Pith. sign in

REVIEW 3 major objections 6 minor 36 references

A Modular Residual Learning Framework to Enhance Model-Based Approach for Robust Locomotion

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

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

Pith's one-line read By pairing a reinforcement-learned footstep correction with a supervised-learned dynamics correction, this paper argues, a nominal convex MPC quadruped controller can handle heavy payloads, external kicks, and out-of-distribution…

desk verdict A genuinely modular residual-learning result with credible simulation ablations and a real hardware demo, but the residual-dynamics loop is undertested in closed loop and the hardware trials are single shots. read the letter →

arxiv 2507.18138 v1 pith:7NWU5SFV submitted 2025-07-24 cs.RO

Min-Gyu Kim , Dongyun Kang , Hajun Kim , Hae-Won Park This is my paper · ORCID
classification cs.RO
keywords leggedrobotsquadrupedlocomotionmodelpredictivecontrolresiduallearningreinforcementsupervisedrobustout-of-distributiongeneralization
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

This paper tries to show that a conventional model-based quadruped controller, built around a simplified rigid-body model and convex model-predictive control, can be made reliable under exactly the uncertainties that break it: heavy payloads, external pushes, and uneven terrain. The method adds two small learned correction modules instead of replacing the controller. A reinforcement-learned module adjusts where each foot lands and how long each stance lasts, while a supervised-learned module predicts the difference between the nominal model and what actually happens, feeding that correction into the MPC's prediction. A sympathetic reader would care because the design keeps the predictable, constraint-respecting core of model-based control while shifting only the model-mismatch burden onto learning; the authors report that this reduces tuning sensitivity, improves training efficiency, and keeps a real robot balanced where the nominal controller fails.

What carries the argument

The central object is a pair of residual modules inserted into the nominal loop. A residual footstep module, trained with proximal policy optimization, outputs per-leg foothold and gait-phase corrections to the heuristic footstep planner. A residual dynamics module, trained with supervised learning, takes recent inertial, joint, and torque history and outputs the low-pass-filtered, 10 Hz difference between the nominal model's prediction and the observed state change. That residual term is added as a constant offset to the MPC's dynamics constraint over the 0.1 s horizon, so the optimization stays convex and fast. The design's work is to route each kind of model error to the appropriate learning paradigm, reinforcement learning for discrete contact choices and supervised regression for smooth continuous mismatch, while keeping the nominal controller as the safety and feasibility backbone.

What would settle it

During a repeated 6 kg payload and push-recovery trial, compare the network-predicted residual with the measured residual, defined as the difference between the nominal model's one-step prediction and the observed state change over the same control step. If the predicted residual is not significantly correlated with the measured residual at the moments when the robot regains balance, then the proposed compensation mechanism is not what is stabilizing the robot.

Watch

Extended reading notes

Core claim

The central claim is that model mismatch in legged locomotion can be decomposed into two separate correction problems, discrete contact decisions and continuous dynamics errors, and that each can be learned by the method best suited to it, with both residuals layered on top of a nominal convex MPC. Under this design the paper reports that the full system stays balanced through a 6 kg payload and added disturbances, while the vanilla MPC fails in the payload condition and a dynamics-only variant fails when kicks are added; in simulation it achieves the highest success rates across payload levels and the lowest RMS velocity error in most conditions, and it outperforms an end-to-end RL baseline when the payload is placed outside the training distribution. The paper also claims the residual correction term can be predicted from proprioceptive history alone, making hardware deployment feasible without accurate contact or ground-reaction-force estimates.

Load-bearing premise

The method relies on a neural network trained purely in simulation to predict the real robot's model error from its own recent sensor readings, and on that error staying roughly constant over the 0.1 s planning horizon; neither assumption is directly verified on the hardware.

Editorial extensions

If this is right

  • If the results hold, the same nominal MPC can be kept as a safety layer while small learned residuals compensate its model mismatch, avoiding a full end-to-end replacement and its sample costs.
  • Robots using this scheme should tolerate heavier payloads and pushes than the baseline MPC, and should do so without re-tuning the MPC's cost weights.
  • Because the dynamics residual is predicted from sensor history alone, deployment does not require high-fidelity estimates of ground reaction forces or contact states.
  • The modular split should make training converge faster than end-to-end RL or joint-space residual RL, as measured by the paper's learning curves.
  • Outside the training distribution, such as a payload attached to the legs rather than the trunk, velocity tracking should degrade less than for an end-to-end RL policy.

Reading between the lines

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

  • Editorial inference: because the residual is low-pass filtered at 10 Hz and treated as constant over the 0.1 s horizon, the method is tuned for slow, quasi-static model errors; high-frequency impacts or terrain compliance would be smoothed out, so extending to dynamic or deformable terrain would likely require a higher cutoff or an in-horizon residual model.
  • Editorial inference: the paper's decomposition suggests a recipe for other model-based controllers: separate discrete decisions such as footholds, gait, or contact sequence from continuous model error, and choose the learning tool per component rather than learning one monolithic policy.
  • Editorial inference: the residual dynamics network is trained on simulation labels and never validated against measured real-world residuals, so a direct extension would be to log the analytical residual on hardware and fine-tune or validate the regressor online; this would also reveal whether the demonstrated stability is truly due to the predicted residual.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The manuscript proposes a modular residual learning framework for quadruped locomotion. An RL-trained residual footstep and gait-phase module augments a heuristic footstep planner, while a supervised residual-dynamics network predicts low-frequency model discrepancy for a convex MPC built on the single rigid body model. The nominal MPC remains the core policy, and the learned modules act as auxiliary corrections. The authors evaluate the framework in simulation and on a Unitree Go1, reporting improved robustness to payloads and disturbances, reduced sensitivity to MPC cost weights, better out-of-distribution consistency than an end-to-end RL baseline, and higher learning efficiency than several residual-RL baselines.

Significance. If the claims hold, the framework is a pragmatic contribution to hybrid model-based/learning-based control: it retains the consistency and safety of convex MPC while using small, modular learned corrections to cover model mismatch. The modular division of labor (RL for discrete foothold/phase decisions, supervised learning for continuous dynamics residuals) is a sensible design that reduces the RL search space. Strengths of the paper include the relatively large simulation ablations (100 runs per condition in Table IV), multiple baselines including end-to-end RL and residual-augmented MPC variants, domain randomization, and a hardware feasibility demonstration. The main weakness is that the sim-to-real transfer of the learned residual-dynamics regressor is not directly validated: no closed-loop simulation with the neural regressor is reported, and the hardware evidence consists of single trials. The significance is therefore conditional on closing that validation gap.

major comments (3)
  1. [Sec. III-D, Eqs. (14)-(17); Fig. 2] The deployed residual-dynamics regressor is autoregressive: the observation o_RD in Eq. (17) contains \hat f_res,k-1, which at runtime is the network's own previous output rather than the analytical filtered residual from Eq. (15). The regression labels, by contrast, are computed from simulator states via Eq. (14) and filtered with Eq. (15). The manuscript does not state whether the simulation results in Table IV and Fig. 5 use the analytical residual or the learned network. If they use the analytical residual, the only direct evidence for the learned regressor is the single-trial hardware data in Table III and Fig. 6. Either way, the paper needs a closed-loop simulation evaluation with the neural regressor in the MPC loop, a quantitative comparison of its predictions to the analytical residual (for example, normalized RMSE and a spectral check), and evidence that the autoregressive input distribution at deployment does not drift outside the training distribution. This validation is load-bearing because the hardware payload and disturbance robustness results are attributed to the residual dynamics module.
  2. [Sec. IV-B, Table III; Sec. IV-D, Fig. 6] All hardware results are single trials without repeated runs, error bars, or explicit trial counts. Given that the central robustness claim is supported by these demonstrations, the paper should report at least the number of trials and success/failure outcomes per condition, or provide repeated trials with variance. This is especially important for the headline numbers in Table III, where the 0.0695 rad roll RMS under disturbance is a single number from one run, and for the out-of-distribution comparison in Fig. 6, which is presented without any quantitative error metric beyond the plotted trajectory.
  3. [Sec. IV-E, Table IV] The text states that the proposed method consistently achieves a high success rate, but at M_p = 1.75M the fpos-phase-dynRL baseline achieves 88% success versus 76% for res-all. Moreover, no confidence intervals or statistical tests are reported for the 100-run success rates, so the claimed superiority over this baseline is not established. The comparison should either be accompanied by uncertainty estimates or the claim should be softened to 'comparable success with lower RMS error and better learning efficiency.'
minor comments (6)
  1. [Sec. IV-C and Sec. IV-E] The success-rate criterion is not defined. Please specify what constitutes a failure (for example, trunk contact, fall detection, or timeout) and how it is detected in simulation and on hardware.
  2. [Sec. III-C, Eq. (9)] The reward weights c1 and c2 in Eq. (9) are never given. Reporting their values is necessary for reproducibility.
  3. [Sec. IV-E, Fig. 7] The learning curves appear to come from a single training run. Please report the number of seeds and show variance or error bars, since the learning-efficiency claim rests on this figure.
  4. [Sec. IV-E] The baseline implementations (fpos-phase-dynRL, jpos-dynRL, phase, resdyn-window) are described too briefly to be reproduced. Please provide network architectures, input/output dimensions, and training hyperparameters, or cite public code.
  5. [Sec. III-D] Please clarify how the 10 million residual-dynamics training samples were collected: which policy was used, at what sampling rate, and whether the observation in Eq. (17) during training contains the analytical \hat f_res,k-1 or the network's own previous prediction.
  6. [Sec. IV-D] The out-of-distribution hardware test in Fig. 6 is qualitative. Reporting velocity-tracking RMS error for both controllers with and without the sandbag would make the comparison quantitative.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: residual modules are trained on simulated mismatch labels and evaluated on external/OOD benchmarks; no prediction reduces to a fitted input by construction.

full rationale

The derivation chain is self-contained. The residual dynamics label (Eq. 14) is defined as the difference between the simulated state evolution and the nominal single-rigid-body model; it is not fitted from the outcome metrics being claimed. The supervised regressor (Eq. 16) is trained on these simulated labels, but the reported robustness results are obtained under payloads and disturbances outside the training distribution (e.g., a 6 kg hardware payload vs. the [-1, 5] kg randomization range) and are compared against external baselines (vanilla-MPC, baseline-RL, Chen et al., Yang et al., Sun et al.). The reward function (Eq. 9) being similar to the MPC cost (Eq. 21) is reward shaping, not a fitted parameter renamed as prediction; it cannot by itself force the footstep policy to stabilize the robot under unseen perturbations. Self-citations ([6], [31]) appear only in related-work context and are not load-bearing evidence for the central claims. The autoregressive deployment loop introduced by o_RD containing f_res_hat,k-1 (Eq. 17) is a sim-to-real validation gap, not a circular reduction: the network outputs are not identical to the training labels by construction, and the paper does not claim closed-loop prediction equivalence. No equation in the paper reduces to its own input, so no circularity step is exhibited.

Assumptions & free parameters 5 free parameters · 7 assumptions · 0 invented entities

The central contribution is architectural and introduces no new physical entities such as forces, particles, or latent mechanisms. The free parameters are tuning knobs, filter constants, and training-distribution ranges. The main unstated load-bearing premise is that a residual model learned from synthetic labels, low-pass filtered, and held constant over the MPC horizon captures the real robot's model mismatch well enough to improve stability.

free parameters (5)
  • Reward weights c1, c2 in Eq. (9)
    Defined in the reward function for the residual footstep module but never given numerical values. They shape the RL objective and the paper's convergence and performance claims depend on them.
  • MPC cost weights wp,phi, wv,omega, and wu = wp,phi in [10,30], wv,omega in [0.1,0.3], wu = 1e-5
    Randomized during training and tested on a grid in Fig. 5. The parameter-robustness claim is specifically about sensitivity to these hand-chosen weights.
  • Low-pass filter cutoff Fc and sampling rate Fs = Fc = 10 Hz, Fs = 1 kHz
    Chosen by hand in Section III-D to isolate slowly varying uncertainties. This determines which model discrepancies are compensated and which are ignored.
  • Cartesian PD gains KP, KD in Eq. (24)
    Used in the low-level feedback torque computation but values are not reported. They affect tracking of the residual footstep references.
  • Domain randomization ranges (Table I) = command +/-[2,1,1], mu in [0.4,1.0], payload [-1,5] kg, bumpiness [0,0.1] m, CoM offset +/-0.05 m, initial rotation…
    The training distribution is chosen by hand, and the out-of-distribution claims are defined relative to these ranges.
assumptions (7)
  • domain assumption The single rigid body model with Euler angles linearized at the operating point phi0 is an adequate nominal model for MPC.
    Eq. (4) treats inertia as approximately constant and ignores contact dynamics and whole-body effects; this simplification is the motivation for the residual modules.
  • domain assumption Footholds change instantaneously when the contact state changes.
    Section III-C states 'we assume that footholds instantaneously change whenever contact state changes', which defines the discrete footstep update in Eqs. (5) and (7).
  • ad hoc to paper Residual dynamics is constant over the 0.1 s MPC prediction horizon.
    Section III-D: 'we assume the residual terms remain constant over the MPC prediction horizon'. This is what keeps the MPC convex and fast, but it is not separately validated against time-varying disturbances.
  • ad hoc to paper Low-frequency uncertainty below 10 Hz dominates and captures payload and disturbance effects.
    Eq. (15) filters the residual at Fc = 10 Hz; any high-frequency model mismatch is invisible to the MPC correction and must be handled by the low-level controller.
  • ad hoc to paper A regressor trained on simulated residual-dynamics labels with proprioceptive inputs transfers to the real robot.
    Section III-D: training data are collected from simulation, and real-world residual dynamics are never measured. The hardware success is taken as indirect evidence of transfer.
  • domain assumption RAISIM contact dynamics with the domain randomization in Table I are a valid proxy for the real robot during training.
    All residual labels and RL rollouts come from the RAISIM simulator [36]; the sim-to-real gap is managed only through randomization.
  • domain assumption The baseline-RL controller trained with the same domain randomization is a fair end-to-end comparison.
    Section IV-D compares against a PPO policy with a learned state estimator; differences in convergence or OOD performance are attributed to the framework rather than to training budget or architecture differences.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Modular Residual Learning Framework to Enhance Model-Based Approach for Robust Locomotion." pith.science (2026). https://pith.science/paper/7NWU5SFV

@misc{pith2026250718138,
  author       = {Pith},
  title        = {Pith review of: A Modular Residual Learning Framework to Enhance Model-Based Approach for Robust Locomotion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7NWU5SFV}},
  note         = {Machine review of arXiv:2507.18138}
}
read the original abstract

This paper presents a novel approach that combines the advantages of both model-based and learning-based frameworks to achieve robust locomotion. The residual modules are integrated with each corresponding part of the model-based framework, a footstep planner and dynamic model designed using heuristics, to complement performance degradation caused by a model mismatch. By utilizing a modular structure and selecting the appropriate learning-based method for each residual module, our framework demonstrates improved control performance in environments with high uncertainty, while also achieving higher learning efficiency compared to baseline methods. Moreover, we observed that our proposed methodology not only enhances control performance but also provides additional benefits, such as making nominal controllers more robust to parameter tuning. To investigate the feasibility of our framework, we demonstrated residual modules combined with model predictive control in a real quadrupedal robot. Despite uncertainties beyond the simulation, the robot successfully maintains balance and tracks the commanded velocity.

Figures

Figures reproduced from arXiv: 2507.18138 by the authors.

Figure 1
Figure 1. Snapshots of experimental results. The quadrupedal [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An illustration of the overall architecture. Each residual module finds auxiliary actions for compensating model [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. An experimental result of robustness test with [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Results of control parameter robustness test. Each [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 4
Figure 4. Figure 4: An detailed results of robustness test using [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: An experimental result of out of distribution test. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Comparative results of learning efficiency. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 18 canonical work pages

  1. [1]

    M. H. Raibert, Legged robots that balance . MIT press, 1986

    1986

  2. [2]

    Capture point: A step toward humanoid push recovery,

    J. Pratt, J. Carff, S. Drakunov, and A. Goswami, “Capture point: A step toward humanoid push recovery,” in 2006 6th IEEE-RAS international conference on humanoid robots , pp. 200–207, Ieee, 2006

    2006

  3. [3]

    Dynamic locomotion in the mit cheetah 3 through convex model-predictive control,

    J. Di Carlo, P. M. Wensing, B. Katz, G. Bledt, and S. Kim, “Dynamic locomotion in the mit cheetah 3 through convex model-predictive control,” in 2018 IEEE/RSJ international conference on intelligent robots and systems (IROS) , pp. 1–9, IEEE, 2018

    work page 2018

  4. [4]

    Optimization-based control for dynamic legged robots,

    P. M. Wensing, M. Posa, Y . Hu, A. Escande, N. Mansard, and A. Del Prete, “Optimization-based control for dynamic legged robots,” IEEE Transactions on Robotics , vol. 40, pp. 43–63, 2023

    2023

  5. [5]

    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

    arXiv 1909

  6. [6]

    Real-time constrained nonlinear model predictive control on so (3) for dynamic legged locomotion,

    S. Hong, J.-H. Kim, and H.-W. Park, “Real-time constrained nonlinear model predictive control on so (3) for dynamic legged locomotion,” in 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 3982–3989, IEEE, 2020

    work page 2020

  7. [7]

    Concurrent training of a control policy and a state estimator for dynamic and robust legged locomotion,

    G. Ji, J. Mun, H. Kim, and J. Hwangbo, “Concurrent training of a control policy and a state estimator for dynamic and robust legged locomotion,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 4630–4637, 2022

    2022

  8. [8]

    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

Show all 36 references
  1. [9]

    Robust and versatile bipedal jumping control through multi-task rein- forcement learning,

    Z. Li, X. B. Peng, P. Abbeel, S. Levine, G. Berseth, and K. Sreenath, “Robust and versatile bipedal jumping control through multi-task rein- forcement learning,” arXiv preprint arXiv:2302.09450 , vol. 1, 2023

    2023 arXiv

  2. [10]

    Extreme parkour with legged robots,

    X. Cheng, K. Shi, A. Agarwal, and D. Pathak, “Extreme parkour with legged robots,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , pp. 11443–11450, IEEE, 2024

    2024

  3. [11]

    Preparing for the unknown: Learning a universal policy with online system identification,

    W. Yu, J. Tan, C. K. Liu, and G. Turk, “Preparing for the unknown: Learning a universal policy with online system identification,” 2017

    2017

  4. [12]

    Rma: Rapid motor adaptation for legged robots,

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

    2021

  5. [13]

    Learning fast adapta- tion with meta strategy optimization,

    W. Yu, J. Tan, Y . Bai, E. Coumans, and S. Ha, “Learning fast adapta- tion with meta strategy optimization,” IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 2950–2957, 2020

    2020

  6. [14]

    A walk in the park: Learning to walk in 20 minutes with model-free reinforcement learning,

    L. Smith, I. Kostrikov, and S. Levine, “A walk in the park: Learning to walk in 20 minutes with model-free reinforcement learning,” 2022

    2022

  7. [15]

    Adaptive clf-mpc with application to quadrupedal robots,

    M. V . Minniti, R. Grandia, F. Farshidian, and M. Hutter, “Adaptive clf-mpc with application to quadrupedal robots,” IEEE Robotics and Automation Letters, vol. 7, no. 1, pp. 565–572, 2021

    2021

  8. [16]

    Online learning of unknown dynamics for model-based controllers in legged locomotion,

    Y . Sun, W. L. Ubellacker, W.-L. Ma, X. Zhang, C. Wang, N. V . Csomay-Shanklin, M. Tomizuka, K. Sreenath, and A. D. Ames, “Online learning of unknown dynamics for model-based controllers in legged locomotion,” IEEE Robotics and Automation Letters , vol. 6, no. 4, pp. 8442–8449, 2021

    2021

  9. [17]

    Bayesian multi-task learning mpc for robotic mobile manipulation,

    E. Arcari, M. V . Minniti, A. Scampicchio, A. Carron, F. Farshidian, M. Hutter, and M. N. Zeilinger, “Bayesian multi-task learning mpc for robotic mobile manipulation,” IEEE Robotics and Automation Letters , vol. 8, no. 6, pp. 3222–3229, 2023

    2023

  10. [18]

    Learning-based model predictive control for autonomous racing,

    J. Kabzan, L. Hewing, A. Liniger, and M. N. Zeilinger, “Learning-based model predictive control for autonomous racing,” IEEE Robotics and Automation Letters, vol. 4, no. 4, pp. 3363–3370, 2019

    2019

  11. [19]

    Neurobem: Hybrid aerodynamic quadrotor model,

    L. Bauersfeld*, E. Kaufmann*, P. Foehn, S. Sun, and D. Scaramuzza, “Neurobem: Hybrid aerodynamic quadrotor model,” in Robotics: Sci- ence and Systems XVII , RSS2021, Robotics: Science and Systems Foundation, July 2021

    2021

  12. [20]

    Knode-mpc: A knowledge- based data-driven predictive control framework for aerial robots,

    K. Y . Chee, T. Z. Jiahao, and M. A. Hsieh, “Knode-mpc: A knowledge- based data-driven predictive control framework for aerial robots,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 2819–2826, 2022

    2022

  13. [21]

    Real-time neural mpc: Deep learning model predictive control for quadrotors and agile robotic platforms,

    T. Salzmann, E. Kaufmann, J. Arrizabalaga, M. Pavone, D. Scaramuzza, and M. Ryll, “Real-time neural mpc: Deep learning model predictive control for quadrotors and agile robotic platforms,” IEEE Robotics and Automation Letters, vol. 8, no. 4, pp. 2397–2404, 2023

    2023

  14. [22]

    Glide: Generalizable quadrupedal locomotion in diverse environments with a centroidal model,

    Z. Xie, X. Da, B. Babich, A. Garg, and M. v. de Panne, “Glide: Generalizable quadrupedal locomotion in diverse environments with a centroidal model,” in Algorithmic F oundations of Robotics XV (S. M. LaValle, J. M. O’Kane, M. Otte, D. Sadigh, and P. Tokekar, eds.), (Cham), pp....

    2023

  15. [23]

    Fast and efficient locomotion via learned gait transitions,

    Y . Yang, T. Zhang, E. Coumans, J. Tan, and B. Boots, “Fast and efficient locomotion via learned gait transitions,” in Proceedings of the 5th Conference on Robot Learning (A. Faust, D. Hsu, and G. Neumann, eds.), vol. 164 of Proceedings of Machine Learning Research , pp. 773– ...

    2022

  16. [24]

    Policy search for model predictive control with application to agile drone flight,

    Y . Song and D. Scaramuzza, “Policy search for model predictive control with application to agile drone flight,” IEEE Transactions on Robotics , vol. 38, no. 4, pp. 2114–2130, 2022

    2022

  17. [25]

    Continuous versatile jumping using learned action residuals,

    Y . Yang, X. Meng, W. Yu, T. Zhang, J. Tan, and B. Boots, “Continuous versatile jumping using learned action residuals,” in Proceedings of The 5th Annual Learning for Dynamics and Control Conference (N. Matni, M. Morari, and G. J. Pappas, eds.), vol. 211 of Proceedings of Mach...

    2023

  18. [26]

    Rloc: Terrain-aware legged locomotion using reinforcement learning and optimal control,

    S. Gangapurwala, M. Geisert, R. Orsolino, M. Fallon, and I. Havoutis, “Rloc: Terrain-aware legged locomotion using reinforcement learning and optimal control,” IEEE Transactions on Robotics , vol. 38, no. 5, pp. 2908–2927, 2022

    2022

  19. [27]

    Tossing- bot: Learning to throw arbitrary objects with residual physics,

    A. Zeng, S. Song, J. Lee, A. Rodriguez, and T. Funkhouser, “Tossing- bot: Learning to throw arbitrary objects with residual physics,” IEEE Transactions on Robotics , vol. 36, no. 4, pp. 1307–1319, 2020

    2020

  20. [28]

    Robust quadruped jumping via deep reinforcement learning,

    G. Bellegarda, C. Nguyen, and Q. Nguyen, “Robust quadruped jumping via deep reinforcement learning,” Robotics and Autonomous Systems , vol. 182, p. 104799, 2024

    2024

  21. [29]

    Learning agile locomotion and adaptive be- haviors via rl-augmented mpc,

    Y . Chen and Q. Nguyen, “Learning agile locomotion and adaptive be- haviors via rl-augmented mpc,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , pp. 11436–11442, 2024

    2024

  22. [30]

    Rl + model-based control: Using on-demand optimal control to learn versatile legged locomotion,

    D. Kang, J. Cheng, M. Zamora, F. Zargarbashi, and S. Coros, “Rl + model-based control: Using on-demand optimal control to learn versatile legged locomotion,” IEEE Robotics and Automation Letters , vol. 8, no. 10, pp. 6619–6626, 2023

    2023

  23. [31]

    Imi- tating and finetuning model predictive control for robust and symmetric quadrupedal locomotion,

    D. Youm, H. Jung, H. Kim, J. Hwangbo, H.-W. Park, and S. Ha, “Imi- tating and finetuning model predictive control for robust and symmetric quadrupedal locomotion,” IEEE Robotics and Automation Letters , vol. 8, no. 11, pp. 7799–7806, 2023

    2023

  24. [32]

    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

  25. [33]

    Opt-mimic: Imitation of optimized trajectories for dynamic quadruped behaviors,

    Y . Fuchioka, Z. Xie, and M. Van de Panne, “Opt-mimic: Imitation of optimized trajectories for dynamic quadruped behaviors,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) , pp. 5092–5098, IEEE, 2023

    2023

  26. [34]

    Hybrid zero dynamics of planar biped walkers,

    E. Westervelt, J. Grizzle, and D. Koditschek, “Hybrid zero dynamics of planar biped walkers,” IEEE Transactions on Automatic Control, vol. 48, no. 1, pp. 42–56, 2003

    2003

  27. [35]

    Proximal policy optimization algorithms,

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

    2017

  28. [36]

    Per-contact iteration method for solving contact dynamics,

    J. Hwangbo, J. Lee, and M. Hutter, “Per-contact iteration method for solving contact dynamics,” IEEE Robotics and Automation Letters , vol. 3, no. 2, pp. 895–902, 2018

    2018

Pith tools

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