Pith. sign in

REVIEW 3 major objections 6 minor 26 references

JAEGER: Dual-Level Humanoid Whole-Body Controller

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Splitting a humanoid's upper and lower body into two coordinated controllers cuts whole-body tracking error below single-controller baselines in simulation and on real hardware.

desk verdict A genuinely different dual-level controller for humanoid whole-body control, with solid internal ablations, but the SOTA comparison is confounded by reward engineering and the real-world claim is unquantified. read the letter →

arxiv 2505.06584 v2 pith:EX7MMMEH submitted 2025-05-10 cs.RO cs.AI

classification cs.ROcs.AI
keywords whole-bodycontrolhumanoidrobotsdual-levelcontrollermulti-agentreinforcementlearningmotionretargetingcurriculumposetrackingsim-to-realtransfer
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 proposes JAEGER, a whole-body controller for adult-sized humanoid robots that splits control into two independent policies—one for the lower body, one for the upper body—instead of training one network to move all joints. The authors argue this separation reduces mutual interference between balance and expressive arm/head motion, shrinks the effective action space for each learner, and makes the system more fault-tolerant. They also introduce an MLP-based retargeting step that converts human motion-capture poses to humanoid joint targets faster and smoother than optimization-based inverse kinematics, and a curriculum that starts with supervised imitation and then refines with reinforcement learning. In simulation on two adult-sized platforms, JAEGER reports lower tracking errors than three open-source baselines in both root-velocity and pose-tracking modes, and the controller is demonstrated on real hardware. A sympathetic reader would take the core claim to be that decoupled, coordinated agents plus supervised initialization form a better inductive bias for whole-body humanoid control than a single monolithic policy.

What carries the argument

The load-bearing object is the dual-level controller, a Dec-POMDP with two agents: a lower-body policy implemented as a gated Transformer-XL taking 15 timesteps of observation history, and an upper-body policy implemented as a 3-layer MLP taking 3 timesteps. The curriculum's supervised initialization uses a per-step squared-error loss $\mathcal{L}_{\text{SL}}(\pi_{\text{upper}}) = \|\pi_{\text{upper}}(x_{t,3}) - g^{\text{upper}}_t\|$ to make the upper body imitate the reference pose directly, and a KL term $\mathcal{L}_{\text{KL}}(\pi_{\text{lower}}) = D_{\text{KL}}(\pi_{\text{lower}}(\cdot|x_{t,15})\,\|\,\pi_{\text{base}}(\cdot|x_{t,15}))$ to keep the lower body near a pretrained base policy. These mechanisms, plus a three-layer MLP retargeting network trained on optimized human-to-humanoid pose pairs, carry the argument: the dual-level structure isolates interference, the supervised initialization prevents the upper body from becoming over-conservative and the lower body from forgetting balance, and the retargeting MLP supplies smooth, high-rate reference targets.

What would settle it

Retrain a single-controller baseline (a monolithic transformer with the same observations, actions, and reward as JAEGER) using JAEGER's supervised initialization, curriculum schedule, and domain randomization on the same motion dataset, then measure the same tracking metrics on the same evaluation set; if its errors match or beat JAEGER's, the dual-level architecture is not what drives the improvement.

Watch

Extended reading notes

Core claim

JAEGER's central claim is that whole-body humanoid control should be formulated as a two-agent Dec-POMDP (decentralized partially observable Markov decision process): a Transformer-based lower-body policy handles balance, root velocity, and lower-limb pose, while a lightweight MLP upper-body policy independently tracks reference arm and torso angles. The two policies share observations and rewards, and their actions are concatenated into target joint positions for PD controllers. The paper reports that this dual-level controller, trained with supervised initialization followed by reinforcement learning, achieves significantly lower mean absolute tracking errors than single-controller baselines—for example on the 19-DoF platform in root-based mode, upper-body joint error of about $0.109$ rad versus at least $0.169$ rad for one baseline, and angular velocity error of $0.094$ rad/s versus $0.31$ rad/s for another. In pose-based mode, its upper- and lower-body joint errors are roughly 30% to 60% of the baselines' errors. The paper also claims the retargeting MLP runs above 1 kHz with smoother output than optimization-based inverse kinematics and lower error than directly copying human angles.

Load-bearing premise

The load-bearing premise of the headline comparison is that the external baselines were evaluated under conditions that make JAEGER's advantage attributable to its dual-level architecture; because the baselines were not retrained with JAEGER's reward function and curriculum, the reported gap could be caused by reward engineering rather than by the controller split.

Editorial extensions

If this is right

  • If JAEGER's claims hold, a humanoid controller can serve both coarse root-velocity commands and fine pose-tracking commands in one system, without one capability degrading the other.
  • The dual-level design lets each half of the body use its own network architecture—a long-horizon transformer below, a cheap MLP above—so computational cost concentrates where balance decisions are hard.
  • Separating the controllers improves fault tolerance: if the upper-body policy fails or is disabled, the lower-body policy can keep the robot balanced and moving, because the two were trained independently.
  • The retargeting MLP's sub-millisecond per-frame cost makes real-time teleoperation practical, converting motion-capture input to joint targets at control rates above 1 kHz.
  • Supervised initialization before reinforcement learning appears to prevent the upper body from freezing into an over-conservative policy and keeps the lower body from drifting off its balance behavior during joint training.

Reading between the lines

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

  • Editorial inference: the cleanest way to separate architecture from reward is to retrain a monolithic baseline with JAEGER's exact reward, curriculum, and randomization; if that baseline closes the gap, the dual-level structure is not the active ingredient.
  • Editorial inference: because the two policies are independent, a user could swap in a stronger upper-body policy or a different lower-body architecture without retraining the other half—an upgrade path the paper does not explicitly test.
  • Editorial inference: JAEGER omits keypoint-position commands; extending the upper-body policy to accept keypoint targets could combine the dual-level separation with the global-pose expressiveness of keypoint-based methods.
  • Editorial inference: the real-world stability reported here likely depends on the hand-tuned penalty terms in the reward; a systematic ablation of those penalties would show how much of the sim-to-real success is architectural rather than reward engineering.
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 / 6 minor

Summary. This manuscript presents JAEGER, a whole-body controller for adult-size humanoids that decouples upper- and lower-body control into two policies (a lightweight MLP for the upper body and a Gated Transformer-XL for the lower body), supports both root-velocity and pose-tracking command modes, and uses an MLP retargeting network plus a curriculum that starts with supervised initialization and continues with PPO. Experiments in IsaacGym on Unitree H1 and H1-2 compare against HumanPlus, ExBody, and OmniH2O on a held-out AMASS subset, reporting lower tracking errors; ablations isolate the dual-level architecture and the supervised-initialization curriculum. The paper also reports qualitative real-robot demonstrations on H1-2 and evaluates retargeting speed, accuracy, and smoothness against H2O and HumanPlus.

Significance. If the simulation results are taken at face value, the paper makes a useful empirical contribution: decoupling upper- and lower-body control improves upper-body pose tracking without a consistent penalty on lower-body tracking, the supervised-initialization curriculum stabilizes joint-angle tracking, and the MLP retargeter provides large speed gains with accuracy comparable to an optimization-based IK method. The internal ablations (Tables 4 and 5) are the strongest evidence because JAEGER, JAEGER Single, and JAEGER w.o. Init share the same reward and therefore isolate the architectural and curriculum effects. The retargeting comparison against H2O and HumanPlus is also a genuinely external benchmark. The main weaknesses are that the external SOTA comparison does not isolate the architecture from reward engineering, and the abstract's real-world superiority claim is not supported by quantitative hardware metrics. These gaps currently prevent the paper from fully supporting its strongest claims.

major comments (3)
  1. [Section 4.2, Tables 2 and 3] The headline comparison against HumanPlus, ExBody, and OmniH2O does not control for reward function or training curriculum. JAEGER uses a heavily hand-tuned reward (Table 1 and the appendix tables with weights spanning several orders of magnitude) and a multi-stage supervised-initialization curriculum, while the baselines were trained with their own reward designs and curricula and are not retrained under JAEGER's reward. Consequently the reported gaps (e.g., E_upper-j 0.1093 vs 0.2187 on H1 in Table 2) could reflect reward engineering rather than the dual-level controller itself. The internal ablations support the architectural claim, but they do not resolve this confound in the external comparison. The manuscript should either retrain the baselines with a matched reward or explicitly reframe the external comparison as a system-level comparison and soften the causal language. Section 6's own admission that the framework 'requires extensive reward engineering' makes this more than a stylistic point.
  2. [Abstract and Section 4.2 vs. Section A.1] The abstract claims superiority over state-of-the-art methods 'in both simulation and real environments,' and Section 4.2 repeats the simulation-side claim, but Section A.1 provides only deployment details (50 Hz policy inference, LCM communication, 500 Hz low-level interface) and qualitative video demonstrations. No quantitative hardware metrics are reported, such as tracking error, command-following success rate, or fall counts. The real-environment superiority claim is therefore currently unsupported. Please add quantitative real-world evaluations or revise the abstract and conclusion to claim only simulation superiority with qualitative real-world feasibility.
  3. [Section 4.2 and Tables 2-3] The text states that JAEGER 'significantly outperforms' the baselines, but the only evidence is mean error with standard deviations across five seeds; no significance test is reported, and in some cells the intervals overlap (e.g., H1 root-linear-velocity in Table 2: 0.1504 +/- 0.0264 vs. HumanPlus 0.1696 +/- 0.0013). Given that the central claim is a comparative one, a paired or repeated-seed significance test, or at least a statement of whether the same evaluation trajectories and command samples are used across methods, would make the comparison more rigorous.
minor comments (6)
  1. [Section 3.1] The heading 'Regtargeting Network' contains a typo and should read 'Retargeting Network.'
  2. [Appendix C, Tables 11 and 12] The captions for Tables 11 and 12 appear to be swapped: Table 11 is labeled 'Reward components and weights' but contains what look like PPO hyperparameters, while Table 12 is labeled 'Hyperparameters for RL Controller' but contains reward terms. Please check the numbering and make the captions match the content.
  3. [Section 3.3, final paragraph] The sentence 'After training the models for both modes separately, we distill them into a single network via imitation learning' is inconsistent with the dual-level description elsewhere, which presents two independent policies. Clarify whether the deployed controller is one distilled network or two networks, and specify what is distilled (the two command modes, or the upper/lower policies).
  4. [Section 4.2, Table 3] On H1-2, JAEGER's root-r (0.0398) and root-y (0.8051) are worse than OmniH2O's (0.0314 and 0.6540), so the claim that JAEGER achieves 'comparable accuracy in root tracking' should be made metric-specific rather than presented as a blanket statement.
  5. [Section 4.1] For OmniH2O, the paper says reference keypoint positions are converted to joint angles 'to ensure a fair comparison,' but the conversion procedure is not described. Please specify the conversion method so readers can judge whether it favors any particular method.
  6. [General] The paper does not announce a code or model release despite providing a project URL. Since the baselines are openly available and the evaluation is on a held-out AMASS subset, a public release of JAEGER's code and checkpoints would materially improve reproducibility and the value of the comparison.

Circularity Check

1 steps flagged · score 2.0 of 10

One in-sample retargeting evaluation; the central dual-level controller claims are empirical and not circular.

  1. fitted input called prediction [Section 3.1 (Regtargeting Network); Section 4.3 (Retargeting Analysis), Tables 6-7]
    "We first we adopt a two-stage optimization approach as H2O [1] to process the AMASS dataset [11], generating a large set of human and humanoid pose pairs. ... Then, a lightweight three-layer MLP is employed to learn the mapping between them. ... Since retargeting does not need to consider whether the pose is feasible for the humanoid, we use the entire AMASS dataset as the evaluation set."

    The MLP retargeter is a supervised regressor fitted to H2O optimization outputs computed on AMASS, and the accuracy comparison in Table 6 is then run on the entire AMASS dataset with no stated train/test split. The reported retargeting error is therefore an in-sample reconstruction of the training labels, so the claimed superiority over H2O is largely a fitting artifact and data-overlap effect rather than an out-of-sample prediction. The smoothness comparison is similarly a direct consequence of using a continuous MLP on filtered data rather than a derived result. This circularity is confined to the retargeting sub-claim (Contribution 1) and does not infect the central dual-level controller comparison.

full rationale

The central whole-body controller claim is an empirical systems result, not a derivation: JAEGER trains two body-decoupled policies in a Dec-POMDP with supervised initialization followed by PPO, and it is compared with open-sourced baselines (HumanPlus, ExBody, OmniH2O) on a designated evaluation subset of AMASS. The internal ablations (JAEGER Single and JAEGER w.o. Init) share JAEGER's reward, so the architectural and curriculum conclusions are not definitionally forced by the reward. No load-bearing self-citation chain exists; the cited priors are from external groups. The external baseline comparison is confounded by reward and curriculum differences, and the paper's Limitations concede that 'the current framework requires extensive reward engineering,' but that is an identification/fairness concern rather than a circularity under the rubric. The one genuinely circular element is the retargeting evaluation: the MLP is trained on H2O-generated AMASS labels and then evaluated on the same AMASS source without a stated split, so its reported accuracy is a training-set fit rather than a prediction. This affects the retargeting sub-claim only, so the overall circularity score is low.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The central claim rests primarily on hand-tuned reward weights and a set of standard RL and sim-to-real assumptions. No new physical entities are introduced. The reward weights in Tables 1 and 11 are chosen by hand and are load-bearing: the paper's Limitations state that 'the current framework requires extensive reward engineering.' The assumptions listed are domain assumptions common to the legged-RL literature.

free parameters (9)
  • Upper-body DoF position reward weight = 20.0
    Hand-chosen weight in Table 1 for tracking upper-body joint angles; central to the upper-body tracking performance claim.
  • Lower-body DoF position reward weight = 100.0
    Hand-chosen weight in Table 1 for lower-body joint angle tracking; central to pose-based mode performance.
  • Feet height reward weight = 100.0
    Hand-chosen weight in Table 1 for foot height tracking; tuned for stability and sim-to-real.
  • Linear and angular velocity reward weights = 20.0 / 20.0
    Hand-chosen weights in Table 1 for root velocity tracking in both command modes.
  • Roll and pitch orientation reward weight = 1.0
    Hand-chosen weight in Table 1 for root orientation tracking; affects balance behavior.
  • Feet air time reward weight = 10000
    Hand-chosen weight in Table 11 (appendix) encouraging periodic foot liftoff; copied from prior work [26] but still a free parameter.
  • Lower-body action rate penalty weight = -20
    Hand-chosen penalty in Table 11 to smooth lower-body actions and reduce jitter.
  • Pose amplitude scaling in curriculum = 0.5
    During supervised initialization, the reference pose amplitude is halved to simplify learning; the factor 0.5 is chosen by hand (Section 3.3).
  • Other penalty and regularization weights = various
    Table 11 lists additional hand-tuned weights (torque limits -10, DoF position/velocity limits -10, acceleration -3e-8, upper-body action rate -5, action smoothness -10, torque -0.0001, feet contact force -10, stumble -0.00125, orientation -200). These are chosen by hand and affect the trained policy.
assumptions (5)
  • domain assumption Human motion capture data from AMASS, after retargeting and filtering, provides a sufficient and valid training distribution for humanoid whole-body control.
    The entire training pipeline relies on AMASS [11] as the source of reference motions (Sections 3.1 and 4.1). If the retargeted poses are not dynamically feasible for the target robots, the learned policies would not transfer.
  • domain assumption IsaacGym simulation with the listed domain randomization is a faithful proxy for real-robot dynamics, so a policy trained in simulation transfers to the real Unitree H1/H1-2 without adaptation.
    Section 4.1 and Appendix C (domain randomization table). The authors deploy to real hardware directly; the paper's Limitations concede 'the current framework requires extensive reward engineering to ... address the sim-to-real challenge,' weakening but not removing this assumption.
  • domain assumption The low-level PD controller at 500 Hz can execute the policy's target joint positions with negligible tracking error.
    Appendix A.1 states the policy runs at 50 Hz and the low-level interface at 500 Hz; the whole pipeline assumes the PD controller realizes the commanded positions.
  • standard math PPO's convergence behavior and the Gated Transformer-XL architecture will find a good policy when trained with the given rewards and curriculum.
    Standard RL machinery; the paper does not provide any convergence proof for the dual-policy curriculum, it is an empirical assumption.
  • domain assumption The two-agent Dec-POMDP with shared observations and shared rewards is an appropriate model for the whole-body control problem.
    Section 3.2 formulates the problem as a Dec-POMDP with two agents; the policies are trained jointly with PPO, not with a multi-agent algorithm, so the formulation is a modeling choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of JAEGER: Dual-Level Humanoid Whole-Body Controller." pith.science (2026). https://pith.science/paper/EX7MMMEH

@misc{pith2026250506584,
  author       = {Pith},
  title        = {Pith review of: JAEGER: Dual-Level Humanoid Whole-Body Controller},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EX7MMMEH}},
  note         = {Machine review of arXiv:2505.06584}
}
read the original abstract

This paper presents JAEGER, a dual-level whole-body controller for humanoid robots that addresses the challenges of training a more robust and versatile policy. Unlike traditional single-controller approaches, JAEGER separates the control of the upper and lower bodies into two independent controllers, so that they can better focus on their distinct tasks. This separation alleviates the dimensionality curse and improves fault tolerance. JAEGER supports both root velocity tracking (coarse-grained control) and local joint angle tracking (fine-grained control), enabling versatile and stable movements. To train the controller, we utilize a human motion dataset (AMASS), retargeting human poses to humanoid poses through an efficient retargeting network, and employ a curriculum learning approach. This method performs supervised learning for initialization, followed by reinforcement learning for further exploration. We conduct our experiments on two humanoid platforms and demonstrate the superiority of our approach against state-of-the-art methods in both simulation and real environments.

Figures

Figures reproduced from arXiv: 2505.06584 by the authors.

Figure 1
Figure 1. Some real-world demonstrations of JAEGER deployed on the H1-2. For the root-based [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The framework of JAEGER. The left shows the retargeting network, which uses an MLP [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 15 canonical work pages

  1. [1]

    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

  2. [2]

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

  3. [3]

    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. 2024. 9

  4. [4]

    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

  5. [5]

    T. He, W. Xiao, T. Lin, Z. Luo, Z. Xu, Z. Jiang, C. Liu, G. Shi, X. Wang, L. Fan, and Y . Zhu. Hover: Versatile neural whole-body controller for humanoid robots.arXiv preprint arXiv:2410.21229, 2024

  6. [6]

    Cheng, Y

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

  7. [7]

    R. Lowe, Y . I. Wu, A. Tamar, J. Harb, O. Pieter Abbeel, and I. Mordatch. Multi-agent actor- critic for mixed cooperative-competitive environments.Advances in neural information pro- cessing systems, 30, 2017

  8. [8]

    Rashid, M

    T. Rashid, M. Samvelyan, C. S. De Witt, G. Farquhar, J. Foerster, and S. Whiteson. Monotonic value function factorisation for deep multi-agent reinforcement learning.Journal of Machine Learning Research, 21(178):1–51, 2020

Show all 26 references
  1. [9]

    C. Yu, A. Velu, E. Vinitsky, J. Gao, Y . Wang, A. Bayen, and Y . Wu. The surprising effectiveness of ppo in cooperative multi-agent games.Advances in Neural Information Processing Systems, 35:24611–24624, 2022

  2. [10]

    Gronauer and K

    S. Gronauer and K. Diepold. Multi-agent deep reinforcement learning: a survey.Artificial Intelligence Review, 55(2):895–943, 2022

  3. [11]

    Mahmood, N

    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, pages 5442–5451, Oct. 2019

  4. [12]

    Zhuang, S

    Z. Zhuang, S. Yao, and H. Zhao. Humanoid parkour learning. In8th Annual Conference on Robot Learning, 2024. URLhttps://openreview.net/forum?id=fs7ia3FqUM

  5. [13]

    J. Long, J. Ren, M. Shi, Z. Wang, T. Huang, P. Luo, and J. Pang. Learning humanoid locomo- tion with perceptive internal model.arXiv preprint arXiv:2411.14386, 2024

  6. [14]

    Agarwal, A

    A. Agarwal, A. Kumar, J. Malik, and D. Pathak. Legged locomotion in challenging terrains using egocentric vision. InConference on robot learning, pages 403–415. PMLR, 2023

  7. [15]

    Radosavovic, T

    I. Radosavovic, T. Xiao, B. Zhang, T. Darrell, J. Malik, and K. Sreenath. Real-world humanoid locomotion with reinforcement learning, 2023. URLhttps://arxiv.org/abs/2303 .03381

  8. [16]

    Ho and S

    J. Ho and S. Ermon. Generative adversarial imitation learning.Advances in neural information processing systems, 29, 2016

  9. [17]

    Radosavovic, J

    I. Radosavovic, J. Rajasegaran, B. Shi, B. Zhang, S. Kamat, K. Sreenath, T. Darrell, and J. Ma- lik. Humanoid locomotion as next token prediction. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024. URLhttps://openreview.net/f orum?id=GrMczQGTlA

  10. [18]

    F. A. Oliehoek, C. Amato, et al.A concise introduction to decentralized POMDPs, volume 1. Springer, 2016

  11. [19]

    A. Vaswani. Attention is all you need.Advances in Neural Information Processing Systems, 2017

  12. [20]

    Schmidhuber, S

    J. Schmidhuber, S. Hochreiter, et al. Long short-term memory.Neural Comput, 9(8):1735– 1780, 1997. 10

  13. [21]

    Parisotto, F

    E. Parisotto, F. Song, J. Rae, R. Pascanu, C. Gulcehre, S. Jayakumar, M. Jaderberg, R. L. Kaufman, A. Clark, S. Noury, et al. Stabilizing transformers for reinforcement learning. In International conference on machine learning, pages 7487–7498. PMLR, 2020

  14. [22]

    Schulman, F

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

  15. [23]

    Makoviychuk, L

    V . Makoviychuk, L. Wawrzyniak, Y . Guo, M. Lu, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa, and G. State. Isaac gym: High performance GPU based physics simu- lation for robot learning. InThirty-fifth Conference on Neural Information Processing Systems D...

  16. [24]

    A. S. Huang, E. Olson, and D. C. Moore. Lcm: Lightweight communications and marshalling. In2010 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 4057– 4062, 2010. doi:10.1109/IROS.2010.5649358

  17. [25]

    Gu, Y .-J

    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

  18. [26]

    Rudin, D

    N. Rudin, D. Hoeller, P. Reist, and M. Hutter. Learning to walk in minutes using massively parallel deep reinforcement learning. In5th Annual Conference on Robot Learning, 2021. URLhttps://openreview.net/forum?id=wK2fDDJ5VcF. 11 A Environment Details A.1 Real-world Deployment ...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.