Real-time Whole-body Model Predictive Control for Bipedal Locomotion with a Novel Kino-dynamic Model and Warm-start Method

Hokyun Lee; Jaeheung Park; Junhyung Kim

REVIEW 3 major objections 6 minor 18 references

Real-time Whole-body Model Predictive Control for Bipedal Locomotion with a Novel Kino-dynamic Model and Warm-start Method

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A kino-dynamic model that merges the LIPFM with full-body kinematics and a modularized MLP warm-start lets whole-body MPC for bipedal walking run within a 20 ms control cycle, solving 60-step problems with 4620 variables in under 17 ms.

desk verdict Real whole-body MPC at 50 Hz on a humanoid is a genuine engineering win, but the model's COM-height consistency and some baseline choices need scrutiny before this is treated as settled. read the letter →

arxiv 2505.19540 v1 pith:H5TA7SU7 submitted 2025-05-26 cs.RO

Junhyung Kim , Hokyun Lee , Jaeheung Park This is my paper

classification cs.RO

keywords whole-bodymodelpredictivecontrolkino-dynamiclinearinvertedpendulumplusflywheelZMP-basedmodularizedMLPwarm-startbipedallocomotionreal-timehumanoidrobot

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 claims that whole-body model predictive control (WB-MPC) for bipedal walking can be executed inside a 20 ms control cycle when the planner uses a kino-dynamic model built from the linear inverted pendulum plus flywheel (LIPFM) and full-body kinematics, with the zero-moment point (ZMP) replacing contact wrenches as the contact representation. The accompanying warm-start strategy uses a modularized multi-layer perceptron to predict a good initial guess for each control cycle, reducing warm-start latency from hundreds of milliseconds to under a millisecond. The authors report solving a 60-time-step, 4620-variable, 2400-constraint WB-MPC in at most about 17 ms, including under external pushes, and validate the framework in simulation and on the TOCABI humanoid. The paper's own conclusion restricts the claim to flat-ground walking; stairs, steep slopes, and flight phases are outside the model.

What carries the argument

The load-bearing object is the kino-dynamic model that merges the LIPFM with full-body kinematics. Equations (1)-(2) give COM acceleration as a linear function of ZMP and the derivative of centroidal angular momentum at constant COM height $z_c$; equation (3) connects joint accelerations to that momentum derivative through the centroidal momentum matrix; and equations (4)-(5) enforce COM and foot placements through forward kinematics. Because the contact quantity is the ZMP rather than distributed contact wrenches, the optimization constraints do not change form when the foot support phase changes, and no inverse dynamics is needed to build the model. The other central piece is the warm-start: a set of small single-hidden-layer MLPs, one for each walking phase and each variable group, maps a low-dimensional input state to RBF/PCA coefficients of a near-optimal state trajectory, which the DDP solver then refines.

What would settle it

Run the framework with the same solver settings on a walking task with a vertical step large enough to force the center of mass to rise or fall by several centimeters, and check whether the robot stays balanced and the solve stays under 20 ms; a failure would show that the constant-COM-height kino-dynamic model, not the warm-start, is the limiting assumption.

Watch

Extended reading notes

Core claim

The central claim is that a whole-body MPC for bipedal locomotion can use ZMP-based dynamics instead of contact-wrench whole-body dynamics and still capture the stabilizing effects of whole-body motion. The proposed kino-dynamic model couples the LIPFM equations of the center of mass to the centroidal momentum matrix and forward kinematics, so joint motion, feet placement, and angular momentum enter the optimization without an inverse-dynamics pass. Replacing contact wrenches with ZMP keeps the problem structure unchanged across support phases, which removes the peak latency observed at contact transitions in wrench-based whole-body MPC and lowers the average per-iteration cost by roughly 44 percent in the paper's comparisons. With the modularized MLP warm-start, the solver converges in about two DDP iterations per cycle, and the combined system keeps the robot balanced against pushes in real time; on hardware, it withstands a 30.4 Ns frontal impact and a 15.3 Ns lateral impact.

Load-bearing premise

The model assumes the center of mass stays at a fixed height, so the real-time planning claim only covers flat-ground walking without flight phases; stairs, steep slopes, running, and jumping are outside what the MPC can represent.

Editorial extensions

If this is right

Humanoid controllers can embed whole-body dynamic effects into the MPC layer itself, since the planned trajectories for pelvis, waist, shoulders, and feet are computed reactively rather than by a separate simplified footstep planner.
The reported worst-case latency of about 17 ms means the same 20 ms control loop can close on a torque-controlled humanoid without splitting planning and control into separate rates.
Warm-start latency no longer scales with the size of the motion database, so the training set can include perturbation scenarios and varied initial states without breaking the real-time budget.
ZMP-based whole-body control can be merged with the MPC objective so that impulses at foot landing and ZMP regulation are handled in the same feedback law.

Reading between the lines

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

The constant-COM-height assumption suggests the framework is a flat-ground, stance-phase planner; extending it to stairs, slopes, running, or jumping would require replacing or augmenting the LIPFM equations, and the reported latency would likely change.
Because the MLP is trained per walking phase and per variable type, its coverage of unseen gaits is bounded by the training distribution; the paper's successful unseen-state walk is suggestive, but a systematic sweep over step length, speed, and terrain height would map the true generalization region.
The phase-invariant structure of the model implies that the latency advantage over wrench-based whole-body MPC should persist on higher-degree-of-freedom humanoids, though the absolute 17 ms number is tied to the 19-actuator configuration tested.

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

Here's the short version: this is a real engineering achievement—whole-body MPC on a torque-controlled humanoid at 50 Hz with peak latency under 17 ms, including during pushes—but the physical consistency of the model is not fully nailed down, and the comparison baselines are somewhat cherry-picked.

What's genuinely new: the LIPFM-plus-FKM formulation that uses ZMP instead of contact wrenches, avoiding the latency spikes during contact transitions; and the modularized MLP warm-start with PCA/RBF compression, which cuts prediction latency from ~500 ms to ~0.3 ms while keeping accuracy close to GPR. The DDP solver stats are well measured: average 2-3 iterations, low variance, and real robot data supporting the latency claims. That part holds up.

The soft spots are real. First, the constant-COM-height LIPFM (Eqs. 1-2) and the full-body COM constraint (Eq. 5) are never reconciled. The paper doesn't state whether the MPC enforces X_com,z(q)=z_c. If it does, the upper-body recovery motions shown in Fig. 7 are artificially constrained; if it doesn't, the LIPFM is being used outside its validity. Either way the optimized plan may not describe the executed centroidal motion, and the robustness could be coming from the WBC and PD loop. This is not just a stairs problem; it can matter on flat ground during large push recovery. The authors' own conclusion concedes vertical COM limitations but doesn't address the flat-ground case.

Second, the comparison handicaps CP-MPC by removing its footstep adjustment, and the authors admit that an ideal-real-time WBD baseline handles larger perturbations. So the contribution is computational speed, not improved robustness over an unconstrained WBD.

Third, no code, data, or weight values are released, which hurts reproducibility.

On the other hand, the circularity of training the warm-start on the same solver is partially mitigated by benchmarking against cold start and other learned predictors. The latency evidence is concrete. For a robotics audience, this is a solid, significant step: it makes real-time whole-body MPC practical, which is a genuine bottleneck. The model inconsistency should be addressed—at minimum, report the vertical COM error during the experiments and quantify how much the MPC constraint is violated.

Who should read it: anyone working on humanoid MPC or warm-starting DDP. It deserves a serious referee even with these caveats; the core claim is important and mostly supported. My recommendation: send it to peer review, but require the authors to clarify the COM-height constraint and release hyperparameters, and rerun the CP-MPC comparison without handicapping the stepping.

Referee Report

3 major / 6 minor

Summary. This paper proposes a real-time whole-body model predictive control framework for bipedal locomotion. The authors introduce a 'kino-dynamic' model that couples the linear inverted pendulum plus flywheel model with full-body forward kinematics, replacing contact-wrench formulations with ZMP-based constraints. They also propose a modularized multi-layer perceptron warm-start strategy that predicts an initial guess for the DDP solver at each control cycle, and a ZMP-based whole-body controller. The reported results include latency comparisons against WBD and centroidal-dynamics baselines, warm-start accuracy comparisons, and simulated and real-robot push-recovery experiments on the TOCABI humanoid, with a central quantitative claim of solving a 60-step, 4620-variable, 2400-constraint WB-MPC within 17 ms.

Significance. If the central claims hold, the paper is a useful practical step toward real-time whole-body MPC on torque-controlled humanoids: the proposed model keeps per-iteration latency low and, importantly, low-variance across contact transitions, and the real-robot validation under perturbations is valuable. The explicit reporting of worst-case latencies and the candid statement of the model's vertical-COM limitations are also strengths. The significance is tempered by an unresolved consistency question between the LIPFM's constant-COM-height assumption and the full-body kinematic constraints, and by a robustness comparison that removes the footstep-adjustment mechanism from the CP-MPC baseline. These issues affect the scope of the central claim more than the measured latency numbers.

major comments (3)

[Section II-A/II-B, Eqs. (1)-(2), (5), and (9)] The manuscript does not reconcile the constant-COM-height assumption of the LIPFM with the full-body COM map \bar X_com(q) that appears in the h1 constraints. Because the MPC state contains both the generalized coordinates q and the LIPFM COM variables (x, y, \dot x, \dot y), the formulation must state whether h1 enforces X_com,z(q) = z_c, and whether it ties the LIPFM COM position/velocity to the COM position/velocity implied by (q, \dot q). Neither is specified. If the vertical COM coordinate is left free, the whole-body recovery motions shown in Fig. 7 (pelvis pitch, waist pitch, shoulder motion) can change z_c while Eqs. (1)-(2) treat z_c as fixed, so the planned ZMP/COM trajectory may not describe the executed centroidal motion. If X_com,z(q) = z_c is imposed instead, the dynamic effects of whole-body motion are artificially restricted. In either case the 17 ms solver latency does not by itself establish that the optimized plan is physically valid in the demonstrated scenarios. The conclusion's caveat about stairs and slopes is the same vertical-COM issue restricted to those terrains; it can also arise on flat ground under perturbation. Please state the exact form of h1 and report the vertical COM trajectory from the Fig. 7 experiment.
[Section V-A and Fig. 6(a)] The robustness comparison against CP-MPC removes the footstep-adjustment strategy of CP-MPC, despite that strategy being CP-MPC's primary mechanism for responding to perturbations. As a result, 'our proposed WB-MPC outperforms CP-MPC' is not a comparison against CP-MPC as designed. Please either include the footstep strategy in the baseline and compare under the same footstep constraints, or rephrase the conclusion to state that the proposed controller outperforms a CP-MPC variant without stepping. The acknowledgment that an ideal-real-time WBD baseline withstands larger perturbations is helpful, but it implies the robustness advantage over WBD rests on that baseline's non-real-time latency; the summary comparison should say so explicitly.
[Section III and Fig. 5] The modularized MLP is trained on trajectories generated by the same WB-MPC solver that it warm-starts, so the reported reductions in DDP iterations and residual cost are partly a measure of how well the regressor reproduces solver outputs within its training distribution. The paper mentions stable walking for 'unseen states' and different step lengths, but it does not quantify generalization to perturbation magnitudes or directions that were absent from the training set. Please state the perturbation sampling distribution used to create the motion datasets and report held-out results (for example, residual cost and average iterations for push magnitudes outside the sampled range) so that the warm-start claim is tested beyond interpolation of the solver's own outputs.

minor comments (6)

[Fig. 6] The caption contains a typo: 'referes' should be 'refers'. Please also report the number of trials underlying the mean and maximum latencies in Fig. 6(b).
[Eq. (8)] The dynamics g in Eq. (8) is never written out. In particular, the evolution of the CAM components h_x and h_y is not stated; please provide the explicit continuous-time or discretized dynamics, including \dot h_x = \tau_x and \dot h_y = \tau_y (or equivalent), so the LIPFM state evolution is well-defined.
[Section II-B cost weights] The numerical values of the cost weights W_p, W_pl, W_cp, W_e, W_u and the WBC weights W_c, W_q, W_F are not reported, and the RBF/PCA hyperparameters are described only as 'fine-tuned via line search'. Without these values, the comparative experiments are difficult to reproduce.
[Eqs. (4)-(5)] Notation is inconsistent: Eq. (4) writes the end-effector map as \bar X_e while the text refers to \bar X_c; please unify the notation and define T_{e,d} more precisely as a desired SE(3) placement.
[Table III] The R-squared values 0.99975 and 0.99953 are cited for PCA and RBF, but it is not stated whether they measure reconstruction accuracy in the reduced space or the original trajectory space; please clarify.
[Section V-B] The 12 real-robot trials are summarized only by maxima and averages; reporting per-condition mean and standard deviation of the achieved DDP latency and iteration count would strengthen the claim.

Circularity Check

1 steps flagged · score 2.0 of 10

Only minor self-referentiality in the learned warm-start; the core real-time WB-MPC claim is externally benchmarked and does not reduce to the fit.

fitted input called prediction [Section III-A, Problem Definition and Training Strategies]
"So we collect WB-MPC motion datasets by varying the initial joint positions, velocities, and the positions and velocities of the COM and ZMP. The convergence threshold for DDP is set a low value at 10−8 to generate near-optimal motion trajectory datasets."

The modularized MLP is trained to output RBF/PCA-compressed versions of the WB-MPC trajectory, and those training trajectories are generated by the same WB-MPC DDP solver that the network is meant to warm-start. The 'initial guess prediction' is therefore a regression of the solver's own output, and the accuracy reported in Tables II and III measures agreement with that same solver rather than with an independent ground truth. The reduction in DDP iterations with warm-start is partly a measure of how well the network has memorized the solver's solution manifold.

full rationale

The paper's central claims are (1) a kino-dynamic LIPFM+FKM model that avoids contact-transition discontinuities and lowers per-iteration latency, and (2) a modularized MLP warm-start that reduces DDP iterations. Claim (1) is derived from model equations (1)-(5) and is compared against WBD and centroidal-dynamics baselines; it does not reduce to any fitted quantity. Claim (2) is a learned warm-start, so its training on the same solver's trajectories is a mild self-referentiality: the network is essentially recalling the solver's solution manifold rather than providing an independent prediction. However, the paper does not present the MLP output as physical evidence; it is an initialization for the DDP solver, and the real-time feasibility and robustness are validated by wall-clock latency measurements and by hardware experiments against external baselines (cold start, previous optimal, WBD, CP-MPC). The acknowledged limitation (no vertical COM motion, stairs, slopes, or flight phase) is a scope restriction, not a circular step. No self-citation chain or imported uniqueness theorem is load-bearing in the derivation. Overall, the partial self-referentiality of the warm-start does not undermine the independent content of the kino-dynamic model and the benchmarked real-time performance.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the constant-COM-height ZMP assumption, the soft penalty handling of full-body kinematics, and an implicit generalization assumption for the MLP warm-start. No invented physical entities are introduced.

free parameters (4)

MPC running cost weights (W_p, W_pl, W_cp, W_e, W_u) = not reported
Enter the five cost terms in Section II-B; hand-tuned to balance ZMP tracking, boundary enforcement, capture-point tracking, feet tracking, and upper-body regularization. No values or tuning procedure are given.
WBC weights and joint gains (W_c, W_q, W_F, kp, kv) = not reported
Used in the QP (10) and final torque command in Section II-C; these shape impulse control and tracking. Values are not disclosed.
RBF means and standard deviations for output compression = fine-tuned via line search; values not reported
Section III-A/B: 55 RBFs are used to compress WB-MPC trajectories before PCA. The paper states the means and standard deviations are tuned by line search, so they are fitted to the training data.
Dimensionality and model size hyperparameters = 55 RBFs, 32 PCA components, 80 hidden LReLUs, max 5 DDP iterations
Chosen by the authors in Section IV-A without sensitivity analysis; these affect prediction accuracy and solve latency.

assumptions (6)

domain assumption The COM height zc is constant in the LIPFM equations (1) and (2).
Central to the kino-dynamic model; the conclusion acknowledges this limits use on stairs, slopes, and flight phases.
domain assumption ZMP alone is a sufficient contact representation for the MPC problem, so contact wrenches and contact-force feasibility can be handled downstream by the WBC.
Used to keep the MPC problem structure invariant during contact transitions; the paper does not analyze friction or contact-force limits inside the MPC.
domain assumption The FKM constraints (3)-(5) can be treated as soft penalty terms without materially changing the solution.
Section II-B: FKM constraints are handled efficiently using a penalty method, as employed in [15]; exact feasibility is not proven.
domain assumption The relation H(q)qdd+Hdot(q)qdot=tau from the centroidal momentum matrix can be combined with the LIPFM dynamics without consistency issues.
Equations (1)-(3); the paper does not prove that the tau produced by joint accelerations matches the tau used as LIPFM control when penalty constraints are active.
domain assumption Offline WB-MPC trajectories generated with low convergence threshold represent near-optimal motions that the MLP can imitate.
Section III-A: datasets are generated by solving WB-MPC with varied initial states; the paper assumes these are good targets for supervised warm-start learning.
domain assumption The modularized MLP generalizes to unseen states and contact perturbations.
Section V-B reports walking with unseen states and step lengths, but no distribution analysis or formal generalization guarantee is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Real-time Whole-body Model Predictive Control for Bipedal Locomotion with a Novel Kino-dynamic Model and Warm-start Method." pith.science (2026). https://pith.science/paper/H5TA7SU7

@misc{pith2026250519540,
  author       = {Pith},
  title        = {Pith review of: Real-time Whole-body Model Predictive Control for Bipedal Locomotion with a Novel Kino-dynamic Model and Warm-start Method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H5TA7SU7}},
  note         = {Machine review of arXiv:2505.19540}
}

read the original abstract

Advancements in optimization solvers and computing power have led to growing interest in applying whole-body model predictive control (WB-MPC) to bipedal robots. However, the high degrees of freedom and inherent model complexity of bipedal robots pose significant challenges in achieving fast and stable control cycles for real-time performance. This paper introduces a novel kino-dynamic model and warm-start strategy for real-time WB-MPC in bipedal robots. Our proposed kino-dynamic model combines the linear inverted pendulum plus flywheel and full-body kinematics model. Unlike the conventional whole-body model that rely on the concept of contact wrenches, our model utilizes the zero-moment point (ZMP), reducing baseline computational costs and ensuring consistently low latency during contact state transitions. Additionally, a modularized multi-layer perceptron (MLP) based warm-start strategy is proposed, leveraging a lightweight neural network to provide a good initial guess for each control cycle. Furthermore, we present a ZMP-based whole-body controller (WBC) that extends the existing WBC for explicitly controlling impulses and ZMP, integrating it into the real-time WB-MPC framework. Through various comparative experiments, the proposed kino-dynamic model and warm-start strategy have been shown to outperform previous studies. Simulations and real robot experiments further validate that the proposed framework demonstrates robustness to perturbation and satisfies real-time control requirements during walking.

Figures

Figures reproduced from arXiv: 2505.19540 by the authors.

**Figure 2.** The proposed control framework using novel kino-dynamic model. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Modularized MLP for finding good initial guess [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Latency graph for 1 DDP iteration during walking [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 5.** Figure 5: The residual cost of WB-MPC under perturbation [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: (a) The maximum impulse force a robot can withstand [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

18 extracted references · 16 canonical work pages

[1]

Biped walking pattern generation by using preview control of zero-moment point,

S. Kajita et al. , “Biped walking pattern generation by using preview control of zero-moment point,” IEEE International Conference on Robotics and Automation , pp. 1620-1626, 2003

work page 2003
[2]

Versatile reactive bipedal locomotion planning through hierarchical optimization,

J. Ding et al., “Versatile reactive bipedal locomotion planning through hierarchical optimization,” IEEE International Conference on Robotics and Automation, pp. 256–262, 2019

work page 2019
[3]

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

J. Di Carlo et al., “Dynamic locomotion in the mit cheetah 3 through convex model-predictive control,” IEEE/RSJ International Conference on Intelligent Robots and Systems , pp. 256–262, 2019

work page 2019
[4]

Optimization based locomotion planning, esti- mation, and control design for the atlas humanoid robot,

S. Kuindersma et al., “Optimization based locomotion planning, esti- mation, and control design for the atlas humanoid robot,” Autonomous Robots, vol. 40, pp. 429–455, 2016

work page 2016
[5]

Whole-body motion planning with centroidal dy- namics and full kinematics,

H. Dai et al. , “Whole-body motion planning with centroidal dy- namics and full kinematics,” IEEE-RAS International Conference on Humanoid Robots, pp. 651–656, 2015

work page 2015
[6]

A Unified MPC Framework for Whole-Body Dy- namic Locomotion and Manipulation,

J. -P. Sleiman et al., “A Unified MPC Framework for Whole-Body Dy- namic Locomotion and Manipulation,” IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4688-4695, 2021

work page 2021
[7]

Analysis of Effects of Leg Inertia Distribution on Contact Stability and Dynamic Performance During Bipedal Walking,

J. Kim et al. , “Analysis of Effects of Leg Inertia Distribution on Contact Stability and Dynamic Performance During Bipedal Walking,” IEEE-RAS International Conference on Humanoid Robots , 2022

work page 2022
[8]

From centroidal to whole-body models for legged locomotion: a comparative analysis,

E. Dantec et al. , “From centroidal to whole-body models for legged locomotion: a comparative analysis,” IEEE-RAS International Confer- ence on Humanoid Robots , 2024

work page 2024

Show all 18 references

[9]

Learning How to Walk: Warm-Starting Optimal Control Solver with memory of motion,

T. S. Lembono et al., “Learning How to Walk: Warm-Starting Optimal Control Solver with memory of motion,” IEEE International Confer- ence on Robotics and Automation , pp. 1357-1363, 2020

2020
[10]

Whole Body Model Predictive Control with a memory of motion: Experiments on a Torque-Controlled Talos,

E. Dantec et al. , “Whole Body Model Predictive Control with a memory of motion: Experiments on a Torque-Controlled Talos,” IEEE International Conference on Robotics and Automation, pp. 8202-8208, 2021

2021
[11]

Whole-Body Model Predictive Control for Biped Locomotion on a Torque-Controlled Humanoid Robot,

E. Dantec et al. . “Whole-Body Model Predictive Control for Biped Locomotion on a Torque-Controlled Humanoid Robot,” IEEE-RAS International Conference on Humanoid Robots , pp. 638-644, 2022

2022
[12]

Tailoring Solution Accuracy for Fast Whole- Body Model Predictive Control of Legged Robots,

C. Khazoom et al. , “Tailoring Solution Accuracy for Fast Whole- Body Model Predictive Control of Legged Robots,” IEEE Robotics and Automation Letters , vol. 9, no. 12, pp. 11074-11081, 2024

2024
[13]

Dynamic Locomotion For Passive-Ankle Biped Robots And Humanoids Using Whole-Body Locomotion Control,

D. Kim et al., “Dynamic Locomotion For Passive-Ankle Biped Robots And Humanoids Using Whole-Body Locomotion Control,” The Inter- national Journal of Robotics Research , vol. 39, pp. 936-956, 2020

2020
[14]

Highly Dynamic Quadruped Locomotion via Whole-Body Impulse Control and Model Predictive Control,

D. Kim et al. , “Highly Dynamic Quadruped Locomotion via Whole-Body Impulse Control and Model Predictive Control,” arXiv:1909.06586v1, 2019

1909 arXiv
[15]

Whole-Body MPC and Online Gait Sequence Generation for Wheeled-Legged Robots,

M. Bjelonic et al. , “Whole-Body MPC and Online Gait Sequence Generation for Wheeled-Legged Robots,” IEEE/RSJ International Conference on Intelligent Robots and Systems , pp. 8388-8395, 2021

2021
[16]

Crocoddyl: An efficient and versatile framework for multi-contact optimal control,

C. Mastalli et al. , “Crocoddyl: An efficient and versatile framework for multi-contact optimal control,” IEEE International Conference on Robotics and Automation , pp.2536-2542, 2020

2020
[17]

Design of the humanoid robot TOCABI,

M. Schwartz et al., “Design of the humanoid robot TOCABI,” IEEE- RAS International Conference on Humanoid Robots , pp. 323-329, 2022

2022
[18]

A Model Predictive Capture Point Control Frame- work for Robust Humanoid Balancing via Ankle, Hip, and Stepping Strategies,

M. Kim et al. , “A Model Predictive Capture Point Control Frame- work for Robust Humanoid Balancing via Ankle, Hip, and Stepping Strategies,” arXiv:2307.13243, 2024

2024 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Biped walking pattern generation by using preview control of zero-moment point,

S. Kajita et al. , “Biped walking pattern generation by using preview control of zero-moment point,” IEEE International Conference on Robotics and Automation , pp. 1620-1626, 2003

work page 2003

[2] [2]

Versatile reactive bipedal locomotion planning through hierarchical optimization,

J. Ding et al., “Versatile reactive bipedal locomotion planning through hierarchical optimization,” IEEE International Conference on Robotics and Automation, pp. 256–262, 2019

work page 2019

[3] [3]

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

J. Di Carlo et al., “Dynamic locomotion in the mit cheetah 3 through convex model-predictive control,” IEEE/RSJ International Conference on Intelligent Robots and Systems , pp. 256–262, 2019

work page 2019

[4] [4]

Optimization based locomotion planning, esti- mation, and control design for the atlas humanoid robot,

S. Kuindersma et al., “Optimization based locomotion planning, esti- mation, and control design for the atlas humanoid robot,” Autonomous Robots, vol. 40, pp. 429–455, 2016

work page 2016

[5] [5]

Whole-body motion planning with centroidal dy- namics and full kinematics,

H. Dai et al. , “Whole-body motion planning with centroidal dy- namics and full kinematics,” IEEE-RAS International Conference on Humanoid Robots, pp. 651–656, 2015

work page 2015

[6] [6]

A Unified MPC Framework for Whole-Body Dy- namic Locomotion and Manipulation,

J. -P. Sleiman et al., “A Unified MPC Framework for Whole-Body Dy- namic Locomotion and Manipulation,” IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4688-4695, 2021

work page 2021

[7] [7]

Analysis of Effects of Leg Inertia Distribution on Contact Stability and Dynamic Performance During Bipedal Walking,

J. Kim et al. , “Analysis of Effects of Leg Inertia Distribution on Contact Stability and Dynamic Performance During Bipedal Walking,” IEEE-RAS International Conference on Humanoid Robots , 2022

work page 2022

[8] [8]

From centroidal to whole-body models for legged locomotion: a comparative analysis,

E. Dantec et al. , “From centroidal to whole-body models for legged locomotion: a comparative analysis,” IEEE-RAS International Confer- ence on Humanoid Robots , 2024

work page 2024