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.

  17. HERO: Learning Humanoid End-Effector Control for Visual Whole-Body Open-Vocabulary Object Grasping

    cs.RO 2026-02 conditional novelty 6.0 of 10

    HERO achieves 2.44 cm end-effector tracking error on a Unitree G1 humanoid and uses it, with open-vocabulary perception, to grasp novel objects at up to 90% success in diverse real scenes.

  18. EgoHumanoid: Unlocking In-the-Wild Loco-Manipulation with Robot-Free Egocentric Demonstration

    cs.RO 2026-02 conditional novelty 6.0 of 10

    Co-training a vision-language-action humanoid policy on aligned egocentric human demonstrations plus limited robot data improves real-world loco-manipulation success by 20% in-domain and 51% in environments the robot ...

  19. Thor: Towards Human-Level Whole-Body Reactions for Intense Contact-Rich Environments

    cs.RO 2025-10 conditional novelty 6.0 of 10

    A decoupled whole-body RL policy with a force-based lean reward enables a Unitree G1 humanoid to pull with up to 167.7 N, beating prior controllers by 69–75%.

  20. PHUMA: Physically Reliable Humanoid Locomotion Dataset

    cs.RO 2025-10 conditional novelty 6.0 of 10

    PHUMA is a curated 73-hour humanoid locomotion corpus whose physical-reliability metrics are partly defined by the same losses used to optimize it, and whose imitation success claims are confounded by in-distribution ...

  21. A Scalable Whole-body Motion Transfer via Implicit Kinodynamic Motion Retargeting

    cs.RO 2025-09 conditional novelty 6.0 of 10

    A neural retargeting pipeline maps human motion to humanoid robot motion at 5000+ frames per second using a shared latent space and physics-based fine-tuning, filtering noise and producing physically feasible trajectories.

  22. HITTER: A HumanoId Table TEnnis Robot via Hierarchical Planning and Learning

    cs.RO 2025-08 conditional novelty 6.0 of 10

    A humanoid robot with a model-based planner and a reinforcement-learning controller returns table tennis balls and sustained a 106-shot rally against a human.

  23. FARM: Frame-Accelerated Augmentation and Residual Mixture-of-Experts for Physics-Based High-Dynamic Humanoid Control

    cs.RO 2025-08 conditional novelty 6.0 of 10

    FARM combines frame-accelerated augmentation with a residual mixture-of-experts to track high-dynamic humanoid motions, cutting tracking failures by 42.8% on a new HDHM benchmark.

  24. Athena-WBC: Capability-Aligned Policy Experts for Long-Tail Humanoid Whole-Body Control

    cs.RO 2026-07 conditional novelty 5.5 of 10

    Capability-aligned dynamic and balance experts recover residual humanoid whole-body tracking failures better than data reallocation alone, then distill into one stronger deployable controller.

  25. PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball

    cs.RO 2026-07 conditional novelty 5.0 of 10

    Link-CBF training with masked onboard depth lets a humanoid dodge whole-body contact near oracle rates in sim and on 95% of real throws, while stronger Joint-CBF needs better ball observability.

  26. ZeroWBC: Learning Natural Whole-Body Humanoid Interaction from Human Egocentric Data

    cs.RO 2026-03 conditional novelty 5.0 of 10

    An open-loop generation-then-tracking system maps one egocentric image plus language into Unitree G1 whole-body interactions using only human egocentric motion data.

  27. RobotDancing: Residual-Action Reinforcement Learning Enables Robust Long-Horizon Humanoid Motion Tracking

    cs.RO 2025-09 conditional novelty 5.0 of 10

    Residual-action reinforcement learning, with selective corrections on hip and knee pitch joints, enables zero-shot long-horizon dance tracking on real humanoid robots.

  28. Switch4EAI: Leveraging Console Game Platform for Benchmarking Robotic Athletics

    cs.RO 2025-08 conditional novelty 5.0 of 10

    Switch4EAI uses Just Dance on a Nintendo Switch as a benchmark: a Unitree G1 robot scored 5,707 versus a human's 9,361 on three routines.

  29. Feature-Based vs. GAN-Based Learning from Demonstrations: When and Why

    cs.LG 2025-07 conditional novelty 3.0 of 10

    Feature-based and GAN-based imitation learning should be selected by task priorities (fidelity, diversity, interpretability, adaptability), not by paradigm loyalty.

Reference graph

Works this paper leans on

49 extracted references · 18 canonical work pages · cited by 29 Pith papers

  1. [1]

    X. B. Peng, P. Abbeel, S. Levine, and M. Van de Panne. Deepmimic: Example-guided deep re- inforcement learning of physics-based character skills.ACM Transactions On Graphics (TOG), 37(4):1–14, 2018

    2018

  2. [2]

    X. B. Peng, Y . Guo, L. Halper, S. Levine, and S. Fidler. Ase: Large-scale reusable adversarial skill embeddings for physically simulated characters. ACM Transactions On Graphics (TOG), 41(4):1–17, 2022

    work page 2022

  3. [3]

    Z. Luo, J. Cao, K. Kitani, W. Xu, et al. Perpetual humanoid control for real-time simulated avatars. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 10895–10904, 2023

    work page 2023

  4. [4]

    T. E. Truong, M. Piseno, Z. Xie, and K. Liu. Pdp: Physics-based character animation via diffusion policy. In SIGGRAPH Asia 2024 Conference Papers, pages 1–10, 2024

    work page 2024

  5. [5]

    Tessler, Y

    C. Tessler, Y . Guo, O. Nabati, G. Chechik, and X. B. Peng. Maskedmimic: Unified physics- based character control through masked motion inpainting. ACM Transactions on Graphics (TOG), 43(6):1–21, 2024

    work page 2024

  6. [6]

    Mahmood, N

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

    work page 2019

  7. [7]

    M. Ji, X. Peng, F. Liu, J. Li, G. Yang, X. Cheng, and X. Wang. Exbody2: Advanced expressive humanoid whole-body control. arXiv preprint arXiv:2412.13196, 2024

    arXiv 2024

  8. [8]

    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

    arXiv 2024

Show all 49 references
  1. [9]

    Z. Fu, Q. Zhao, Q. Wu, G. Wetzstein, and C. Finn. Humanplus: Humanoid shadowing and imitation from humans. arXiv preprint arXiv:2406.10454, 2024

    2024 arXiv

  2. [10]

    Cheng, Y

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

    2024 arXiv

  3. [11]

    T. He, J. Gao, W. Xiao, Y . Zhang, Z. Wang, J. Wang, Z. Luo, G. He, N. Sobanbab, C. Pan, et al. Asap: Aligning simulation and real-world physics for learning agile humanoid whole- body skills. arXiv preprint arXiv:2502.01143, 2025

    2025 arXiv

  4. [12]

    Miura and I

    H. Miura and I. Shimoyama. Dynamic walk of a biped. The International Journal of Robotics Research, 3(2):60–74, 1984. 9

    1984

  5. [13]

    Sreenath, H.-W

    K. Sreenath, H.-W. Park, I. Poulakakis, and J. W. Grizzle. A compliant hybrid zero dynamics controller for stable, efficient and fast bipedal walking on mabel. The International Journal of Robotics Research, 30(9):1170–1193, 2011

    2011

  6. [14]

    Geyer, A

    H. Geyer, A. Seyfarth, and R. Blickhan. Positive force feedback in bouncing gaits? Proceed- ings of the Royal Society of London. Series B: Biological Sciences , 270(1529):2173–2183, 2003

    2003

  7. [15]

    Radosavovic, B

    I. Radosavovic, B. Zhang, B. Shi, J. Rajasegaran, S. Kamat, T. Darrell, K. Sreenath, and J. Ma- lik. Humanoid locomotion as next token prediction. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  8. [16]

    Radosavovic, S

    I. Radosavovic, S. Kamat, T. Darrell, and J. Malik. Learning humanoid locomotion over chal- lenging terrain. arXiv preprint arXiv:2410.03654, 2024

    2024 arXiv

  9. [17]

    Radosavovic, T

    I. Radosavovic, T. Xiao, B. Zhang, T. Darrell, J. Malik, and K. Sreenath. Real-world humanoid locomotion with reinforcement learning. Science Robotics, 9(89):eadi9579, 2024

    2024

  10. [18]

    Gu, Y .-J

    X. Gu, Y .-J. Wang, X. Zhu, C. Shi, Y . Guo, Y . Liu, and J. Chen. Advancing humanoid loco- motion: Mastering challenging terrains with denoising world model learning. arXiv preprint arXiv:2408.14472, 2024

    2024 arXiv

  11. [19]

    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

  12. [20]

    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

  13. [21]

    Zhang, P

    Q. Zhang, P. Cui, D. Yan, J. Sun, Y . Duan, G. Han, W. Zhao, W. Zhang, Y . Guo, A. Zhang, et al. Whole-body humanoid robot locomotion with human reference. In 2024 IEEE/RSJ In- ternational Conference on Intelligent Robots and Systems (IROS), pages 11225–11231. IEEE, 2024

    2024

  14. [22]

    Huang, S

    R. Huang, S. Zhu, Y . Du, and H. Zhao. Moe-loco: Mixture of experts for multitask locomotion. arXiv preprint arXiv:2503.08564, 2025

    2025 arXiv

  15. [23]

    T. He, W. Xiao, T. Lin, Z. Luo, Z. Xu, Z. Jiang, J. Kautz, C. Liu, G. Shi, X. Wang, et al. Hover: Versatile neural whole-body controller for humanoid robots.arXiv preprint arXiv:2410.21229, 2024

    2024 arXiv

  16. [24]

    T. He, Z. Luo, W. Xiao, C. Zhang, K. Kitani, C. Liu, and G. Shi. Learning human-to-humanoid real-time whole-body teleoperation. arXiv preprint arXiv:2403.04436, 2024

    2024 arXiv

  17. [25]

    Zhang, W

    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

  18. [26]

    Zhuang, S

    Z. Zhuang, S. Yao, and H. Zhao. Humanoid parkour learning. arXiv preprint arXiv:2406.10759, 2024

    2024 arXiv

  19. [27]

    Y . Xue, W. Dong, M. Liu, W. Zhang, and J. Pang. A unified and general humanoid whole-body controller for fine-grained locomotion. arXiv preprint arXiv:2502.03206, 2025

    2025 arXiv

  20. [28]

    X. He, R. Dong, Z. Chen, and S. Gupta. Learning getting-up policies for real-world humanoid robots. arXiv preprint arXiv:2502.12152, 2025

    2025 arXiv

  21. [29]

    Huang, J

    T. Huang, J. Ren, H. Wang, Z. Wang, Q. Ben, M. Wen, X. Chen, J. Li, and J. Pang. Learning humanoid standing-up control across diverse postures.arXiv preprint arXiv:2502.08378, 2025. 10

    2025 arXiv

  22. [30]

    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) , 40(4): 1–20, 2021

    2021

  23. [31]

    Tessler, Y

    C. Tessler, Y . Kasten, Y . Guo, S. Mannor, G. Chechik, and X. B. Peng. Calm: Conditional ad- versarial latent models for directable virtual characters. In ACM SIGGRAPH 2023 Conference Proceedings, pages 1–9, 2023

    2023

  24. [32]

    Hassan, Y

    M. Hassan, Y . Guo, T. Wang, M. Black, S. Fidler, and X. B. Peng. Synthesizing physical character-scene interactions. In ACM SIGGRAPH 2023 Conference Proceedings, pages 1–9, 2023

    2023

  25. [33]

    Y . Yuan, V . Makoviychuk, Y . Guo, S. Fidler, X. Peng, and K. Fatahalian. Learning physically simulated tennis skills from broadcast videos. ACM Trans. Graph, 42(4), 2023

    2023

  26. [34]

    C. Lu, X. Cheng, J. Li, S. Yang, M. Ji, C. Yuan, G. Yang, S. Yi, and X. Wang. Mobile- television: Predictive motion priors for humanoid whole-body control. arXiv preprint arXiv:2412.07773, 2024

    2024 arXiv

  27. [35]

    Serifi, R

    A. Serifi, R. Grandia, E. Knoop, M. Gross, and M. B ¨acher. Vmp: Versatile motion priors for robustly tracking motion on physical characters. In Computer Graphics Forum, page e15175. Wiley Online Library, 2024

    2024

  28. [36]

    J. Mao, S. Zhao, S. Song, T. Shi, J. Ye, M. Zhang, H. Geng, J. Malik, V . Guizilini, and Y . Wang. Learning from massive human videos for universal humanoid pose control. arXiv preprint arXiv:2412.14172, 2024

    2024 arXiv

  29. [37]

    F. Liu, Z. Gu, Y . Cai, Z. Zhou, S. Zhao, H. Jung, S. Ha, Y . Chen, D. Xu, and Y . Zhao. Opt2skill: Imitating dynamically-feasible whole-body trajectories for versatile humanoid loco- manipulation. arXiv preprint arXiv:2409.20514, 2024

    2024

  30. [38]

    Schulman, F

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

  31. [39]

    S. Ross, G. Gordon, and D. Bagnell. A reduction of imitation learning and structured predic- tion to no-regret online learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics, pages 627–635. JMLR Workshop and Conference Pro-...

    2011

  32. [40]

    F. G. Harvey, M. Yurick, D. Nowrouzezahrai, and C. Pal. Robust motion in-betweening. ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH), 39(4), 2020

    2020

  33. [41]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 2012

    2012

  34. [42]

    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), pages 3803–3810. IEEE, 2018

    2018

  35. [43]

    Rudin, D

    N. Rudin, D. Hoeller, P. Reist, and M. Hutter. Learning to walk in minutes using massively parallel deep reinforcement learning. InConference on Robot Learning, pages 91–100. PMLR, 2022

    2022

  36. [44]

    Makoviychuk, L

    V . Makoviychuk, L. Wawrzyniak, Y . Guo, M. Lu, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa, et al. Isaac gym: High performance gpu-based physics simulation for robot learning. arXiv preprint arXiv:2108.10470, 2021

    2021 arXiv

  37. [45]

    Unitree g1 – humanoid agent ai avatar, 2025

    Unitree Robotics. Unitree g1 – humanoid agent ai avatar, 2025. URL https://www. unitree.com/g1. 11

    2025

  38. [46]

    Tevet, S

    G. Tevet, S. Raab, B. Gordon, Y . Shafir, D. Cohen-Or, and A. H. Bermano. Human motion diffusion model. arXiv preprint arXiv:2209.14916, 2022

    2022 arXiv

  39. [47]

    Todorov, T

    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 , pages 5026–

    2012

  40. [48]

    R. S. Sutton, A. G. Barto, et al. Reinforcement learning: An introduction . MIT press Cam- bridge, 1998. 12 Appendix 6.1 Goal-Conditioned Reinforcement Learning In this work, motion tracking problem is defined as a goal-conditioned RL problem where given the goal, the agent in...

    1998

  41. [5033]

    doi:10.1109/IROS.2012.6386109

    IEEE, 2012. doi:10.1109/IROS.2012.6386109

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.