PUMA: Perception-driven Unified Foothold Prior for Mobility Augmented Quadruped Parkour

Jun Wu; Kanzhong Yao; Liang Wang; Qiuguo Zhu; Weikai Qin; Yang Liu; Zhe Sun

REVIEW 4 major objections 5 minor 27 references

PUMA: Perception-driven Unified Foothold Prior for Mobility Augmented Quadruped Parkour

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read PUMA shows that an egocentric polar foothold prior — two front-foot distances plus two heading errors — beats coordinate-based foothold targets for end-to-end quadruped parkour, reaching 94.7–98.7% simulation success and near-perfect real-w

desk verdict Solid empirical case for polar foothold priors in quadruped parkour, but the oracle-to-depth recoverability gap and thin real-world statistics keep it conditional rather than accept. read the letter →

arxiv 2601.15995 v2 pith:DIZHNNVY submitted 2026-01-22 cs.RO cs.AIcs.LG

Liang Wang , Kanzhong Yao , Yang Liu , Weikai Qin , Jun Wu , Zhe Sun , Qiuguo Zhu This is my paper

classification cs.ROcs.AIcs.LG

keywords footholdpriorquadrupedparkourreinforcementlearningegocentricpolarrepresentationvisualperceptionsim-to-realtransfersingle-stagepolicytraining

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

PUMA is an end-to-end reinforcement-learning framework for quadruped parkour over discrete terrain — stepping stones, gaps, and high platforms. Its central move is to represent the next foothold not as an (x, y, z) coordinate to track but as an egocentric polar prior: two front-foot distances to the expected contact point plus two heading errors. This four-scalar cue is predicted from an onboard depth camera and proprioception, then fed to a velocity-tracking actor, guiding posture adaptation (twisting to kick off a wall) without strict foothold tracking. In simulation it reaches 94.7–98.7% success across five terrain configurations, beating Cartesian foothold priors and recent baselines; zero-shot real-world trials succeed in 8–10 out of 10 runs. The paper argues that the format of the guidance signal is a decisive inductive bias for agile legged locomotion.

What carries the argument

The load-bearing object is the egocentric polar foothold prior, f_t = {d_t^(L), d_t^(R), ψ_t, ψ_{t+1}} — four scalars giving the distances from the left and right forefeet to the expected foothold and the heading errors toward the current and next foothold. Its work is to convert a potentially error-prone 3D foot-coordinate target into a minimal directional-and-distance cue that can guide posture adaptation without dictating exact foot placement, preserving the policy's exploratory freedom. Two mechanisms make the single-stage training work: probability annealing selection (PAS), a cosine schedule that gradually hands control from the ground-truth prior to the predicted prior; and a multi-cr

What would settle it

Compare polar and Cartesian priors under oracle execution: feed the ground-truth prior (no estimation error) to the actor for both representations on the wall-assisted gap task. If success rates become equal, the polar format's benefit is purely in easing the regression problem; if polar remains superior, the representation changes what the policy itself can learn. Either result would settle the paper's explanation for its advantage.

Watch

Extended reading notes

Core claim

The paper's central claim is that egocentric polar foothold priors are a better motion-guide for agile quadruped locomotion than coordinate-based foothold targets. The prior decomposes a foothold target into four scalars: the distances from the left and right forefeet to the expected contact point, and the heading errors toward the current and next foothold. A shared depth-plus-proprioception estimator regresses this prior alongside base velocity and a terrain latent, and the actor uses it to adjust body posture rather than to track an exact foot position. On five discrete-terrain tasks, the polar prior yields simulation success rates of 94.7–98.7%, exceeding explicit and implicit Cartesian

Load-bearing premise

Everything rests on the assumption that an oracle can pre-select the geometrically useful contact points — the centers of exploitable inclined walls, filtered by distance to terrain edges — and that the depth-only estimator can recover those same targets from pixels at deployment; the paper's own failure analysis shows this assumption is sometimes violated.

Editorial extensions

If this is right

End-to-end parkour training is feasible without hierarchical foothold planning or teacher–student distillation: the robot discovers galloping gaits and wall-kick propulsion from a four-scalar cue.
Compact polar guidance roughly halves regression error compared with Cartesian foot coordinates, making sim-to-real transfer more robust with noisy onboard depth.
PAS provides a general recipe for annealing from privileged ground truth to predicted estimates in single-stage RL, accelerating convergence without hurting final performance.
Multi-critic reward separation is necessary on inclined terrain, allowing the policy to sacrifice velocity tracking in favor of momentum-building contact phases.

Reading between the lines

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

The advantage of the polar prior may stem from the information bottleneck it imposes; a direct test would feed ground-truth priors (polar vs Cartesian) at execution time and see if the policy gap persists once estimation error is removed.
The oracle-sampled foothold set — hand-chosen centers of inclined walls filtered by edge distance — could be replaced by a learned affordance estimator, removing the need for terrain-specific annotation and making the prior adaptive to unseen structures.
Because the paper's real-world evaluation uses only 10 trials per condition, a larger deployment study is a natural next test to confirm the robustness claims beyond single-run anecdotes.
The polar representation might generalize to bipedal or industrial robots that must aim feet at discrete features, where the same decomposition of distance and heading could reduce the perceptual burden.

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.

Desk Editor's Note

read the letter

Short version: PUMA makes a credible empirical case that a polar foothold prior—two distances plus two headings—is a better representation for single-stage quadruped parkour than Cartesian foothold coordinates. The simulation ablations are solid: 1000 trials per condition, consistent margins over explicit and implicit Cartesian priors, and a fair comparison against PIE and Extreme Parkour using their own reward functions and the teacher policy for Extreme Parkour. The real robot demos add weight, and the failure-case breakdown is honestly done—they even show a case where foothold estimation error causes a missed wall contact.

What's genuinely new is the egocentric polar decomposition, combined with PAS annealing and multi-critic training to make the single-stage pipeline converge. The Table III regression MSE data support the claim that the polar form is easier to recover from depth than Cartesian coordinates, and the downstream success rates line up with that. The authors also show that the biggest benefit appears on terrain that requires actively exploiting inclined walls, which is the interesting case.

The soft spots are real but not fatal. The ground-truth foothold prior is defined using privileged terrain knowledge: candidates sampled at 1 m intervals, filtered by d_safe, and explicitly including the centers of inclined walls. At deployment the estimator has to recover those same targets from a 10 Hz depth stream. The paper reports regression MSE but never shows how much task success degrades as prior error increases. The real-world failure in Fig. 7d is exactly this failure mode, so the concern is not hypothetical. That said, the same recoverability issue applies to the Cartesian baselines, so it doesn't undermine the core comparison—it does temper the real-world robustness claims and the 'exceptional' language in the abstract.

Second, the real-world evaluation is thin: 10 trials per method, no confidence intervals, and the hardest course has a 0.8 success rate. That's enough for a proof-of-concept but not for strong claims. Third, no code or exact hyperparameters are released, and while the reward table and many settings are listed, independent reproduction would be painful. The PAS finding—that it mainly speeds training rather than improving final performance—is honest and not oversold. The multi-critic ablation shows a clear drop on hard terrain, and the qualitative analysis in Fig. 4 is reasonable.

Who is this for: researchers working on legged locomotion or RL with learned priors. It's a useful representation to build on. I'd want code or a detailed training recipe before citing it heavily, but the direction is sound.

Recommendation: send it to review. It's a competent, well-ablated empirical paper with a clear representation hypothesis. The referee should push for code/data release, more real-world trials with uncertainty quantification, and an explicit analysis of sensitivity to prior error. With those changes, it would be a solid contribution.

Referee Report

4 major / 5 minor

Summary. The paper presents PUMA, an end-to-end reinforcement-learning framework for quadruped parkour. A depth-plus-proprioception estimator predicts an egocentric polar foothold prior—two forefoot distances and two heading errors—that is annealed from ground-truth to predicted values during training via Probability Annealing Selection (PAS). The policy is trained with PPO and a multi-critic reward structure on three discrete terrain types: stepping stones, wall-assisted gaps, and surmounting platforms. Experiments in Isaac Gym (1,000 trials per configuration) and on a DeepRobotics Lite3 robot (10 trials per method) compare PUMA with Cartesian-prior variants and the PIE and Extreme Parkour baselines. The paper reports higher simulation success rates (94.7–98.7%) and better real-world success for PUMA, attributing the gain to the polar representation of foothold priors.

Significance. If the findings hold, the polar foothold prior is a simple and effective inductive bias for agile legged locomotion, avoiding explicit foothold tracking and hierarchical planning while improving perception-to-control transfer. The simulation protocol is a strength: 1,000 trials per configuration, multiple ablations, and a shared terrain curriculum. The zero-shot real-world deployment across three dynamic tasks is valuable, and the paper honestly lists typical failure cases. However, the real-world evidence is limited to 10 trials per condition, and the gap between oracle-generated foothold priors and depth-only estimation is not directly quantified. The central claim is therefore plausible but not yet fully established.

major comments (4)

[Section IV-C, Fig. 5] The real-world evaluation uses only 10 trials per method and reports no confidence intervals or statistical tests. For the decisive comparisons—e.g., PUMA 1.0 vs explicit Cartesian 0.6/0.2 on wall-assisted gap, and 0.8 vs 0.1/0.1 on surmounting—the binomial 95% CI for 10/10 is about [0.69,1.0] and for 8/10 is about [0.44,0.97]. The observed differences are not statistically significant at n=10. The abstract's real-world robustness claim requires more trials, repeated runs, or at least error bars.
[Section III-A (Eq. 1) and III-B.2, Fig. 7d] The ground-truth foothold prior is generated from privileged terrain knowledge: candidates sampled at 1 m intervals including 'centers of exploitable inclined walls', filtered by d_safe. At deployment, a depth-only estimator must recover the same four scalars from 10 Hz depth images. Table III reports regression MSE but does not quantify how prior error affects task success. Fig. 7d explicitly attributes a failure to 'Foothold estimation error causing missed wall contact'. Without a controlled-noise experiment on f_input_t or an analysis of success as a function of estimator error, the oracle-to-estimator gap is a load-bearing unquantified assumption.
[Section IV-B.1, Table III] The MSE comparison between polar and Cartesian priors is not apples-to-apples. The polar target contains distances (meters) and angles (radians); the Cartesian target contains x,y,z (meters). The scales and units differ, and the reported 'MSE(%)' normalization is not defined. The claim that the polar representation 'achieves significantly higher accuracy' is therefore not supported by the data as presented. To make the representation argument, the authors should report a normalized per-component error on a common scale or compare task success under matched prediction error.
[Section III-A/C, Table II] The comparison between PUMA and the Cartesian-prior ablations is confounded by reward alignment. The foothold rewards (Eqs. 4–6) are defined on d_L, d_R, and ψ—exactly the components of f_t that are fed to the actor in PUMA. The Cartesian-prior variants receive raw x,y,z coordinates and must internally compute these reward-relevant quantities. The higher success of PUMA may reflect direct reward-feature alignment rather than an intrinsically better prior representation. A control that provides Cartesian variants with polar features computed from their predicted coordinates, or that gives PUMA a Cartesian reward, would isolate the representation effect.

minor comments (5)

[Section III-B.2] Clarify how f_input_t, the estimated velocity, and the environment latent are concatenated and fed to the actor. The sentence 'These representations are then concatenated with the current observations' is ambiguous about which inputs go to which network head.
[Eq. (3), Fig. 6b] Define T and state its value. In Fig. 6b, define what '5K' and '8K' mean and whether curves are averaged over seeds; add error bars or shading.
[Table I, Eq. (7)] The group weights 'Tracking Group 3', 'Foothold Group 1.5', and 'Regularization Group 1.0' are not explicitly tied to w_task, w_foothold, and w_style in Eq. (7). Specify how group weights and per-term weights combine.
[Section IV-C] State unambiguously whether the '10 trials' are per terrain condition or per method overall, and report trial-level outcomes or a table with per-trial results. The captions in Fig. 5 are abbreviated and hard to interpret.
[Reproducibility] No code or exact hyperparameters are released. Provide learning rate, PPO clip, annealing total steps T, d_safe, sparse-reward threshold ϵ, and domain-randomization ranges for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: central claim rests on empirical ablation with external baselines.

full rationale

PUMA is an empirical RL paper; it does not claim to derive a prediction from first principles. The load-bearing comparison is the ablation of foothold prior representation (polar vs explicit/implicit Cartesian, w/o prior) and against external baselines (PIE [2], Extreme Parkour [3]) in Table II, all measured by simulation success/traverse rate and real-world trials. The ground-truth f_t (Eq. 1 and definition) is an oracle geometric label used as both regression target and reward shaping (Eqs. 4-6); this makes the prior part of the training objective, but it is not a case of predicting a fitted quantity: the estimator \hat f_t is trained to match f_t, and the policy is evaluated on task success, not on matching f_t. The w/o PAS ablation (Table IIc) shows the self-cited PAS [23] is not load-bearing for performance, and the PIE estimator [2] is an architectural starting point, not an invoked uniqueness or force result. The acknowledged failure case 'Foothold estimation error causing missed wall contact' (Fig. 7d) is a robustness limitation, not a circular step. No equation reduces to its inputs by construction; hence no significant circularity.

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

The central claim is an empirical method result. The main burden is carried by hand-chosen definitions and assumptions: the ground-truth foothold sample generator, the representation choice, reward weights, PAS schedule, and sim-to-real assumptions. There are no free parameters in the physics sense; the listed hyperparameters are necessary to reproduce the system exactly.

free parameters (7)

d_safe (edge safety threshold) = not specified
Filters candidate footholds near edges in Eq. (1); hand-chosen and affects which footholds define the prior.
Foothold sampling interval = 1 m
Candidates sampled every 1 m along commanded direction (Section III-A); defines resolution of the prior.
Sparse reward threshold epsilon = not specified
Defines when sparse foothold reward fires in Eq. (6); must be tuned for the robot foot size and terrain.
Annealing total steps T = not explicitly stated; ablation compares 5K and 8K
Controls PAS schedule p_t = 1 - cos(pi t/(2T)); convergence and final performance depend on it.
Reward weights = Table I (e.g., foothold group weights 1.0; regularization weights -0.05 to -2.5e-7)
Hand-tuned weights balancing tracking, foothold, and regularization; central to the learned behavior.
PD gains on real robot = P=20, D=0.5
Converts policy joint targets to motor torques in deployment; chosen for the Lite3 hardware.
Domain randomization ranges = not specified
Sim-to-real transfer relies on randomized observations/physics/depth delay; ranges are not reported, so exact simulation cannot be reproduced.

assumptions (6)

ad hoc to paper Foothold candidates exist at 1 m intervals along the commanded direction, including centers of exploitable inclined walls.
Introduced in Section III-A as the definition of ground-truth footholds; not derived from terrain or perception; the whole prior depends on this oracle-like choice.
domain assumption Depth images plus proprioceptive history are sufficient to regress the egocentric polar foothold prior f_t.
The regression estimator in Section III-B.2 must recover f_hat from partial observations; if the mapping is ambiguous, the policy receives noisy guidance (cf. Fig. 7d).
domain assumption Privileged critic with ground-truth footholds and heightfield provides a valid learning signal for the depth-only actor.
Asymmetric actor-critic training (Section III-B) assumes privileged information can be distilled into the actor; standard in the literature but unproven here.
domain assumption PPO with multi-critic advantages converges to a good policy under the specified rewards.
No convergence or stability proof is offered; the paper relies on empirical training curves (Fig. 6).
domain assumption Sim-to-real transfer is achieved with domain randomization and depth delay.
Section IV-A states randomization and visual delay are applied; limited real-world trials are the only evidence the gap is closed.
ad hoc to paper PAS schedule p_t = 1 - cos(pi t/(2T)) is an appropriate annealing shape.
The schedule is chosen in Eq. (3) and only its duration is ablated, not its functional form.

how reviews work

0 comments

Cite this review

Pith. "Pith review of PUMA: Perception-driven Unified Foothold Prior for Mobility Augmented Quadruped Parkour." pith.science (2026). https://pith.science/paper/DIZHNNVY

@misc{pith2026260115995,
  author       = {Pith},
  title        = {Pith review of: PUMA: Perception-driven Unified Foothold Prior for Mobility Augmented Quadruped Parkour},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DIZHNNVY}},
  note         = {Machine review of arXiv:2601.15995}
}

read the original abstract

Parkour tasks for quadrupeds have emerged as a promising benchmark for agile locomotion. While human athletes can effectively perceive environmental characteristics to select appropriate footholds for obstacle traversal, endowing legged robots with similar perceptual reasoning remains a significant challenge. Existing methods often rely on hierarchical controllers that follow pre-computed footholds, thereby constraining the robot's real-time adaptability and the exploratory potential of reinforcement learning. To overcome these challenges, we present PUMA, an end-to-end learning framework that integrates visual perception and foothold priors into a single-stage training process. This approach leverages terrain features to estimate egocentric polar foothold priors, composed of relative distance and heading, guiding the robot in active posture adaptation for parkour tasks. Extensive experiments conducted in simulation and real-world environments across various discrete complex terrains, demonstrate PUMA's exceptional agility and robustness in challenging scenarios.

Figures

Figures reproduced from arXiv: 2601.15995 by the authors.

**Figure 1.** PUMA enables quadruped robots to fuse proprioception with visual perception to estimate adaptive footholds for traversing complex discrete terrains. Top Row: The robot twists its posture to forcefully kick off the inclined wall, propelling itself across a wide gap. Middle Row: The robot sequentially traverses uneven stepping stones. Bottom Row: The robot leverages an inclined wall to surmount a high platform. Abstra… view at source ↗

**Figure 2.** Overview of PUMA training framework. A velocity-tracking locomotion policy takes proprioception and depth [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Terrain difficulty gradually increasing from the left [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Temporal evolution of body velocity and total contact [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Real-world experimental results: (left) Wall-assisted Gap terrain, featuring stepping walls at 60° and 80° angles [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: Training curves analysis. (a) PUMA obtains sig [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

**Figure 7.** Figure 7: Typical failure cases in the real-world experiment: [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 1 linked inside Pith

[1]

High-speed control and navigation for quadrupedal robots on complex and discrete terrain,

H. Kim, H. Oh, J. Park, Y . Kim, D. Youm, M. Jung, M. Lee, and J. Hwangbo, “High-speed control and navigation for quadrupedal robots on complex and discrete terrain,”Science Robotics, vol. 10, no. 102, p. eads6192, 2025

2025
[2]

Pie: Parkour with implicit-explicit learning framework for legged robots,

S. Luo, S. Li, R. Yu, Z. Wang, J. Wu, and Q. Zhu, “Pie: Parkour with implicit-explicit learning framework for legged robots,”IEEE Robotics and Automation Letters, 2024

2024
[3]

Extreme parkour with legged robots,

X. Cheng, K. Shi, A. Agarwal, and D. Pathak, “Extreme parkour with legged robots,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 11 443–11 450

2024
[4]

Robot parkour learning,

Z. Zhuang, Z. Fu, J. Wang, C. G. Atkeson, S. Schwertfeger, C. Finn, and H. Zhao, “Robot parkour learning,” inConference on Robot Learning. PMLR, 2023, pp. 73–92

2023
[5]

Attention-based map encoding for learning generalized legged loco- motion,

J. He, C. Zhang, F. Jenelten, R. Grandia, M. B ¨acher, and M. Hutter, “Attention-based map encoding for learning generalized legged loco- motion,”Science Robotics, vol. 10, no. 105, p. eadv3604, 2025

2025
[6]

Start: Traversing sparse footholds with terrain reconstruction,

R. Yu, Q. Wang, H. Li, Z. Jun, Z. Wang, J. Wu, and Q. Zhu, “Start: Traversing sparse footholds with terrain reconstruction,”IEEE Robotics and Automation Letters, pp. 1–8, 2025

2025
[7]

Anymal parkour: Learning agile navigation for quadrupedal robots,

D. Hoeller, N. Rudin, D. Sako, and M. Hutter, “Anymal parkour: Learning agile navigation for quadrupedal robots,”Science Robotics, vol. 9, no. 88, p. eadi7566, 2024

2024
[8]

Coupling vision and proprioception for navigation of legged robots,

Z. Fu, A. Kumar, A. Agarwal, H. Qi, J. Malik, and D. Pathak, “Coupling vision and proprioception for navigation of legged robots,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 17 273–17 283

2022

Show all 27 references

[9]

Learning vision-guided quadrupedal locomotion end-to-end with cross-modal transformers,

R. Yang, M. Zhang, N. Hansen, H. Xu, and X. Wang, “Learning vision-guided quadrupedal locomotion end-to-end with cross-modal transformers,” inDeep RL Workshop NeurIPS 2021

2021
[10]

Legged locomotion in challenging terrains using egocentric vision,

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

2023
[11]

Dtc: Deep tracking control,

F. Jenelten, J. He, F. Farshidian, and M. Hutter, “Dtc: Deep tracking control,”Science Robotics, vol. 9, no. 86, p. eadh5401, 2024

2024
[12]

Hierar- chical reinforcement learning and value optimization for challenging quadruped locomotion,

J. Coholich, M. A. Murtaza, S. Hutchinson, and Z. Kira, “Hierar- chical reinforcement learning and value optimization for challenging quadruped locomotion,”arXiv preprint arXiv:2506.20036, 2025

2025 arXiv
[13]

Optimized jumping on the mit cheetah 3 robot,

Q. Nguyen, M. J. Powell, B. Katz, J. Di Carlo, and S. Kim, “Optimized jumping on the mit cheetah 3 robot,” in2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 7448–7454

2019
[14]

Continuous jumping for legged robots on stepping stones via trajectory optimization and model predictive control,

C. Nguyen, L. Bao, and Q. Nguyen, “Continuous jumping for legged robots on stepping stones via trajectory optimization and model predictive control,” in2022 IEEE 61st Conference on Decision and Control (CDC). IEEE, 2022, pp. 93–99

2022
[15]

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
[16]

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
[17]

Adversarial motion priors make good substitutes for complex 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 complex reward functions,” in2022 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS). IEEE, 2022, pp. 25–32

2022
[18]

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
[19]

Spatio-temporal motion retargeting for quadruped robots,

T. Yoon, D. Kang, S. Kim, J. Cheng, M. Ahn, S. Coros, and S. Choi, “Spatio-temporal motion retargeting for quadruped robots,”IEEE Transactions on Robotics, 2025

2025
[20]

Dfm: Deep fourier mimic for expressive dance motion learning,

R. Watanabe, C. Li, and M. Hutter, “Dfm: Deep fourier mimic for expressive dance motion learning,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 9644– 9650

2025
[21]

Dreamwaq: Learning robust quadrupedal locomotion with implicit terrain imagination via deep reinforcement learning,

I. M. A. Nahrendra, B. Yu, and H. Myung, “Dreamwaq: Learning robust quadrupedal locomotion with implicit terrain imagination via deep reinforcement learning,” in2023 IEEE International Conference on Robotics and Automation, ICRA 2023. Institute of Electrical and Electronics En...

2023
[22]

Asymmetric actor critic for image-based robot learning,

L. Pinto, M. Andrychowicz, P. Welinder, W. Zaremba, and P. Abbeel, “Asymmetric actor critic for image-based robot learning,” inProceed- ings of Robotics: Science and Systems, Pittsburgh, Pennsylvania, June 2018

2018
[23]

Saro: Space- aware robot system for terrain crossing via vision-language model,

S. Zhu, D. Li, L. Mou, Y . Liu, N. Xu, and H. Zhao, “Saro: Space- aware robot system for terrain crossing via vision-language model,” pp. 14 820–14 827, 2025

2025
[24]

Robotkeyframing: Learning locomotion with high-level objectives via mixture of dense and sparse rewards,

F. Zargarbashi, J. Cheng, D. Kang, R. Sumner, and S. Coros, “Robotkeyframing: Learning locomotion with high-level objectives via mixture of dense and sparse rewards,” inConference on Robot Learning. PMLR, 2025, pp. 916–932

2025
[25]

Multi-critic learning for whole-body end-effector twist tracking,

A. E. Vijayan, A. Cramariuc, M. Risiglione, C. Gehring, and M. Hut- ter, “Multi-critic learning for whole-body end-effector twist tracking,” inConference on Robot Learning. PMLR, 2025, pp. 1470–1485

2025
[26]

Multi-critic actor learning: Teaching rl policies to act with style,

S. Mysore, G. Cheng, Y . Zhao, K. Saenko, and M. Wu, “Multi-critic actor learning: Teaching rl policies to act with style,” inInternational Conference on Learning Representations, 2022

2022
[27]

Learning Humanoid Standing-up Control across Diverse Postures,

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,” inProceedings of Robotics: Science and Systems, LosAngeles, CA, USA, June 2025

2025

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

High-speed control and navigation for quadrupedal robots on complex and discrete terrain,

H. Kim, H. Oh, J. Park, Y . Kim, D. Youm, M. Jung, M. Lee, and J. Hwangbo, “High-speed control and navigation for quadrupedal robots on complex and discrete terrain,”Science Robotics, vol. 10, no. 102, p. eads6192, 2025

2025

[2] [2]

Pie: Parkour with implicit-explicit learning framework for legged robots,

S. Luo, S. Li, R. Yu, Z. Wang, J. Wu, and Q. Zhu, “Pie: Parkour with implicit-explicit learning framework for legged robots,”IEEE Robotics and Automation Letters, 2024

2024

[3] [3]

Extreme parkour with legged robots,

X. Cheng, K. Shi, A. Agarwal, and D. Pathak, “Extreme parkour with legged robots,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 11 443–11 450

2024

[4] [4]

Robot parkour learning,

Z. Zhuang, Z. Fu, J. Wang, C. G. Atkeson, S. Schwertfeger, C. Finn, and H. Zhao, “Robot parkour learning,” inConference on Robot Learning. PMLR, 2023, pp. 73–92

2023

[5] [5]

Attention-based map encoding for learning generalized legged loco- motion,

J. He, C. Zhang, F. Jenelten, R. Grandia, M. B ¨acher, and M. Hutter, “Attention-based map encoding for learning generalized legged loco- motion,”Science Robotics, vol. 10, no. 105, p. eadv3604, 2025

2025

[6] [6]

Start: Traversing sparse footholds with terrain reconstruction,

R. Yu, Q. Wang, H. Li, Z. Jun, Z. Wang, J. Wu, and Q. Zhu, “Start: Traversing sparse footholds with terrain reconstruction,”IEEE Robotics and Automation Letters, pp. 1–8, 2025

2025

[7] [7]

Anymal parkour: Learning agile navigation for quadrupedal robots,

D. Hoeller, N. Rudin, D. Sako, and M. Hutter, “Anymal parkour: Learning agile navigation for quadrupedal robots,”Science Robotics, vol. 9, no. 88, p. eadi7566, 2024

2024

[8] [8]

Coupling vision and proprioception for navigation of legged robots,

Z. Fu, A. Kumar, A. Agarwal, H. Qi, J. Malik, and D. Pathak, “Coupling vision and proprioception for navigation of legged robots,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 17 273–17 283

2022