Pith. sign in

REVIEW 3 major objections 5 minor 32 references

ADP: Adversarial Dynamics Priors for Physically Grounded Humanoid Locomotion

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Regularizing momentum and contact dynamics rather than joint poses makes humanoid policies substantially more push-resistant, the paper argues.

desk verdict A plausible and well-controlled demonstration that dynamics-feature adversarial priors improve push recovery in humanoids, but the headline robustness is anchored to a nominal gait schedule and the single-seed numbers need verification. read the letter →

arxiv 2607.03454 v2 pith:YAXYPWSO submitted 2026-07-03 cs.RO cs.LG

Seokju Lee , Jeongtae Lee , Jeonghyeok Lim , Jeonguk Kang , Byungwook Lee , Seungho Han , Keun Ha Choi , Dongil Park
show 1 more author
Kyung-Soo Kim
This is my paper
classification cs.ROcs.LG
keywords adversarialdynamicspriorshumanoidlocomotionperturbationrecoverycentroidalmomentumcontactforcestrajectoryoptimizationimitationpush
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 argues that a humanoid locomotion policy recovers from external pushes much better when the adversarial prior guiding its behavior is defined over dynamics features — center-of-mass velocity, centroidal angular momentum, normalized foot contact forces, and binary contact indicators over a short time window — instead of the joint-level kinematic features used by existing motion priors. The authors build a reference set by optimizing single-rigid-body trajectories offline, then train a discriminator to distinguish policy-generated windows from reference windows, rewarding the policy for staying close to the reference distribution after an impulse. On the simulated G1 humanoid they report that this 'adversarial dynamics prior' raises the 80%-success impulse threshold by 16.7% over the kinematic AMP baseline and cuts recovery time and velocity error roughly in half, with qualitative hardware confirmation. The significance is that push recovery hinges on momentum regulation and contact timing, quantities that kinematic features barely expose, so a dynamics-based prior gives the policy a more direct physical target.

What carries the argument

The load-bearing piece is the adversarial dynamics prior over a temporal window: a discriminator that classifies eight-timestep windows of normalized dynamics features (heading-frame CoM velocity, centroidal angular momentum and its planar rate, weight-normalized foot contact forces, and binary contact indicators, conditioned on the velocity command and tracking error) as reference-like or policy-like, and a PPO policy rewarded for windows the discriminator scores as reference-like. The reference distribution is generated offline by single-rigid-body trajectory optimization, so the prior encodes physically consistent momentum–force–contact relationships rather than a particular kinematic sty

What would settle it

Run ADP under a push whose optimal recovery requires a contact-timing change not in the nominal gait schedule (e.g., an extra step or a delayed swing phase) and observe whether the policy falls precisely when it must leave the reference contact schedule. Alternatively, retrain ADP with a reference set that includes perturbed windows and check whether J80 materially improves; if it does, the reference support is the limiting constraint.

Watch

Extended reading notes

Core claim

The central claim is that replacing the kinematic motion-style prior in adversarial imitation with a dynamics-feature prior — CoM velocity, centroidal angular momentum, normalized contact forces, and binary contact indicators over a temporal window — yields substantially better post-push recovery without requiring reference poses, phase, or end-effector tracking. The authors construct the reference set from single-rigid-body trajectory optimization, train a discriminator on windowed dynamics features, and show that the resulting reward pulls policy rollouts back toward the reference support after a push. Quantitatively, they report a J80 of 115.5 N·s, a 91.4% direction-averaged success rate,

Load-bearing premise

The TO reference set is built only for nominal locomotion modes under fixed commands and contains no perturbed trajectories; the recovery mechanism assumes the desired post-push behavior is to be pulled back into the nominal dynamics-feature support, so if strong perturbations require temporarily leaving that support (e.g., an extra step or a yielding posture), the discriminator reward could suppress the very actions needed to recover.

Editorial extensions

If this is right

  • Policies trained with ADP withstand impulses roughly 17% stronger than those trained with a kinematic motion prior before crossing an 80% success threshold.
  • Direction-averaged recovery time after a 3 m/s impulse push drops from 4.76 s to 2.48 s, and velocity tracking error from 1.30 to 0.84 m/s, relative to AMP.
  • Dynamics features expose a perturbation within 20 ms (6-fold deviation) versus 160 ms for kinematic features, so they can serve as an early detection signal for balance failures.
  • Contact timing and centroidal momentum are the most critical feature groups; removing the contact indicator or momentum degrades success from 91.4% to 30.5% or 45.3%, respectively.
  • No reference poses, phase variables, or end-effector tracking are needed, so the prior can be built purely from trajectory optimization and used for robots without motion-capture data.

Reading between the lines

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

  • Because the TO reference set contains only nominal, unperturbed trajectories, the recovered behavior is defined as a pull-back to nominal dynamics. If large pushes require a non-nominal step or contact schedule, the discriminator reward may suppress the correct recovery action — a testable boundary of the method.
  • The binary contact indicator is effectively the prescribed gait schedule; the ablation suggests the discriminator leans heavily on gait phase. This raises the question of whether the prior is learning a dynamics distribution or a timing schedule, and whether the benefit persists under pushes that demand a phase change.
  • The paper compares against an AMP baseline built from the same TO-derived reference, not a high-quality mocap-based prior. A richer kinematic reference might shrink the gap, though dynamics features would likely still show earlier transient exposure.
  • A direct extension would be to add a small number of perturbed reference windows (e.g., from push-recovery trajectory optimization) and test whether J80 increases further; if it does, the reference support is the limiting factor.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. This paper proposes Adversarial Dynamics Priors (ADP) for humanoid locomotion control. The method replaces the kinematic features used in Adversarial Motion Priors with selected dynamics features—CoM velocity, centroidal angular momentum and its rate, normalized foot contact forces, and binary contact indicators—extracted from trajectory-optimization (TO) solutions of a single-rigid-body dynamics model. A discriminator is trained on temporal windows of these features to distinguish TO-derived reference windows from policy-generated windows, and the resulting adversarial reward is added to the task reward during PPO training. In simulation on the Unitree G1, the authors compare ADP against Vanilla RL, AMP, and a cluster-based Dynamics Reward baseline. The reported results show that ADP improves the 80%-success impulse threshold J80 by 16.7%, reduces direction-averaged recovery time by 47.9% and velocity error by 35.4% relative to AMP. Ablations show the importance of the temporal window length and of individual features, particularly the binary contact indicator. A representation-sensitivity analysis indicates that the dynamics representation responds earlier and more strongly to pushes than the joint-level kinematic representation. The paper also includes qualitative hardware demonstrations.

Significance. If the empirical results are robust, the contribution is significant for legged locomotion: it demonstrates that a TO-derived dynamics-feature prior can be more effective than a kinematic motion prior for perturbation recovery, and the controlled comparison using the same reference source for AMP and ADP is a strong experimental design. The representation-sensitivity analysis is a useful diagnostic. However, the central empirical claims currently lack statistical backing (no training seeds), and the perturbation envelope tested is narrow relative to the claim of general recovery. These issues must be addressed before the paper can be accepted.

major comments (3)
  1. [Sec. IV-A / Table I] The central quantitative comparisons are presented as single numbers with no error bars or number of training seeds. Because the methods are stochastic and the claimed gains (e.g., J80 115.5 vs 99.0 N·s, recovery time 2.48 vs 4.76 s) are large, the lack of variance estimates is a load-bearing gap. The IQR bands in Fig. 3 are across 32 evaluation environments from one training run and do not quantify training stochasticity. Please provide means and standard deviations over multiple seeds (e.g., 3–5), and ideally a bootstrap significance test for success rate and J80. Without this, the reader cannot judge whether the improvements are consistent.
  2. [Sec. III-B, Eq. (3); Sec. III-C, Eq. (4); Table III] The reference dataset D_dyn is generated from TO under fixed commands and a per-motion prescribed contact schedule {c_t^L, c_t^R}. The ablation in Table III shows that removing the binary contact indicator from the dynamics feature drops direction-averaged success from 91.4% to 30.5%, demonstrating that the ADP reward is critically dependent on enforcing the nominal contact schedule. The evaluation sweeps impulses only up to Δv = 4.5 m/s and does not test pushes at different gait phases or forces that would require temporarily leaving the nominal support (extra steps, altered contact timing). Consequently, the paper's central claim that ADP 'encourages policy rollouts to remain close to the reference support, even after perturbations' is only demonstrated within an envelope where the nominal contact schedule remains a viable recovery target. Please either (i) test perturbations that forc
  3. [Sec. IV-C, Fig. 6] The default window length K=8 and the feature set are selected on the basis of the same direction-averaged success metric that is later reported as the headline result. This selection on the test metric can lead to optimistic estimates. The K-sweep shows high sensitivity (success 28.1% at K=1, 49.2% at K=4, 91.4% at K=8), so the final result hinges on this choice. Please provide evidence that the design choices generalize, e.g., by selecting K and features on a validation set of different push directions or commands, or by reporting the full ablation across a range of K values on multiple seeds. At minimum, state the selection protocol explicitly.
minor comments (5)
  1. [Sec. IV-A] The value of the robot mass is omitted; since J80 is defined as mΔv, please report the mass of the G1 model used in the experiments.
  2. [Eq. (11) and Sec. IV] The cluster-based Dynamics Reward baseline requires hyperparameters (τ, number of clusters, command-space bandwidth) that are not specified, making the baseline difficult to reproduce.
  3. [Reproducibility] The code release is promised 'on the paper website' but is not available at submission; for reproducibility, please provide code or a detailed appendix with the exact observation, architecture, and training hyperparameters.
  4. [Fig. 6] The normalization of 'lower-is-better metrics' is described only in the caption; please explain the normalization in the text so the reader can interpret the y-axis.
  5. [General] The phrase 'direction-averaged' is sometimes hyphenated inconsistently; also check the header date '16 Jul 2026' for correctness.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the TO-derived dynamics prior is an independent regularization target, and the headline metrics are task-level, not defined by the discriminator.

full rationale

The claimed derivation chain is: SRBD-based TO (Eq. 3) produces reference trajectories from which a dynamics-feature dataset D_dyn is extracted (Eq. 4); a discriminator is trained to distinguish reference windows from policy windows (Eq. 9); the policy receives reward r_D (Eq. 10) and is trained with PPO. None of the headline results (success rate, J80, recovery time, velocity error) are defined in terms of r_D or D_dyn; they are based on fall termination conditions, fixed error thresholds, and RMS tracking error. The reference set is generated by an independent optimization objective (tracking commanded CoM velocity, penalizing momentum and forces), not by fitting to the evaluation metrics. The comparison against AMP controls the reference source, and the Dynamics Reward baseline controls the feature set, so the reported improvements are empirical outcomes rather than identities. The only mild concern is that the diagnostic d_dyn in Table III is computed in the same feature space as the dynamics-prior reward, so component-ablation effects on that diagnostic are partly expected; however, the paper explicitly labels this analysis as not a task-level metric and bases its central claims on the task-level metrics, which are external to the reward. No load-bearing self-citations or imported uniqueness theorems were found. The conclusion's note that AMP is compared against a TO-derived rather than mocap-based reference is a scope limitation, not circularity.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central empirical claim rests mainly on simulation evidence and standard RL/adversarial-training practice. The method introduces no new physical entities. Most numerical hyperparameters are unreported, and the key domain assumption—that nominal SRBD-TO trajectories with prescribed gait contact schedules form a valid target for post-push recovery—is not independently validated. The ablation shows the binary contact indicator is the dominant feature, so the hand-chosen gait schedules strongly influence the result.

free parameters (7)
  • Reward weights w_G, w_D, w_R (Eq. 1)
    Balances task reward, dynamics prior reward, and regularization; values not reported, tuned before evaluation.
  • ADP reward scale c_D (Eq. 10)
    Scales discriminator reward; value not reported.
  • Gradient penalty λ_gp (Eq. 9)
    Stabilizes discriminator training; value not reported.
  • Temporal window length K (Eq. 8) = K=8
    Selected in Fig. 6 as best on the direction-averaged success metric that is also the headline result; no held-out validation described.
  • TO objective weights Q_v, Q_L, R_f (Eq. 3)
    Shape the SRBD reference distribution; values not reported.
  • Contact model parameters F_max, μ (Eq. 3)
    Maximum normal force and friction-cone coefficient in the TO; chosen by hand.
  • Prescribed gait contact schedule c_L, c_R
    Per-motion binary contact schedule fixed before optimization; ablation shows contact indicator is the most important feature, so this hand choice strongly influences the results.
assumptions (6)
  • domain assumption SRBD equations (2) adequately capture the dynamics needed for locomotion priors for the G1 humanoid.
    Reference dataset and feature space are built from SRBD; full-body effects such as swing-leg dynamics and joint torque limits are absent.
  • domain assumption Trajectories from SRBD TO with prescribed gait schedules are a valid reference for post-perturbation recovery.
    The reference set contains no perturbed trajectories; recovery is treated as returning to the nominal feature support. This is the load-bearing assumption.
  • domain assumption The least-squares discriminator with gradient penalty (9) provides a stable and informative reward signal.
    Adversarial training stability is assumed; no analysis of mode collapse or reward hacking is provided.
  • domain assumption PPO with the combined reward (1) converges to a policy whose dynamics features remain near the reference support.
    Standard RL convergence assumptions; no guarantees.
  • domain assumption Simulator dynamics, domain randomization, and privileged critic suffice for sim-to-real transfer.
    Hardware results are qualitative; no quantitative transfer metrics.
  • domain assumption The hand-selected feature vector in (4) sufficiently exposes perturbation transients.
    Authors select features; ablation supports it but the set is not derived from a formal criterion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ADP: Adversarial Dynamics Priors for Physically Grounded Humanoid Locomotion." pith.science (2026). https://pith.science/paper/YAXYPWSO

@misc{pith2026260703454,
  author       = {Pith},
  title        = {Pith review of: ADP: Adversarial Dynamics Priors for Physically Grounded Humanoid Locomotion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YAXYPWSO}},
  note         = {Machine review of arXiv:2607.03454}
}
abstract

In this paper, we propose Adversarial Dynamics Priors (ADP) for perturbation-resilient humanoid locomotion control. Existing motion prior-based methods induce natural motion styles by imitating kinematic motion features, but they do not directly regularize dynamics features, such as CoM motion, centroidal momentum, contact forces, and contact states. To address this limitation, we replace kinematic motion-style feature with selected dynamics features extracted from locomotion trajectories as the target of adversarial regularization. To this end, we use trajectory optimization to construct a reference dataset and train a discriminator to evaluate whether policy-induced temporal windows are consistent with the resulting reference distribution. Without explicit motion tracking, ADP encourages policy rollouts to remain close to the reference support, even after perturbations. Experimental results show that, compared with AMP, the strongest baseline in our evaluation, ADP improves the $80\%$-success impulse threshold ($J_{80}$) by $16.7\%$, while reducing direction-averaged recovery time and velocity tracking error by $47.9\%$ and $35.4\%$, respectively.

Figures

Figures reproduced from arXiv: 2607.03454 by the authors.

Figure 1
Figure 1. Dynamics-feature alignment for perturbation recovery. An external [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed Adversarial Dynamics Priors (ADP) framework. (a) SRBD-based trajectory optimization generates reference locomotion [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Time-series perturbation recovery under a lateral push. Red, orange, blue, and green curves denote Vanilla RL, Dynamics Reward, AMP, and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative hardware demonstration of ADP on the real robot [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Normalized reference deviation in dynamics-feature and kinematic [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 3 linked inside Pith

  1. [1]

    Generalizable humanoid manipulation with 3d diffusion policies,

    Y . Ze, Z. Chen, W. Wang, T. Chen, X. He, Y . Yuan, X. B. Peng, and J. Wu, “Generalizable humanoid manipulation with 3d diffusion policies,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 2873–2880, IEEE, 2025

    2025

  2. [2]

    Opt2skill: Imitating dynamically-feasible whole-body trajectories for versatile humanoid loco-manipulation,

    F. Liu, Z. Gu, Y . Cai, Z. Zhou, H. Jung, J. Jang, S. Zhao, S. Ha, Y . Chen, D. Xu,et al., “Opt2skill: Imitating dynamically-feasible whole-body trajectories for versatile humanoid loco-manipulation,” IEEE Robotics and Automation Letters, 2025

    2025

  3. [3]

    Demohlm: From one demonstration to generalizable humanoid loco-manipulation,

    Y . Fu, F. Xie, C. Xu, J. Xiong, H. Yuan, and Z. Lu, “Demohlm: From one demonstration to generalizable humanoid loco-manipulation,” IEEE Robotics and Automation Letters, 2026

    2026

  4. [4]

    Learning quadrupedal locomotion over challenging terrain,

    J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter, “Learning quadrupedal locomotion over challenging terrain,”Science robotics, vol. 5, no. 47, p. eabc5986, 2020

    2020

  5. [5]

    Learning robust perceptive locomotion for quadrupedal robots in the wild,

    T. Miki, J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter, “Learning robust perceptive locomotion for quadrupedal robots in the wild,”Science robotics, vol. 7, no. 62, p. eabk2822, 2022

    2022

  6. [6]

    Learning quadrupedal locomotion on deformable terrain,

    S. Choi, G. Ji, J. Park, H. Kim, J. Mun, J. H. Lee, and J. Hwangbo, “Learning quadrupedal locomotion on deformable terrain,”Science Robotics, vol. 8, no. 74, p. eade2256, 2023

    2023

  7. [7]

    ASAP: Aligning Simulation and Real- World Physics for Learning Agile Humanoid Whole-Body Skills,

    T. He, J. Gao, W. Xiao, Y . Zhang, Z. Wang, J. Wang, Z. Luo, G. He, N. Sobanbabu, C. Pan, Z. Yi, G. Qu, K. Kitani, J. K. Hodgins, L. Fan, Y . Zhu, C. Liu, and G. Shi, “ASAP: Aligning Simulation and Real- World Physics for Learning Agile Humanoid Whole-Body Skills,” inProceedings of Robotics: Science and Systems, (LosAngeles, CA, USA), June 2025

    2025

  8. [8]

    Visual imitation enables contextual humanoid control,

    A. Allshire, H. Choi, J. Zhang, D. McAllister, A. Zhang, C. M. Kim, T. Darrell, P. Abbeel, J. Malik, and A. Kanazawa, “Visual imitation enables contextual humanoid control,” inConference on Robot Learning, pp. 794–815, PMLR, 2025

    2025

Show all 32 references
  1. [9]

    Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion,

    Q. Liao, T. E. Truong, X. Huang, Y . Gao, G. Tevet, K. Sreenath, and C. K. Liu, “Beyondmimic: From motion tracking to versatile humanoid control via guided diffusion,”arXiv preprint arXiv:2508.08241, 2025

    2025 arXiv

  2. [10]

    Zest: Zero-shot embodied skill transfer for athletic robot control,

    J. P. Sleiman, H. Li, A. Adu-Bredu, R. Deits, A. Kumar, K. Bergamin, M. Bhardwaj, S. Biddlestone, N. Burger, M. A. Estrada,et al., “Zest: Zero-shot embodied skill transfer for athletic robot control,”arXiv preprint arXiv:2602.00401, 2026

    2026

  3. [11]

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

    X. B. Peng, P. Abbeel, S. Levine, and M. Van de Panne, “Deepmimic: Example-guided deep reinforcement learning of physics-based char- acter skills,”ACM Transactions On Graphics (TOG), vol. 37, no. 4, pp. 1–14, 2018

    2018

  4. [12]

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

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

    2021

  5. [13]

    Adversarial motion priors make good substitutes for com- plex reward functions,

    A. Escontrela, X. B. Peng, W. Yu, T. Zhang, A. Iscen, K. Goldberg, and P. Abbeel, “Adversarial motion priors make good substitutes for com- plex reward functions,” in2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 25–32, IEEE, 2022

    2022

  6. [14]

    Learning robust and agile legged locomotion using adversarial motion priors,

    J. Wu, G. Xin, C. Qi, and Y . Xue, “Learning robust and agile legged locomotion using adversarial motion priors,”IEEE Robotics and Automation Letters, vol. 8, no. 8, pp. 4975–4982, 2023

    2023

  7. [15]

    Humanmimic: Learning natural locomo- tion and transitions for humanoid robot via wasserstein adversarial imitation,

    A. Tang, T. Hiraoka, N. Hiraoka, F. Shi, K. Kawaharazuka, K. Kojima, K. Okada, and M. Inaba, “Humanmimic: Learning natural locomo- tion and transitions for humanoid robot via wasserstein adversarial imitation,” in2024 IEEE International Conference on Robotics and Automation (I...

    2024

  8. [16]

    Centroidal dynamics of a humanoid robot,

    D. E. Orin, A. Goswami, and S.-H. Lee, “Centroidal dynamics of a humanoid robot,”Autonomous robots, vol. 35, no. 2, pp. 161–176, 2013

    2013

  9. [17]

    Whole-body motion planning with centroidal dynamics and full kinematics,

    H. Dai, A. Valenzuela, and R. Tedrake, “Whole-body motion planning with centroidal dynamics and full kinematics,” in2014 IEEE-RAS International Conference on Humanoid Robots, pp. 295–302, IEEE, 2014

    2014

  10. [18]

    Dynamic complementarity conditions and whole-body trajectory optimization for humanoid robot locomotion,

    S. Dafarra, G. Romualdi, and D. Pucci, “Dynamic complementarity conditions and whole-body trajectory optimization for humanoid robot locomotion,”IEEE Transactions on Robotics, vol. 38, no. 6, pp. 3414– 3433, 2022

    2022

  11. [19]

    Fast trajectory optimization for legged robots using vertex-based zmp constraints,

    A. W. Winkler, F. Farshidian, D. Pardo, M. Neunert, and J. Buchli, “Fast trajectory optimization for legged robots using vertex-based zmp constraints,”IEEE Robotics and Automation Letters, vol. 2, no. 4, pp. 2201–2208, 2017

    2017

  12. [20]

    Robust whole- body motion control of legged robots,

    F. Farshidian, E. Jelavi ´c, A. W. Winkler, and J. Buchli, “Robust whole- body motion control of legged robots,” in2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 4589–4596, IEEE, 2017

    2017

  13. [21]

    Trajec- tory optimization through contacts and automatic gait discovery for quadrupeds,

    M. Neunert, F. Farshidian, A. W. Winkler, and J. Buchli, “Trajec- tory optimization through contacts and automatic gait discovery for quadrupeds,”IEEE Robotics and Automation Letters, vol. 2, no. 3, pp. 1502–1509, 2017

    2017

  14. [22]

    Computationally-robust and efficient prioritized whole-body con- troller with contact constraints,

    D. Kim, J. Lee, J. Ahn, O. Campbell, H. Hwang, and L. Sentis, “Computationally-robust and efficient prioritized whole-body con- troller with contact constraints,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1–8, IEEE, 2018

    2018

  15. [23]

    Highly dynamic quadruped locomotion via whole-body impulse control and model predictive control,

    D. Kim, J. Di Carlo, B. Katz, G. Bledt, and S. Kim, “Highly dynamic quadruped locomotion via whole-body impulse control and model predictive control,”arXiv preprint arXiv:1909.06586, 2019

    1909 arXiv

  16. [24]

    Online dcm trajectory adap- tation for push and stumble recovery during humanoid locomotion,

    G. Mesesan, J. Englsberger, and C. Ott, “Online dcm trajectory adap- tation for push and stumble recovery during humanoid locomotion,” in2021 IEEE international conference on robotics and automation (ICRA), pp. 12780–12786, IEEE, 2021

    2021

  17. [25]

    Online walking pattern generation for humanoid robot with compliant motion control,

    M. Kim, D. Lim, and J. Park, “Online walking pattern generation for humanoid robot with compliant motion control,” in2019 International Conference on Robotics and Automation (ICRA), pp. 1417–1422, IEEE, 2019

    2019

  18. [26]

    Fast whole-body motion control of humanoid robots with inertia constraints,

    G. Ficht and S. Behnke, “Fast whole-body motion control of humanoid robots with inertia constraints,” in2020 IEEE International Confer- ence on Robotics and Automation (ICRA), pp. 6597–6603, IEEE, 2020

    2020

  19. [27]

    Model-free reinforce- ment learning for robust locomotion using demonstrations from tra- jectory optimization,

    M. Bogdanovic, M. Khadiv, and L. Righetti, “Model-free reinforce- ment learning for robust locomotion using demonstrations from tra- jectory optimization,”Frontiers in Robotics and AI, vol. 9, p. 854212, 2022

    2022

  20. [28]

    Integrating model-based footstep planning with model-free reinforcement learning for dynamic legged locomotion,

    H. J. Lee, S. Hong, and S. Kim, “Integrating model-based footstep planning with model-free reinforcement learning for dynamic legged locomotion,” in2024 IEEE/RSJ International Conference on Intelli- gent Robots and Systems (IROS), pp. 11248–11255, IEEE, 2024

    2024

  21. [29]

    Dynaflow: Dynamics- embedded flow matching for physically consistent motion generation from state-only demonstrations,

    S. Lee, D. Kang, J. Park, and H.-W. Park, “Dynaflow: Dynamics- embedded flow matching for physically consistent motion generation from state-only demonstrations,”arXiv preprint arXiv:2509.19804, 2025

    2025

  22. [30]

    Agile perceptive multiskill locomotion for quadrupedal robots in the wild,

    J.-G. Kang, J. Park, T.-G. Song, J.-H. Kim, S. Hong, and H.-W. Park, “Agile perceptive multiskill locomotion for quadrupedal robots in the wild,”Science Robotics, vol. 11, no. 116, p. eadz7397, 2026

    2026

  23. [31]

    Proximal policy optimization algorithms,

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

    2017 arXiv

  24. [32]

    Unitree G1 Humanoid Robot

    Unitree Robotics, “Unitree G1 Humanoid Robot.”https://www. unitree.com/g1/

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.