Pith. sign in

REVIEW 3 major objections 5 minor 52 references

A two-stage physics-guided training scheme produces a fully proprioceptive humanoid policy that walks continuous outdoor grass slopes up to 32.1° without collapsing into a crouched low-CoM gait.

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

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

T0 review · grok-4.5

2026-07-10 17:09 UTC pith:V5KEN4WC

load-bearing objection Solid engineering advance on under-served steep-slope humanoid locomotion; the outdoor 32° grass result is real, the soft-prior transfer story is the only soft link. the 3 major comments →

arxiv 2607.07830 v1 pith:V5KEN4WC submitted 2026-07-08 cs.RO

Physics-Guided Biomechanical Gait Adaptation for Humanoid Locomotion on Extreme Sloped Terrains

Xuanyu Chen , Mohan Liu , Dengchen Mei , Zhihao Gu , Haitian Zhang , Kaimin Mao , Haiyue Zhu , Shijun Yan
show 1 more author
Lin Wang
This is my paper
classification cs.RO
keywords humanoid locomotionsloped terrainsreinforcement learningZero Moment Pointbiomechanical gait adaptationproprioceptive controlSim-to-Realcenter of mass regulation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Steep continuous slopes put a constant gravitational bias on a humanoid, so ordinary reinforcement-learning rewards often settle on a slow, crouched “Groucho” gait that buys short-term balance at the cost of posture and further slope capability. This paper claims that the problem can be solved by a two-stage procedure called HumoSlope. Stage I first learns a balance prior by measuring Zero-Moment-Point deviation on the local inclined support plane rather than a world-horizontal plane. Stage II then uses a training-only five-dimensional PCA slope descriptor to gate soft biomechanical reward terms that raise CoM height and switch between hip-dominant uphill propulsion and knee-oriented downhill braking. The deployed controller never sees the terrain sensors; it runs on proprioception alone. In simulation the policy finishes compound tracks up to 36°; outdoors it continuously traverses wet grass slopes of 62.7 % grade (32.1°). A sympathetic reader cares because the same gravitational-bias problem appears on everyday ramps and hillsides, and the method shows that physics-aligned balance plus biomechanically motivated reward gates can replace both crouching and online vision.

Core claim

Generic model-free rewards for humanoid slope locomotion converge to an undesired low-center-of-mass crouched gait. HumoSlope prevents that degeneration by first installing a slope-adaptive ZMP regularizer evaluated on the local support plane, then using a privileged macroscopic terrain descriptor to gate soft biomechanical priors that modulate CoM height and lower-limb coordination. The resulting actor remains purely proprioceptive yet achieves continuous blind traversal of outdoor grass slopes up to 32.1° and simulated compound slopes up to 36°.

What carries the argument

HumoSlope: a two-stage framework whose Stage-I slope-adaptive ZMP regularizer (terrain-aligned deviation from a force-weighted support anchor) supplies a balance prior, and whose Stage-II Biomechanical Slope Gait Adapter (BSGA) gates CoM-height, hip/knee, and swing-leg soft rewards from a five-dimensional PCA terrain descriptor available only at training time.

Load-bearing premise

A five-number PCA summary of a privileged height-scan patch, together with soft reward gates fitted from Stage-I rollouts, is rich enough that a purely proprioceptive actor can later walk unseen outdoor grass slopes without any online terrain sensing.

What would settle it

Deploy the identical proprioceptive actor, trained without the BSGA reward gates or without the slope-adaptive ZMP term, on the same outdoor grass slope of measured grade ≥30°; if that ablated policy still completes continuous traversal while keeping mean CoM height comparable to the full model, the claimed necessity of the two-stage physics-guided adaptation is falsified.

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

If this is right

  • Humanoid policies trained with ordinary tracking-and-survival rewards will systematically prefer low-CoM crouches on continuous inclines unless the balance metric is evaluated on the local support plane.
  • Training-time macroscopic slope descriptors can encode uphill/downhill joint-work asymmetry without requiring the deployed controller to carry cameras or depth sensors.
  • Compound uphill–downhill tracks with friction tiers normalized to tan(θ) become a stricter and more informative benchmark than isolated constant ramps.
  • Once the Stage-I balance prior exists, soft biomechanical gates alone are enough to convert a crouched warm-start into a faster, more upright slope gait.

Where Pith is reading between the lines

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

  • The same local-plane ZMP construction may transfer to other persistent gravitational biases such as walking under constant external force or on banked curves.
  • If the PCA descriptor can be replaced by a short history of proprioceptive accelerations, the entire pipeline could become fully unsupervised with respect to height maps.
  • Abrupt slope transitions will remain a failure mode until some form of look-ahead cue is added, exactly as the paper’s own limitations section anticipates.
  • Peak knee-torque diagnostics from the ablations suggest that uncontrolled crouching is not merely aesthetic; it is a measurable overload that limits maximum grade.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper presents HumoSlope, a two-stage physics-guided RL framework for blind humanoid locomotion on continuous steep slopes. Stage I trains a proprioceptive actor–critic with a slope-adaptive ZMP regularizer that evaluates balance deviation on the local inclined support plane rather than a world-horizontal reference, producing a terrain-consistent balance prior. Stage II warm-starts from that actor and introduces the Biomechanical Slope Gait Adapter (BSGA), which uses a training-only five-dimensional PCA terrain descriptor (Eq. 1) extracted from height-scan patches to gate soft reward priors for slope-conditioned CoM height (Eq. 3), uphill/downhill lower-limb asymmetry (Eq. 4), and swing-hip guidance fitted from Stage-I rollouts (Eq. 5). The deployed actor remains purely proprioceptive. On held-out compound slope-track benchmarks the method reaches 77.1% success at 30° and a max grade of 36°, outperforming proprioceptive and one exteroceptive baseline; outdoor Unitree G1 demos show continuous traversal of grass slopes up to 32.1° with qualitative posture adaptation.

Significance. Continuous steep slopes are a distinct and under-studied regime for humanoid RL: they impose a persistent gravitational bias rather than discrete foothold selection, and generic rewards readily induce low-CoM “Groucho” gaits. The combination of a terrain-aligned ZMP prior with biomechanically motivated, descriptor-gated soft rewards is a concrete and transferable design pattern. Strengths include a held-out compound-track protocol with friction-tier normalization, three-checkpoint averaging, a full ablation suite (Table 2), biomechanical diagnostics (Fig. 5), and real outdoor video evidence on deformable grass. If the claimed transfer holds, the work supplies both a practical recipe for extreme-slope humanoid locomotion and a clear demonstration that physics- and biomechanics-informed reward shaping can mitigate posture degeneration without online exteroception.

major comments (3)
  1. [§4.2 Real-world experiments / Abstract] The headline outdoor claim (continuous blind traversal of grass slopes up to 32.1°) rests on qualitative figures (Figs. 1, 6) and narrative description. Unlike the simulation protocol (Table 1: three checkpoints × three friction tiers, SR/MXD/T_trav), the real-world section supplies no repeated-trial success rates, distance statistics, failure modes, or wetness/friction conditions. Because the load-bearing transfer argument is that Stage-II soft priors (Eqs. 3–5) remain informative once the privileged PCA descriptor is removed and the surface becomes deformable grass, quantitative outdoor metrics (or at least a clear statement of trial counts and observed failure modes) are needed to support the central claim at the strength asserted in the abstract.
  2. [§3.2 Eqs. (3)–(5); Table 2; Fig. 6] Table 2 shows that removing BSGA collapses SR to 0% at 20°, while Stage I alone retains 100% SR but with markedly lower CoM height and longer traversal time. This establishes necessity of BSGA in simulation, yet leaves open whether the fitted swing-hip trend (Eq. 5) and asymmetric CoM offsets (Eq. 3) are tuned to rigid sim geometry. A sensitivity or re-fit experiment (e.g., perturbing β0, β1 or b_up/b_down and re-evaluating both sim and real posture) would strengthen the causal link between the claimed physics-guided mechanism and the outdoor posture adaptation shown in Fig. 6.
  3. [§3.1 Slope-Adaptive ZMP Regularization] The slope-adaptive ZMP regularizer (Sec. 3.1) is a point-mass apparent-force surrogate with a contact-force-weighted support anchor. The paper correctly notes that a full ZMP/ZML formulation is noisy in massively parallel RL, but does not quantify how often the terrain-aligned intersection falls outside the actual support polygon or how sensitive r_ta_zmp is to the smoothing constant ε and scale σ_zmp. A short diagnostic (distribution of d_ta_zmp on steep segments, or ablation of the force-weighted anchor versus a simple mid-foot anchor) would make the Stage-I contribution more transparent and reproducible.
minor comments (5)
  1. [§3.2 / Implementation] Reward weights (w_com, w_bio, w_swing), ascent/descent offsets, and the PCA clip threshold θ_clip are free parameters listed only conceptually; a table or appendix of numerical values (and any tuning protocol) would aid reproducibility.
  2. [Fig. 5] Fig. 5 panels lack error bands or trial counts; given the three-checkpoint averaging used in Table 1, the same protocol should be stated for the biomechanical diagnostics.
  3. [§4.1 / Table 1] The friction-tier definition µ = tan(|θ|) + ∆ is clear, but the text should note whether the same tiers were used for the Max Grade sweep column of Table 1.
  4. [Eq. (1) and Eq. (4)] Minor notation: 1_up / 1_down appear both as indicators and as gating factors; a single sentence clarifying that they are binary regime flags would avoid ambiguity.
  5. [§2 Dynamic Balance] Related-work discussion of multi-contact ZMP/ZML (Caron et al., Brecelj & Petrič) is appropriate; a brief remark on why those geometric formulations were not used as hard constraints (rather than soft rewards) would help readers unfamiliar with the RL setting.

Circularity Check

0 steps flagged

No significant circularity: empirical two-stage RL method whose success metrics are external held-out tracks and real-world demos, not quantities forced by the Stage-I fits or soft priors.

full rationale

HumoSlope is an engineering RL pipeline (PPO actor-critic with asymmetric observations). Stage I adds a terrain-aligned ZMP deviation reward (point-mass surrogate on the local support plane estimated from stance foot); Stage II gates soft, low-weight biomechanical priors (CoM height target, hip/knee directional biases, swing hip-pitch reference) using a training-only 5-D PCA descriptor of the privileged height scan. The swing coefficients (Eq. 5) are fitted from Stage-I rollouts and used only as a soft training prior, not as a claimed first-principles prediction of an external observable. Performance claims (SR/MXD/Ttrav on held-out compound slope tracks up to 36°, real outdoor grass up to 32.1°) are measured against external environments and ablations that remove components; they are not recovered by construction from the fitted priors or the PCA definition. Biomechanical motivation cites independent human-locomotion literature. Minor co-author citations exist on unrelated topics and are not load-bearing for the slope claims. No self-definitional loop, no fitted quantity re-labeled as prediction, no uniqueness theorem imported from the authors, and no renaming of a known result. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 3 invented entities

The central claim rests on classical ZMP geometry, human biomechanics literature, and a set of hand-chosen reward gates and PCA descriptors introduced by the authors. No new physical constants are fitted; the free parameters are the usual RL reward weights and the linear fit for swing-hip targets. Invented entities are the named modules (HumoSlope, BSGA, slope-adaptive ZMP regularizer) whose only evidence is the performance tables themselves.

free parameters (4)
  • BSGA reward weights (w_com, w_bio, w_swing) and ascent/descent offsets (b_up, b_down)
    Hand-tuned soft priors that gate CoM height and joint preferences; values are not derived from first principles and directly affect posture metrics in Table 2.
  • Linear coefficients β0, β1 for swing-hip target (Eq. 5)
    Fitted from Stage-I rollouts; the swing reward tracks this fitted curve.
  • ZMP deviation scale σ_zmp and support-anchor smoothing ε
    Shape the dense Stage-I reward; chosen for training stability rather than measured.
  • PCA clip threshold θ_clip and normalized slope intensity ρ_slope
    Ad-hoc normalizations that control when biomechanical gates activate.
axioms (4)
  • domain assumption Evaluating ZMP deviation on the local inclined support plane (rather than a world-horizontal plane) yields a terrain-consistent balance prior useful for RL.
    Stated in Sec. 3.1 and Fig. 3; classical ZMP theory is adapted but not re-proved for the point-mass surrogate used in massively parallel simulation.
  • domain assumption Human uphill/downhill joint-work asymmetries (hip propulsion, knee braking) transfer as useful soft priors for a bipedal humanoid of different morphology.
    Invoked via citations [16,18] and encoded in r_bio (Eq. 4); no independent validation that the same redistribution is optimal for the Unitree G1.
  • ad hoc to paper A five-dimensional PCA descriptor of a height-scan patch is a sufficient macroscopic slope cue for gating rewards and for the privileged critic.
    Defined in Eq. 1; introduced specifically for BSGA and not shown to be optimal among possible terrain summaries.
  • standard math PPO with asymmetric actor-critic and domain randomization is a valid optimization procedure for the claimed sim-to-real transfer.
    Standard in the cited locomotion literature; used without modification.
invented entities (3)
  • HumoSlope two-stage framework no independent evidence
    purpose: Organize the slope-adaptive ZMP warm-start and subsequent BSGA adaptation into a single training pipeline.
    Named contribution of the paper; evidence is the performance tables and real-world demos.
  • Biomechanical Slope Gait Adapter (BSGA) no independent evidence
    purpose: Gate CoM-height, hip/knee, and swing-leg soft rewards from a training-only PCA descriptor.
    Core Stage-II module; no external falsifiable prediction beyond the reported success rates.
  • Slope-adaptive ZMP regularizer (terrain-aligned d_zmp^ta) no independent evidence
    purpose: Provide a dense balance reward evaluated on the local support plane rather than a horizontal reference.
    Defined via the point-mass ray-plane intersection in Sec. 3.1; utility shown only by ablation.

pith-pipeline@v1.1.0-grok45 · 18000 in / 3298 out tokens · 86925 ms · 2026-07-10T17:09:36.874332+00:00 · methodology

0 comments
read the original abstract

Model-free reinforcement learning has enabled impressive humanoid locomotion; however, control on steep slopes remains largely unexplored. Unlike flat or discrete terrains, sloped terrains impose a persistent gravitational bias that demands simultaneous stability and posture control. Consequently, under generic reward formulations, policies can converge to slow, conservative low-center-of-mass (CoM) crouched gaits. In this work, we propose a novel two-stage physics-guided framework, dubbed HumoSlope, dedicated to robust humanoid locomotion on diverse sloped terrains. Specifically, Stage I establishes a terrain-consistent balance prior by introducing a slope-adaptive Zero Moment Point (ZMP) regularizer evaluated directly on the local inclined support plane rather than a world-horizontal reference. To prevent the resulting policy from defaulting to a crouched posture, Stage II introduces the Biomechanical Slope Gait Adapter (BSGA). Utilizing extracted macroscopic terrain descriptors as privileged, training-only signals, BSGA dynamically gates soft reward priors to modulate CoM height and lower-limb coordination based on the estimated slope geometry -- encouraging hip-dominant uphill propulsion and knee-oriented downhill braking. Crucially, the deployed actor remains entirely proprioceptive, requiring no online exteroceptive sensing. Extensive Sim-to-Real experiments demonstrate that our framework effectively mitigates posture degeneration and enables blind, continuous traversal of outdoor grass slopes up to 62.7% ($32.1^\circ$), validating a physics-guided approach to challenging slope terrain adaptation.

Figures

Figures reproduced from arXiv: 2607.07830 by Dengchen Mei, Haitian Zhang, Haiyue Zhu, Kaimin Mao, Lin Wang, Mohan Liu, Shijun Yan, Xuanyu Chen, Zhihao Gu.

Figure 1
Figure 1. Figure 1: Real-world locomotion on sloped terrains. Our robot traverses grassy slopes up to 62.7% (32.1 ◦ ) grade and generalizes to slippery surfaces, grass, wavy terrains, and level walkways. Abstract: Model-free reinforcement learning has enabled impressive humanoid locomotion; however, control on steep slopes remains largely unexplored. Un￾like flat or discrete terrains, sloped terrains impose a persistent gravi… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed two-stage blind slope-locomotion framework. The actor uses [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: On slope, the horizontal￾reference deviation dhoriz can differ from the terrain-aligned deviation d ta zmp. Approximation scope and support anchor. A strict ZMP/ZML reward would require estimating centroidal angular-momentum rates and a reliable contact-wrench support region, which are noisy and contact-solver depen￾dent in massively parallel RL. We thus use a point-mass apparent-force surrogate and avoid … view at source ↗
Figure 4
Figure 4. Figure 4: Held-out compound slope-track benchmark and representative policy behaviors. (a) Our [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Biomechanical diagnostics. HumoSlope maintains a higher CoM height with mild slope [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Real-world posture on grass terrain [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 11 internal anchors

  1. [1]

    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

    work page 2022

  2. [2]

    Hwangbo, J

    J. Hwangbo, J. Lee, A. Dosovitskiy, D. Bellicoso, V . Tsounis, V . Koltun, and M. Hutter. Learn- ing agile and dynamic motor skills for legged robots.Science robotics, 4(26):eaau5872, 2019

    work page 2019

  3. [3]

    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

    work page 2024

  4. [4]

    Blind Bipedal Stair Traversal via Sim-to-Real Reinforcement Learning

    J. Siekmann, K. Green, J. Warila, A. Fern, and J. Hurst. Blind bipedal stair traversal via sim- to-real reinforcement learning.arXiv preprint arXiv:2105.08328, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  5. [5]

    H. Wang, Z. Wang, J. Ren, Q. Ben, T. Huang, W. Zhang, and J. Pang. Beamdojo: Learning agile humanoid locomotion on sparse footholds.arXiv preprint arXiv:2502.10363, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    Humanoid Parkour Learning

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

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  7. [7]

    Learning Humanoid Locomotion over Challenging Terrain

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

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    Advancing Humanoid Locomotion: Mastering Challenging Terrains with Denoising World Model Learning

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  9. [9]

    H. Song, H. Zhu, T. Yu, Y . Liu, M. Yuan, W. Zhou, H. Chen, and H. Li. Gait-adaptive per- ceptive humanoid locomotion with real-time under-base terrain reconstruction.IEEE Robotics and Automation Letters, 2026

    work page 2026

  10. [10]

    Suliman, E

    W. Suliman, E. Davydenko, E. Chaikovskaia, and R. Gorbachev. Reinforcement learning- based footstep control for humanoid robots on complex terrain.IEEE Access, 2025

    work page 2025

  11. [11]

    Unitree RL lab: Reinforcement learning implementation for unitree robots, based on IsaacLab, 2024

    Unitree Robotics. Unitree RL lab: Reinforcement learning implementation for unitree robots, based on IsaacLab, 2024. URLhttps://github.com/unitreerobotics/unitree_rl_ lab

    work page 2024

  12. [12]

    T. A. McMahon, G. Valiant, and E. C. Frederick. Groucho running.Journal of applied physi- ology, 62(6):2326–2337, 1987

    work page 1987

  13. [13]

    X. B. Peng, G. Berseth, K. Yin, and M. Van De Panne. Deeploco: Dynamic locomotion skills using hierarchical deep reinforcement learning.Acm transactions on graphics (tog), 36(4): 1–13, 2017

    work page 2017

  14. [14]

    Vukobratovi ´c and J

    M. Vukobratovi ´c and J. Stepanenko. On the stability of anthropomorphic systems.Mathemat- ical biosciences, 15(1-2):1–37, 1972

    work page 1972

  15. [15]

    W. Xie, C. Bai, J. Shi, J. Yang, Y . Ge, W. Zhang, and X. Li. Humanoid whole-body loco- motion on narrow terrain via dynamic balance and reinforcement learning. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 4751–4758. IEEE, 2025

    work page 2025

  16. [16]

    Vernillo, M

    G. Vernillo, M. Giandolini, W. B. Edwards, J.-B. Morin, P. Samozino, N. Horvais, and G. Y . Millet. Biomechanics and physiology of uphill and downhill running.Sports Medicine, 47(4): 615–629, 2017

    work page 2017

  17. [17]

    T. J. Roberts and R. A. Belliveau. Sources of mechanical power for uphill running in humans. Journal of Experimental Biology, 208(10):1963–1970, 2005. 9

    work page 1963

  18. [18]

    A. E. Minetti, C. Moia, G. S. Roi, D. Susta, and G. Ferretti. Energy cost of walking and running at extreme uphill and downhill slopes.Journal of applied physiology, 2002

    work page 2002

  19. [19]

    K. Pearson. Liii. on lines and planes of closest fit to systems of points in space.The London, Edinburgh, and Dublin philosophical magazine and journal of science, 2(11):559–572, 1901

    work page 1901

  20. [20]

    Z. Xie, P. Clary, J. Dao, P. Morais, J. Hurst, and M. Panne. Learning locomotion skills for cassie: Iterative design and sim-to-real. InConference on Robot Learning, pages 317–329. PMLR, 2020

    work page 2020

  21. [21]

    Z. Li, X. B. Peng, P. Abbeel, S. Levine, G. Berseth, and K. Sreenath. Reinforcement learning for versatile, dynamic, and robust bipedal locomotion control.The International Journal of Robotics Research, 44(5):840–888, 2025

    work page 2025

  22. [22]

    Q. Ben, B. Xu, K. Li, F. Jia, W. Zhang, J. Wang, J. Wang, D. Lin, and J. Pang. Gallant: V oxel grid-based humanoid locomotion and local-navigation across 3d constrained terrains.arXiv preprint arXiv:2511.14625, 2025

    work page arXiv 2025

  23. [23]

    J. He, C. Zhang, F. Jenelten, R. Grandia, M. B ¨acher, and M. Hutter. Attention-based map encoding for learning generalized legged locomotion.Science Robotics, 10(105):eadv3604, 2025

    work page 2025

  24. [24]

    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

    work page 2023

  25. [25]

    J. Long, J. Ren, M. Shi, Z. Wang, T. Huang, P. Luo, and J. Pang. Learning humanoid locomo- tion with perceptive internal model. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 9997–10003. IEEE, 2025

    work page 2025

  26. [26]

    W. Sun, L. Chen, Y . Su, B. Cao, Y . Liu, and Z. Xie. Learning humanoid locomotion with world model reconstruction.arXiv preprint arXiv:2502.16230, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

    RMA: Rapid Motor Adaptation for Legged Robots

    A. Kumar, Z. Fu, D. Pathak, and J. Malik. Rma: Rapid motor adaptation for legged robots. arXiv preprint arXiv:2107.04034, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  28. [28]

    Zhang, Y

    Y . Zhang, Y . Seo, J. Chen, Y . Yuan, K. Sreenath, P. Abbeel, C. Sferrazza, K. Liu, R. Duan, and G. Shi. Rpl: Learning robust humanoid perceptive locomotion on challenging terrains.arXiv preprint arXiv:2602.03002, 2026

    work page arXiv 2026

  29. [29]

    Z. Gu, M. Yang, D. Zou, and D. Xu. Learning diffusion policy from primitive skills for robot manipulation.arXiv preprint arXiv:2601.01948, 2026

    work page arXiv 2026

  30. [30]

    LiMoDE: Rethinking Lifelong Robot Manipulation from a Mixture-of-Dynamic-Experts Perspective

    Z. Gu and L. Wang. Limode: Rethinking lifelong robot manipulation from a mixture-of- dynamic-experts perspective.arXiv preprint arXiv:2606.26183, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  31. [31]

    Sardain and G

    P. Sardain and G. Bessonnet. Forces acting on a biped robot. center of pressure-zero moment point.IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, 34 (5):630–637, 2004

    work page 2004

  32. [32]

    Z. Gu, J. Li, W. Shen, W. Yu, Z. Xie, S. McCrory, X. Cheng, A. Shamsah, R. Griffin, C. K. Liu, et al. Humanoid locomotion and manipulation: Current progress and challenges in control, planning, and learning.IEEE/ASME Transactions on Mechatronics, 31(2):2300–2330, 2026

    work page 2026

  33. [33]

    Kim, I.-W

    J.-Y . Kim, I.-W. Park, and J.-H. Oh. Walking control algorithm of biped humanoid robot on uneven and inclined floor.Journal of intelligent and robotic systems, 48(4):457–484, 2007

    work page 2007

  34. [34]

    Seven, T

    U. Seven, T. Akbas, K. C. Fidan, M. Yilmaz, and K. Erbatur. Humanoid robot walking control on inclined planes. In2011 IEEE International Conference on Mechatronics, pages 875–880. IEEE, 2011. 10

    work page 2011

  35. [35]

    Caron, Q.-C

    S. Caron, Q.-C. Pham, and Y . Nakamura. Zmp support areas for multicontact mobility under frictional constraints.IEEE Transactions on Robotics, 33(1):67–80, 2016

    work page 2016

  36. [36]

    Brecelj and T

    T. Brecelj and T. Petri ˇc. Zero moment line—universal stability parameter for multi-contact systems in three dimensions.Sensors, 22(15):5656, 2022

    work page 2022

  37. [37]

    R. C. Sheehan and J. S. Gottschall. At similar angles, slope walking has a greater fall risk than stair walking.Applied ergonomics, 43(3):473–478, 2012

    work page 2012

  38. [38]

    N. T. Pickle, A. M. Grabowski, A. G. Auyang, and A. K. Silverman. The functional roles of muscles during sloped walking.Journal of biomechanics, 49(14):3244–3251, 2016

    work page 2016

  39. [39]

    A. H. Dewolf, Y . Ivanenko, K. E. Zelik, F. Lacquaniti, and P. A. Willems. Kinematic patterns while walking on a slope at different speeds.Journal of Applied Physiology, 125(2):642–653, 2018

    work page 2018

  40. [40]

    R. W. Nuckols, K. Z. Takahashi, D. J. Farris, S. Mizrachi, R. Riemer, and G. S. Sawicki. Mechanics of walking and running up and downhill: A joint-level perspective to guide design of lower-limb exoskeletons.PloS one, 15(8):e0231996, 2020

    work page 2020

  41. [41]

    Papachatzis and K

    N. Papachatzis and K. Z. Takahashi. Mechanics of the human foot during walking on different slopes.PLoS One, 18(9):e0286521, 2023

    work page 2023

  42. [42]

    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

    work page 2018

  43. [43]

    C. Yao, C. Liu, L. Xia, M. Liu, and Q. Chen. Humanoid adaptive locomotion control through a bioinspired cpg-based controller.Robotica, 40(3):762–779, 2022

    work page 2022

  44. [44]

    J. Fang, Y . Jin, B. Wang, K. Zhou, M. Wang, and Z. Liu. Bio-inspired central pattern generator for adaptive gait generation and stability in humanoid robots on sloped surfaces.Biomimetics, 10(9):637, 2025

    work page 2025

  45. [45]

    Liang, Y

    Y . Liang, Y . Zhao, and W. Yan. Terrain-adaptive bipedal locomotion via reinforcement learning with human-inspired stepping strategy. InInternational Conference on Intelligent Robotics and Applications, pages 624–636. Springer, 2025

    work page 2025

  46. [46]

    F. Jin, Y . Wang, P. Ma, G. Yang, P. Zhao, E. Li, and Z. Zhang. Teacher motion priors: Enhanc- ing robot locomotion over challenging terrain. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1100–1107. IEEE, 2025

    work page 2025

  47. [47]

    Padulo, D

    J. Padulo, D. Powell, R. Milia, and L. P. Ardig `o. A paradigm of uphill running.PloS one, 8 (7):e69006, 2013

    work page 2013

  48. [48]

    L. P. Kaelbling, M. L. Littman, and A. R. Cassandra. Planning and acting in partially ob- servable stochastic domains.Artificial Intelligence, 101(1):99–134, 1998. ISSN 0004-3702. doi:https://doi.org/10.1016/S0004-3702(98)00023-X. URLhttps://www.sciencedirect. com/science/article/pii/S000437029800023X

    work page doi:10.1016/s0004-3702(98)00023-x 1998

  49. [49]

    ˚Astr¨om

    K. ˚Astr¨om. Optimal control of markov processes with incomplete state information.Jour- nal of Mathematical Analysis and Applications, 10(1):174–205, 1965. ISSN 0022-247X. doi:https://doi.org/10.1016/0022-247X(65)90154-X. URLhttps://www.sciencedirect. com/science/article/pii/0022247X6590154X

    work page doi:10.1016/0022-247x(65)90154-x 1965

  50. [50]

    Proximal Policy Optimization Algorithms

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov. Proximal policy optimization algorithms, 2017. URLhttps://arxiv.org/abs/1707.06347. 11

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  51. [51]

    Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning

    M. Mittal, P. Roth, J. Tigue, A. Richard, O. Zhang, P. Du, A. Serrano-Mu ˜noz, X. Yao, R. Zurbr ¨ugg, N. Rudin, L. Wawrzyniak, M. Rakhsha, A. Denzler, E. Heiden, A. Borovicka, O. Ahmed, I. Akinola, A. Anwar, M. T. Carlson, J. Y . Feng, A. Garg, R. Gasoto, L. Gulich, Y . Guo, M. Gussert, A. Hansen, M. Kulkarni, C. Li, W. Liu, V . Makoviychuk, G. Malczyk, H...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    Y . Seo, C. Sferrazza, H. Geng, M. Nauman, Z.-H. Yin, and P. Abbeel. Fasttd3: Simple, fast, and capable reinforcement learning for humanoid control.arXiv preprint arXiv:2505.22642, 2025. 12

    work page internal anchor Pith review Pith/arXiv arXiv 2025