pith. sign in

arxiv: 2604.22251 · v3 · submitted 2026-04-24 · 💻 cs.RO

False Feasibility in Variable Impedance MPC for Legged Locomotion

Vishal Ramesh This is my paper

Pith reviewed 2026-05-12 03:28 UTC · model grok-4.3

classification 💻 cs.RO
keywords variable impedance controlmodel predictive controllegged locomotionactuator dynamicsfeasibilityhopping robotstiffness command
0
0 comments X

The pith

Variable impedance MPC for legged robots overestimates what stiffness actuators can actually deliver.

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

The paper demonstrates that treating joint stiffness as an instantaneous choice in model predictive control creates a feasible set larger than what first-order actuator dynamics allow. It formalizes this gap through the ratio alpha of actuator bandwidth to task timescale and derives a closed-form critical value alpha_crit for a one-dimensional hopping monoped below which no stiffness command can match the prediction. Numerical checks confirm that deviation grows predictably as alpha drops, and the same mismatch appears in planar spring-loaded inverted pendulum models as center-of-mass and timing errors. The work shows that simply tightening stiffness limits cannot fix the problem below a second threshold, while adding stiffness to the prediction state removes the mismatch by construction.

Core claim

In variable impedance MPC the parameter-based feasible set F_param strictly contains the realizable set F_real defined by first-order actuator dynamics. For the 1D hopping monoped, below the analytical threshold alpha_crit derived from task physics, no admissible stiffness command realizes the parameter-based prediction; numerical tests across ten parameter sets show monotonic deviation growth with log-log R-squared of 0.986, and the mismatch transfers to planar dynamics as center-of-mass and stance-timing deviations.

What carries the argument

The dimensionless ratio alpha equals actuator bandwidth times task timescale; it identifies the regime where the assumed instantaneous stiffness change exceeds what the actuator can produce, with alpha_crit serving as the exact boundary for the monoped case.

If this is right

  • Deviation between predicted and realized behavior increases steadily as alpha decreases, with the observed scaling consistent across varied parameter combinations.
  • In planar spring-loaded inverted pendulum models the mismatch produces primary center-of-mass and stance-timing errors plus secondary friction effects.
  • A second lower threshold alpha_infeas shows that restricting the allowed stiffness range cannot restore realizability once alpha drops far enough.
  • Augmenting the MPC prediction model with the stiffness state removes the feasibility gap by construction.

Where Pith is reading between the lines

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

  • The same false-feasibility pattern is likely to appear in any variable-impedance controller whose task timescale approaches or exceeds actuator bandwidth.
  • Robot designers could use the closed-form alpha_crit to set minimum actuator requirements before deploying variable-impedance MPC on new hardware.
  • Hardware experiments that sweep alpha while logging realized versus commanded stiffness would directly test whether the analytical threshold holds outside simulation.

Load-bearing premise

The realizable stiffness set is defined under strictly first-order actuator dynamics; any higher-order or nonlinear actuator behavior would alter the mismatch threshold.

What would settle it

Measure the actual stiffness trajectory on a 1D hopping monoped under MPC commands and check whether the motion deviates from the prediction exactly when the computed alpha falls below the closed-form alpha_crit.

Figures

Figures reproduced from arXiv: 2604.22251 by Vishal Ramesh.

Figure 1
Figure 1. Figure 1: 1D sweep across α for both controllers. Left. Normalized L∞ trajectory deviation Dα. The parameter-based formulation (red) exhibits monotone growth as α decreases, while the stiffness-as-state formulation (blue) remains zero to numerical precision across the sweep. Shaded regions indicate ensemble spread across touchdown velocities. Right. Normalized liftoff-time deviation ∆Tα, showing the same qualitative… view at source ↗
Figure 2
Figure 2. Figure 2: shows that the points cluster tightly around a line of slope one in log-log coordinates with proportionality constant 0.67, with a coefficient of determination of R2 = 0.986. The consistency of this relation across more than a fivefold range in predicted αcrit indicates that the scaling law captured by (19) is not specific to the nominal operating point [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Planar SLIP sweep showing mechanism transfer from 1D. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Conservative tuning and structural reach. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
read the original abstract

Variable impedance model predictive control (MPC) formulations often treat joint stiffness as an instantaneous decision variable. The resulting feasible set strictly contains the physically realizable set under first-order actuator dynamics. We identify this as a formulation error rather than a modeling approximation, formalize the distinction between the parameter-based feasible set F_param and the realizable set F_real, and characterize the regime of mismatch via the dimensionless parameter {\alpha} = {\omega}sT (actuator bandwidth times task timescale). For the 1D hopping monoped, we prove that below an analytical threshold {\alpha}_crit derived in closed form from task physics, no admissible stiffness command realizes the parameter-based prediction. Numerical validation in 1D shows monotonic deviation growth as {\alpha} decreases, with the predicted scaling holding across ten parameter combinations (log-log R2 = 0.986). Mechanism transfer to planar spring-loaded inverted pendulum dynamics confirms center-of-mass and stance-timing deviation as the primary consequence, with regime-dependent friction effects as a tertiary observable. A second threshold {\alpha}_infeas < {\alpha}_crit establishes a floor below which restricting the admissible stiffness range cannot repair realizability, closing the conservative-tuning objection. Augmenting the prediction state with stiffness closes the mismatch by construction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript identifies a formulation error in variable impedance MPC where joint stiffness is treated as an instantaneous decision variable, causing the parameter-based feasible set F_param to strictly contain the physically realizable set F_real under first-order actuator dynamics. It introduces the dimensionless parameter α = ω_s T and provides a closed-form proof for the 1D hopping monoped that below an analytical threshold α_crit (derived from task physics and actuator bandwidth), no admissible stiffness command realizes the F_param prediction. Numerical validation across ten parameter sets confirms monotonic deviation growth with the predicted scaling (log-log R² = 0.986). The analysis transfers to planar SLIP dynamics, highlighting center-of-mass and stance-timing deviations, derives a second threshold α_infeas below which tuning cannot repair realizability, and proposes augmenting the prediction state with stiffness to close the mismatch by construction.

Significance. If the results hold, this work is significant for legged locomotion control because it exposes a subtle but load-bearing inconsistency in a widely used MPC approach that can produce unrealizable commands when actuator bandwidth is insufficient relative to task timescales. Strengths include the closed-form derivation of α_crit, independent numerical validation with high R² across parameter combinations, transfer of the mechanism to SLIP dynamics, and the constructive state-augmentation fix. The identification of α_infeas as an irreparable floor is practically useful and distinguishes the contribution from mere conservative tuning advice.

major comments (2)
  1. [1D hopping monoped analysis and F_real definition] The impossibility result and derivation of α_crit are constructed exclusively under the assumption of strictly first-order actuator dynamics when defining F_real. While the manuscript correctly scopes the claim to this model, it should add an explicit discussion (perhaps in the 1D analysis or conclusions) of how the threshold and realizability floor would shift under higher-order poles, nonlinear stiffness-to-torque maps, or unmodeled delays, as this is the primary modeling assumption underlying the central proof.
  2. [Mechanism transfer to planar SLIP] The transfer to planar SLIP dynamics is presented as confirming COM and stance-timing deviations as primary consequences, but the manuscript should clarify whether this extension relies on additional approximations beyond the 1D case or if the α_crit threshold carries over directly without re-derivation.
minor comments (2)
  1. [Numerical validation] The abstract reports log-log R² = 0.986 for the scaling check but does not name the precise deviation metric (e.g., position error, timing error, or combined cost) being regressed; state this explicitly in the numerical validation section.
  2. [Throughout] Notation for F_param and F_real is introduced in the abstract and early sections; ensure every subsequent reference to feasible sets uses these symbols consistently rather than descriptive phrases.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and the recommendation for minor revision. The comments identify useful opportunities to strengthen the discussion of modeling assumptions and the scope of the SLIP results. We address each point below.

read point-by-point responses

  1. Referee: [1D hopping monoped analysis and F_real definition] The impossibility result and derivation of α_crit are constructed exclusively under the assumption of strictly first-order actuator dynamics when defining F_real. While the manuscript correctly scopes the claim to this model, it should add an explicit discussion (perhaps in the 1D analysis or conclusions) of how the threshold and realizability floor would shift under higher-order poles, nonlinear stiffness-to-torque maps, or unmodeled delays, as this is the primary modeling assumption underlying the central proof.

    Authors: We agree that an explicit discussion of the first-order assumption would strengthen the manuscript. Although the claims are already scoped to this actuator model, we will add a paragraph in the 1D analysis section (with a forward reference in the conclusions) addressing how α_crit and the α_infeas floor would be affected by higher-order poles, nonlinear stiffness-to-torque maps, and unmodeled delays. The discussion will note that the qualitative mismatch between F_param and F_real persists under these extensions, but the precise thresholds would generally require numerical evaluation rather than closed-form derivation. revision: yes

  2. Referee: [Mechanism transfer to planar SLIP] The transfer to planar SLIP dynamics is presented as confirming COM and stance-timing deviations as primary consequences, but the manuscript should clarify whether this extension relies on additional approximations beyond the 1D case or if the α_crit threshold carries over directly without re-derivation.

    Authors: The SLIP results are obtained via numerical simulation of the closed-loop planar system under the same variable-impedance MPC formulation and first-order actuator model; no analytical re-derivation of α_crit is performed for the 2D case. The α_crit threshold is specific to the 1D monoped derivation and does not carry over directly. We will add a clarifying sentence in the SLIP section stating that the extension confirms the mechanism through simulation (highlighting COM and stance-timing deviations as primary effects) without claiming an identical analytical threshold or introducing further approximations beyond those already stated for the 1D analysis. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained analytical proof under explicit assumptions

full rationale

The paper's central result is an analytical closed-form derivation of α_crit from task physics and first-order actuator dynamics for the 1D monoped, followed by numerical validation across parameter sweeps (log-log R2 = 0.986) that is independent of the derivation. No load-bearing step reduces by construction to a fitted input, self-citation chain, or redefinition of the target quantity; the realizable set F_real is defined from actuator dynamics and the proof proceeds from there without circular reduction. The augmentation fix is explicitly labeled 'by construction' as a proposed remedy rather than a claimed prediction. This is the normal case of a self-contained derivation against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim rests on the modeling choice of first-order actuator dynamics and the definition of the two feasible sets; no free parameters are fitted and no new physical entities are postulated.

axioms (1)
  • domain assumption Actuator dynamics are first-order
    Used to define the realizable set F_real and to derive the mismatch regime via α = ω_s T.
invented entities (2)
  • F_param no independent evidence
    purpose: The feasible set assumed by the MPC optimizer when stiffness is treated as an instantaneous decision variable
    Introduced to contrast with the physically realizable set; no independent evidence outside the formulation.
  • F_real no independent evidence
    purpose: The set of stiffness commands that can actually be realized under first-order actuator dynamics
    Introduced to expose the formulation error; no independent evidence outside the model.

pith-pipeline@v0.9.0 · 5513 in / 1564 out tokens · 73019 ms · 2026-05-12T03:28:39.022991+00:00 · methodology

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    Impedance control: An approach to manipulation: Part I— Theory,

    N. Hogan, “Impedance control: An approach to manipulation: Part I— Theory,”ASME J. Dynamic Systems, Measurement, and Control, vol. 107, no. 1, pp. 1–7, Mar. 1985

    work page 1985

  2. [2]

    Impedance control: An approach to manipulation: Part II— Implementation,

    N. Hogan, “Impedance control: An approach to manipulation: Part II— Implementation,”ASME J. Dynamic Systems, Measurement, and Control, vol. 107, no. 1, pp. 8–16, Mar. 1985

    work page 1985

  3. [3]

    Impedance control: An approach to manipulation: Part III— Applications,

    N. Hogan, “Impedance control: An approach to manipulation: Part III— Applications,”ASME J. Dynamic Systems, Measurement, and Control, vol. 107, no. 1, pp. 17–24, Mar. 1985

    work page 1985

  4. [4]

    Series elastic actuators,

    G. A. Pratt and M. M. Williamson, “Series elastic actuators,” inProc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems (IROS), Pittsburgh, PA, USA, Aug. 1995, vol. 1, pp. 399–406

    work page 1995

  5. [5]

    A unified passivity-based control framework for position, torque and impedance control of flexible joint robots,

    A. Albu-Schäffer, C. Ott, and G. Hirzinger, “A unified passivity-based control framework for position, torque and impedance control of flexible joint robots,”Int. J. Robot. Res., vol. 26, no. 1, pp. 23–39, Jan. 2007

    work page 2007

  6. [6]

    Variable impedance actuators: A review,

    B. Vanderborght, A. Albu-Schäffer, A. Bicchi, E. Burdet, D. G. Caldwell, R. Carloni,et al., “Variable impedance actuators: A review,”Robot. Auton. Syst., vol. 61, no. 12, pp. 1601–1614, Dec. 2013

    work page 2013

  7. [7]

    A tank-based approach to impedance control with variable stiffness,

    F. Ferraguti, C. Secchi, and C. Fantuzzi, “A tank-based approach to impedance control with variable stiffness,” inProc. IEEE Int. Conf. Robot. Autom. (ICRA), Karlsruhe, Germany, May 2013, pp. 4948–4953

    work page 2013

  8. [8]

    M. H. Raibert,Legged Robots That Balance. Cambridge, MA, USA: MIT Press, 1986

    work page 1986

  9. [9]

    The spring-mass model for running and hopping,

    R. Blickhan, “The spring-mass model for running and hopping,”J. Biomech., vol. 22, no. 11-12, pp. 1217–1227, 1989

    work page 1989

  10. [10]

    Templates and anchors: Neuromechan- ical hypotheses of legged locomotion on land,

    R. J. Full and D. E. Koditschek, “Templates and anchors: Neuromechan- ical hypotheses of legged locomotion on land,”J. Exp. Biol., vol. 202, no. 23, pp. 3325–3332, Dec. 1999

    work page 1999

  11. [11]

    The dynamics of legged locomotion: Models, analyses, and challenges,

    P. Holmes, R. J. Full, D. Koditschek, and J. Guckenheimer, “The dynamics of legged locomotion: Models, analyses, and challenges,”SIAM Rev., vol. 48, no. 2, pp. 207–304, 2006

    work page 2006

  12. [12]

    Compliant leg behaviour explains basic dynamics of walking and running,

    H. Geyer, A. Seyfarth, and R. Blickhan, “Compliant leg behaviour explains basic dynamics of walking and running,”Proc. Royal Soc. B: Biol. Sci., vol. 273, no. 1603, pp. 2861–2867, Nov. 2006

    work page 2006

  13. [13]

    Design principles for energy-efficient legged locomotion and implementation on the MIT Cheetah robot,

    S. Seok, A. Wang, M. Y . Chuah, D. J. Hyun, J. Lee, D. M. Otten, J. H. Lang, and S. Kim, “Design principles for energy-efficient legged locomotion and implementation on the MIT Cheetah robot,”IEEE/ASME Trans. Mechatronics, vol. 20, no. 3, pp. 1117–1129, Jun. 2015

    work page 2015

  14. [14]

    ANYmal—A highly mobile and dynamic quadrupedal robot,

    M. Hutter, C. Gehring, D. Jud, A. Lauber, C. D. Bellicoso, V . Tsounis, J. Hwangbo, K. Bodie, P. Fankhauser, M. Bloesch, R. Diethelm, S. Bachmann, A. Melzer, and M. Höpflinger, “ANYmal—A highly mobile and dynamic quadrupedal robot,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), Daejeon, Korea, Oct. 2016, pp. 38–44

    work page 2016

  15. [15]

    Proprioceptive actuator design in the MIT Cheetah: Impact mitigation and high-bandwidth physical interaction for dynamic legged robots,

    P. M. Wensing, A. Wang, S. Seok, D. Otten, J. Lang, and S. Kim, “Proprioceptive actuator design in the MIT Cheetah: Impact mitigation and high-bandwidth physical interaction for dynamic legged robots,” IEEE Trans. Robot., vol. 33, no. 3, pp. 509–522, Jun. 2017

    work page 2017

  16. [16]

    Mini Cheetah: A platform for pushing the limits of dynamic quadruped control,

    B. Katz, J. Di Carlo, and S. Kim, “Mini Cheetah: A platform for pushing the limits of dynamic quadruped control,” inProc. IEEE Int. Conf. Robot. Autom. (ICRA), Montreal, Canada, May 2019, pp. 6295–6301

    work page 2019

  17. [17]

    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,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), Madrid, Spain, Oct. 2018, pp. 1–9

    work page 2018

  18. [18]

    MIT Cheetah 3: Design and control of a robust, dynamic quadruped robot,

    G. Bledt, M. J. Powell, B. Katz, J. Di Carlo, P. M. Wensing, and S. Kim, “MIT Cheetah 3: Design and control of a robust, dynamic quadruped robot,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), Madrid, Spain, Oct. 2018, pp. 2245–2252. 12

    work page 2018

  19. [19]

    Whole-body nonlinear model predictive control through contacts for quadrupeds,

    M. Neunert, M. Stäuble, M. Giftthaler, C. D. Bellicoso, J. Carius, C. Gehring, M. Hutter, and J. Buchli, “Whole-body nonlinear model predictive control through contacts for quadrupeds,”IEEE Robot. Autom. Lett., vol. 3, no. 3, pp. 1458–1465, Jul. 2018

    work page 2018

  20. [20]

    Frequency-aware model predictive control,

    R. Grandia, F. Farshidian, A. Dosovitskiy, R. Ranftl, and M. Hutter, “Frequency-aware model predictive control,”IEEE Robot. Autom. Lett., vol. 4, no. 2, pp. 1517–1524, Apr. 2019

    work page 2019

  21. [21]

    Feedback MPC for torque-controlled legged robots,

    R. Grandia, F. Farshidian, R. Ranftl, and M. Hutter, “Feedback MPC for torque-controlled legged robots,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), Macau, China, Nov. 2019, pp. 4730–4737

    work page 2019

  22. [22]

    Perceptive locomotion through nonlinear model-predictive control,

    R. Grandia, F. Jenelten, S. Yang, F. Farshidian, and M. Hutter, “Perceptive locomotion through nonlinear model-predictive control,”IEEE Trans. Robot., vol. 39, no. 5, pp. 3402–3421, Oct. 2023

    work page 2023

  23. [23]

    A unified MPC framework for whole-body dynamic locomotion and manipulation,

    J.-P. Sleiman, F. Farshidian, M. V . Minniti, and M. Hutter, “A unified MPC framework for whole-body dynamic locomotion and manipulation,” IEEE Robot. Autom. Lett., vol. 6, no. 3, pp. 4688–4695, Jul. 2021

    work page 2021

  24. [24]

    TAMOLS: Terrain- aware motion optimization for legged systems,

    F. Jenelten, R. Grandia, F. Farshidian, and M. Hutter, “TAMOLS: Terrain- aware motion optimization for legged systems,”IEEE Trans. Robot., vol. 38, no. 6, pp. 3395–3413, Dec. 2022

    work page 2022

  25. [25]

    Agile maneuvers in legged robots: A predictive control approach,

    C. Mastalli, W. Merkt, G. Xin, J. Shim, M. Mistry, I. Havoutis, and S. Vijayakumar, “Agile maneuvers in legged robots: A predictive control approach,”arXiv preprint arXiv:2203.07554, 2022

    work page arXiv 2022

  26. [26]

    A direct method for trajectory optimization of rigid bodies through contact,

    M. Posa, C. Cantu, and R. Tedrake, “A direct method for trajectory optimization of rigid bodies through contact,”Int. J. Robot. Res., vol. 33, no. 1, pp. 69–81, Jan. 2014

    work page 2014

  27. [27]

    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 Trans. Robot., vol. 37, no. 4, pp. 1154–1171, Aug. 2021

    work page 2021

  28. [28]

    Kinodynamic model predictive control for energy efficient locomotion of legged robots with parallel elasticity,

    Y . Zhuang, Y . Wang, and Y . Ding, “Kinodynamic model predictive control for energy efficient locomotion of legged robots with parallel elasticity,” inProc. IEEE Int. Conf. Robot. Autom. (ICRA), Atlanta, GA, USA, May 2025, pp. 12365–12371

    work page 2025

  29. [29]

    Con- strained model predictive control: Stability and optimality,

    D. Q. Mayne, J. B. Rawlings, C. V . Rao, and P. O. M. Scokaert, “Con- strained model predictive control: Stability and optimality,”Automatica, vol. 36, no. 6, pp. 789–814, Jun. 2000

    work page 2000

  30. [30]

    J. B. Rawlings, D. Q. Mayne, and M. M. Diehl,Model Predictive Control: Theory, Computation, and Design, 2nd ed. Madison, WI, USA: Nob Hill Publishing, 2017

    work page 2017

  31. [31]

    A real-time iteration scheme for nonlinear optimization in optimal feedback control,

    M. Diehl, H. G. Bock, and J. P. Schlöder, “A real-time iteration scheme for nonlinear optimization in optimal feedback control,”SIAM J. Control Optim., vol. 43, no. 5, pp. 1714–1736, 2005

    work page 2005

  32. [32]

    Learning agile and dynamic motor skills for legged robots,

    J. Hwangbo, J. Lee, A. Dosovitskiy, D. Bellicoso, V . Tsounis, V . Koltun, and M. Hutter, “Learning agile and dynamic motor skills for legged robots,”Sci. Robot., vol. 4, no. 26, eaau5872, Jan. 2019

    work page 2019

  33. [33]

    Learning quadrupedal locomotion over challenging terrain,

    J. Lee, J. Hwangbo, L. Wellhausen, V . Koltun, and M. Hutter, “Learning quadrupedal locomotion over challenging terrain,”Sci. Robot., vol. 5, no. 47, eabc5986, Oct. 2020

    work page 2020

  34. [34]

    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,”Sci. Robot., vol. 7, no. 62, eabk2822, Jan. 2022

    work page 2022