REVIEW 3 major objections 4 minor 26 references
Quadrupedal Robot Skateboard Mounting via Reverse Curriculum Learning
T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read A quadrupedal robot can learn to mount a skateboard by training backward from the goal, and the resulting policy transfers to a moving board.
desk verdict A genuinely new task (quadruped mounting a skateboard) with a sensible reverse-curriculum approach, but the mobile-board transfer claim is internally inconsistent and the evaluation is too qualitative to support the core claim. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing mechanism is reverse curriculum learning: the agent starts each episode in the already-mounted state, learns to stay balanced, then the spawn region is widened in stages, up to a 60 cm by 60 cm area around the board's center of mass, with a final stage that lets the board move freely. The policy is a multilayer perceptron trained with Proximal Policy Optimization, using sparse contact rewards for each foot on the deck, dense distance- and orientation-alignment rewards with exponential kernels, and penalties for board overturning and displacement. The skateboard model treats the truck as a roughly 1:1 map from deck tilt angle $\alpha$ to wheel steering angle $\beta$ ($\beta \approx \alpha$) and models the bushings as a torsional spring, which makes the board's response to the robot's weight shift physically reactive.
What would settle it
Run the trained policy on a physical quadruped facing a real skateboard under the same starting offsets used in simulation; if the success rate is near zero because the board shifts during foot placement, the transfer claim fails. A cheaper in-simulation test is to vary skateboard mass, wheel friction, and bushing stiffness beyond the training distribution and watch for success collapse.
Extended reading notes
Core claim
The paper's central claim is that a quadrupedal robot can be trained entirely in simulation to mount a skateboard from various starting positions, and that a policy trained with the skateboard rigidly fixed in the world frame performs well when the board is allowed to move freely, with no fine-tuning on the moving board. The authors frame this as closing the gap between walking and riding: earlier skateboarding systems begin with the robot already on the board, leaving the mounting phase open. They report that the trained policy mounts the board in roughly three seconds from first contact, and that a forward curriculum baseline, which spawns the robot near the board and lets it explore freely, never progresses beyond aligning itself and placing two feet on the deck.
Load-bearing premise
The central claim assumes that a policy trained with the skateboard fixed in place generalizes to a skateboard that can roll and turn, even though the policy never experiences board motion for most of its training.
Editorial extensions
If this is right
- A complete autonomous skateboarding pipeline becomes possible by appending the learned mounting skill to an existing riding controller.
- The same reverse-curriculum schedule could be applied to other contact-rich tasks with a known goal configuration, such as climbing onto a platform or entering a narrow bay.
- Operators no longer need to position the robot precisely before the mounting attempt, since the trained policy tolerates variation in starting position and orientation.
- Because the policy transfers from a fixed to a moving board, training effort can be concentrated on the contact-rich mounting phase rather than on modelling the board's rolling dynamics.
- The observation of board-edge points at 10 cm intervals suggests the policy can accommodate skateboards of different shapes without changing the network.
Reading between the lines
- Beyond the paper: the fixed-to-moving transfer suggests the learned policy relies mostly on board-relative features such as edge points, contact state, and orientation, so the same curriculum may work for other mobile objects whose target configuration is known; the authors do not test this.
- Beyond the paper: a real-world transfer test would be the first decisive check, because the simulation's no-slip wheel contact and simplified truck model may be more forgiving than a physical skateboard; the authors list real-world validation only as future work.
- Beyond the paper: the reward design is not ablated, so it is unknown whether the sparse foot-contact rewards or the dense distance and orientation terms do the heavy lifting; isolating each term would make the method more portable.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a Reverse Curriculum Reinforcement Learning approach to train a simulated Unitree A1 quadruped to mount a skateboard. The training starts with the robot already positioned on a rigidly fixed skateboard, then gradually expands the spawn distribution around the board, and finally addresses an unfixed (mobile) skateboard. The authors report that the trained policy can mount the skateboard from various starting positions and that a policy trained on a fixed board performs well on a mobile board. The paper includes training curves, a qualitative description of the mounting sequence, and a public repository with code and trained models.
Significance. If the central claims are substantiated, the work would provide a useful demonstration of autonomous skateboard mounting, a skill that prior quadrupedal skateboarding work has not addressed. The main positive elements are the open-source release of code and trained models, the use of a physically plausible skateboard model, and the application of a reverse curriculum to a contact-rich underactuated task. However, the paper's current evidence is largely qualitative, and the transfer claim for the mobile board is internally ambiguous. The contribution is more of an empirical demonstration than a methodological advance, but it could be a valuable data point for the community if the evaluation is tightened.
major comments (3)
- [Section IV-F, Section IV-B] The transfer claim is internally inconsistent. Section IV-B states that the skateboard was allowed to move freely 'only in the final stage of training,' which implies the final policy was trained on a mobile board. Section IV-F, however, states that the agent trained on a fixed skateboard 'demonstrated good performance without requiring additional training,' and then immediately introduces a new episode termination condition and negative rewards for overturning and displacement. If the final policy was fine-tuned with these modified rewards, the claim of zero-shot transfer is false. If it was not fine-tuned, the added negative rewards could not have influenced the policy, and the evaluation used a different termination condition than the one used during training. The manuscript must clarify exactly which policy was used for the mobile-board results, whether training was resumed, and what role the new termination condition and rewards played.
- [Section IV-H, Figure 2] The results section provides no quantitative success metric. The claim that the robot 'consistently mounts the skateboard in approximately 3 seconds from first contact' is not supported by any tables, success rates, episode counts, or repeated-seed statistics. Figure 2 shows only mean episode reward curves without variance or separate success-rate curves. Since the central contribution is an empirical policy result, the paper needs quantitative evaluation: success rate over a fixed number of evaluation episodes, distribution of mounting times, and variability across training seeds.
- [Section IV-E] The forward-curriculum baseline is described only narratively: 'the most promising result observed with this method was successful alignment of the robot's center of mass and coordinate axes with those of the skateboard, along with the placement of two legs on the board.' No reward curves, success rates, or quantitative comparisons are given for the baseline. The claimed advantage of the proposed reverse curriculum over this baseline is therefore not evidenced. The paper should include at least final reward values or success rates for both methods under identical evaluation conditions.
minor comments (4)
- [Section IV-D, Table I] Table I lists the supplementary reward terms but does not report their weights or the values of the free parameters sigma, d_th, and k_angular beyond a passing mention in Section IV-B. Since the reward weights directly determine the learned behavior, they should be reported in the text or table even if they are available in the code repository.
- [Section IV-F] The phrase 'The weights of the awards and all other training parameters have not changed' is unclear: it could mean that the reward weights were kept constant across curriculum stages, but the reward terms themselves changed. Please state explicitly which reward terms and weights were active in each curriculum stage.
- [Section IV-C] The observation space includes foot contact status with the skateboard, and the authors note that this is difficult to obtain in a real-world setting and is planned to be removed in future work. This is a relevant limitation for the sim-to-real potential of the method and should be acknowledged in the conclusion as well.
- [Figure 2] The training curves in Figure 2 are not labeled with the corresponding curriculum stages on the plot itself, and the axes are not described in the caption. Please make the figure self-contained by marking the stages and defining the reward metric.
Circularity Check
No circularity found: the RL training/evaluation loop is self-contained, and no prediction reduces to a fitted parameter or self-citation chain.
full rationale
The paper's central output is a trained policy evaluated in the same Isaac Lab simulator used for training. This is ordinary reinforcement-learning evaluation, not a derivation of a prediction from its own input: the reward terms define the optimization objective, and the policy is optimized against that objective, not fitted to the reported success measure. No parameter is calibrated on a subset of the data and then used to predict a closely related quantity. No load-bearing claim is justified by a self-citation: the cited reverse-curriculum method [24], Isaac Lab [26], and prior skateboarding works [5]-[10] are all external, and none of the authors' own prior results are invoked to force a conclusion. The only structural concern is the ambiguity in Section IV-F, where the paper first says the fixed-board agent transferred 'without requiring additional training' and then introduces a new overturning termination condition and negative rewards for overturning and displacement; Section IV-B also states the board was freed 'only in the final stage of training.' These statements are internally inconsistent about whether or how the final policy was refined, but inconsistency is not circularity. Under the hard rule that circularity must be exhibited as a specific reduction (Eq. X = Eq. Y by construction, or a fitted parameter renamed as a prediction), no such reduction exists in the manuscript. Therefore the circularity score is 0.
Assumptions & free parameters
free parameters (4)
- Reward shaping scaling factor sigma and distance threshold d_th
- Reward weights for supplementary reward terms
- Curriculum stage spawn distributions
- Skateboard truck stiffness k_angular =
2 N·m/rad
assumptions (5)
- standard math Rigid body dynamics equations (3) and (4) govern the skateboard and robot motion.
- domain assumption Skateboard truck kinematics satisfy beta approximately equals alpha (Eq. 1) with a 45 degree kingpin angle.
- domain assumption Bushing behavior is modeled as a torsional spring with damping, implemented via a P controller (Eq. 2).
- domain assumption Wheel-ground and foot-deck contacts obey the Coulomb friction constraints in Eqs. (5) and (6).
- standard math PPO with the stated hyperparameters converges to a policy that approximately maximizes the cumulative reward.
Cite this review
Pith. "Pith review of Quadrupedal Robot Skateboard Mounting via Reverse Curriculum Learning." pith.science (2026). https://pith.science/paper/4RG7XC2L
@misc{pith2026250506561,
author = {Pith},
title = {Pith review of: Quadrupedal Robot Skateboard Mounting via Reverse Curriculum Learning},
year = {2026},
howpublished = {\url{https://pith.science/paper/4RG7XC2L}},
note = {Machine review of arXiv:2505.06561}
}
read the original abstract
The aim of this work is to enable quadrupedal robots to mount skateboards using Reverse Curriculum Reinforcement Learning. Although prior work has demonstrated skateboarding for quadrupeds that are already positioned on the board, the initial mounting phase still poses a significant challenge. A goal-oriented methodology was adopted, beginning with the terminal phases of the task and progressively increasing the complexity of the problem definition to approximate the desired objective. The learning process was initiated with the skateboard rigidly fixed within the global coordinate frame and the robot positioned directly above it. Through gradual relaxation of these initial conditions, the learned policy demonstrated robustness to variations in skateboard position and orientation, ultimately exhibiting a successful transfer to scenarios involving a mobile skateboard. The code, trained models, and reproducible examples are available at the following link: https://github.com/dancher00/quadruped-skateboard-mounting
Figures
Reference graph
Works this paper leans on
-
[1]
M. H. Raibert, Legged robots that balance . MIT press, 1986
1986
-
[2]
Actuator design for high force proprioceptive control in fast legged locomotion,
S. Seok, A. Wang, D. Otten, and S. Kim, “Actuator design for high force proprioceptive control in fast legged locomotion,” in 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems . IEEE, 2012, pp. 1970–1975
work page 2012
-
[3]
A low cost modular actuator for dynamic robots,
B. G. Katz, “A low cost modular actuator for dynamic robots,” Ph.D. dissertation, Massachusetts Institute of Technology, 2018
work page 2018
-
[4]
Stanford doggo: An open-source, quasi-direct-drive quadruped,
N. Kau, A. Schultz, N. Ferrante, and P. Slade, “Stanford doggo: An open-source, quasi-direct-drive quadruped,” in 2019 International conference on robotics and automation (ICRA) . IEEE, 2019, pp. 6309–6315
work page 2019
-
[5]
A bipedal walking robot that can fly, slackline, and skateboard,
K. Kim, P. Spieler, E.-S. Lupu, A. Ramezani, and S.-J. Chung, “A bipedal walking robot that can fly, slackline, and skateboard,” Science Robotics, vol. 6, no. 59, p. eabf8136, 2021
work page 2021
-
[6]
Learning skateboarding for humanoid robots through massively parallel rein- forcement learning,
W. Thibault, V . Rajendran, W. Melek, and K. Mombaur, “Learning skateboarding for humanoid robots through massively parallel rein- forcement learning,” arXiv preprint arXiv:2409.07846 , 2024
arXiv 2024
-
[7]
Discrete-time hybrid automata learning: Legged locomotion meets skateboarding,
H. Liu, S. Teng, B. Liu, W. Zhang, and M. Ghaffari, “Discrete-time hybrid automata learning: Legged locomotion meets skateboarding,” arXiv preprint arXiv:2503.01842 , 2025
arXiv 2025
-
[8]
Optimization based dynamic skateboarding of quadrupedal robot,
Z. Xu, M. Al-Khulaqui, H. Ma, J. Wang, Q. Xin, Y . You, M. Zhou, D. Xiang, and S. Zhang, “Optimization based dynamic skateboarding of quadrupedal robot,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 8058–8064
work page 2024
Show all 26 references
-
[9]
Skating with a force controlled quadrupedal robot,
M. Bjelonic, C. D. Bellicoso, M. E. Tiryaki, and M. Hutter, “Skating with a force controlled quadrupedal robot,” in 2018 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS) . IEEE, 2018, pp. 7555–7561
2018
-
[10]
Roller-skating of mammalian quadrupedal robot with passive wheels inspired by human,
J. Chen, K. Xu, and X. Ding, “Roller-skating of mammalian quadrupedal robot with passive wheels inspired by human,” IEEE/ASME Transactions on Mechatronics , vol. 26, no. 3, pp. 1624– 1634, 2020
2020
-
[11]
Representation-free model predictive control for dynamic motions in quadrupeds,
Y . Ding, A. Pandala, C. Li, Y .-H. Shin, and H.-W. Park, “Representation-free model predictive control for dynamic motions in quadrupeds,” IEEE Transactions on Robotics , vol. 37, no. 4, pp. 1154–1171, 2021
2021
-
[12]
Variational contact-implicit tra- jectory optimization,
Z. Manchester and S. Kuindersma, “Variational contact-implicit tra- jectory optimization,” in Robotics Research: The 18th International Symposium ISRR . Springer, 2019, pp. 985–1000
2019
-
[13]
Rapid locomotion via reinforcement learning,
G. B. Margolis, G. Yang, K. Paigwar, T. Chen, and P. Agrawal, “Rapid locomotion via reinforcement learning,” The International Journal of Robotics Research, vol. 43, no. 4, p. 572–587, 2024
2024
-
[14]
Agile but safe: Learning collision-free high speed legged locomotion,
T. He, C. Zhang, W. Xiao, G. He, C. Liu, and G. Shi, “Agile but safe: Learning collision-free high speed legged locomotion,” Robotics: Science and Systems (RSS) , 2024
2024
-
[15]
Quadruped robot traversing 3d complex environments with limited perception,
Y . Cheng, H. Liu, G. Pan, H. Liu, and L. Ye, “Quadruped robot traversing 3d complex environments with limited perception,” 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), p. 9074–9081, 2024
2024
-
[16]
Dribble bot: Dynamic legged manipulation in the wild,
Y . Ji, G. B. Margolis, and P. Agrawal, “Dribble bot: Dynamic legged manipulation in the wild,” 2023 IEEE International Conference on Robotics and Automation (ICRA) , p. 5155–5162, 2023
2023
-
[17]
Extreme parkour with legged robots,
X. Cheng, K. Shi, A. Agarwal, and D. Pathak, “Extreme parkour with legged robots,” 2024 IEEE International Conference on Robotics and Automation (ICRA) , p. 11443–11450, 2024
2024
-
[18]
Actor-critic reinforcement learning for control with stability guarantee,
M. Han, L. Zhang, J. Wang, and W. Pan, “Actor-critic reinforcement learning for control with stability guarantee,” IEEE Robotics and Automation Letters, 2020
2020
-
[19]
Reinforcement learning for control with stability guarantee,
H. Gong, J. Zhang, T. Yan, Y . Zhan, and Y . Xia, “Reinforcement learning for control with stability guarantee,” 2024. [Online]. Available: https://openreview.net/forum?id=vBNTeQ7dPP
2024
-
[20]
Hierarchical reinforcement learning for quadruped locomotion,
D. Jain, A. Iscen, and K. Caluwaerts, “Hierarchical reinforcement learning for quadruped locomotion,” 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 7551–7557, 2019
2019
-
[21]
Reinforcement learning for versatile, dynamic, and robust bipedal locomotion control,
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 , 2024
2024
-
[22]
Robust recovery controller for a quadrupedal robot using deep reinforcement learning,
J. Lee, J. Hwangbo, and M. Hutter, “Robust recovery controller for a quadrupedal robot using deep reinforcement learning,” arXiv preprint arXiv:1901.0751, 2019
1901
-
[23]
Learning symbolic representations of hybrid dynamical systems,
D. L. Ly and H. Lipson, “Learning symbolic representations of hybrid dynamical systems,” Journal of Machine Learning Research , vol. 13, no. 115, pp. 3585–3618, 2012. [Online]. Available: http://jmlr.org/papers/v13/ly12a.html
2012
-
[24]
Re- verse curriculum generation for reinforcement learning,
C. Florensa, D. Held, M. Wulfmeier, M. Zhang, and P. Abbeel, “Re- verse curriculum generation for reinforcement learning,” 1st Annual Conference on Robot Learning (CoRL 2017) , no. 78, pp. 482–495, 2017
2017
-
[25]
Z. Fan. (2024) robot lab: Rl extension library for robots, based on isaaclab. [Online]. Available: https://github.com/fan-ziqi/robot lab
2024
-
[26]
Orbit: A unified simulation framework for interactive robot learning environments,
M. Mittal, C. Yu, Q. Yu, J. Liu, N. Rudin, D. Hoeller, J. L. Yuan, R. Singh, Y . Guo, H. Mazhar, A. Mandlekar, B. Babich, G. State, M. Hutter, and A. Garg, “Orbit: A unified simulation framework for interactive robot learning environments,” IEEE Robotics and Automa- tion Lette...
2023
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.