Pith. sign in

REVIEW 3 major objections 4 minor 12 references

VIMPPI: Enhancing Model Predictive Path Integral Control with Variational Integration for Underactuated Systems

T0 review · 3 major / 4 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 Swapping in a variational integrator extends MPPI planning horizons 4–20x and wins longer balance times on underactuated pendulums.

desk verdict Useful combination of variational integration and MPPI with strong benchmark results, but the no-cost claim is unmeasured and the correction-step Jacobian in Eq. (11) looks inconsistent with the residual in Eq. (12). read the letter →

arxiv 2505.05507 v2 pith:JK7MGVSK submitted 2025-05-07 eess.SY cs.ROcs.SY

Igor Alentev , Lev Kozlov , Ivan Domrachev , Simeon Nedelchev , Jee-Hwan Ryu This is my paper · ORCID
classification eess.SYcs.ROcs.SY
keywords modelpredictivepathintegralcontrolvariationalintegratorsymplecticintegrationunderactuateddoublependulumpendubotacrobotsampling-basedreal-time
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 argues that the bottleneck in Model Predictive Path Integral (MPPI) control is not the optimizer but the numerical integrator used in its rollouts. Replacing the standard explicit Euler integrator with a variational (symplectic) integrator lets each rollout step jump from 0.001–0.005 s to 0.02 s or more while staying accurate, stretching the effective planning horizon by 4–20x at the same compute budget. On two underactuated double pendulums (pendubot and acrobot) the resulting controller, VIMPPI, holds the upper equilibrium for up to 48 s, outperforming prior learning-based baselines and MPPI variants with other integrators. The authors position this as a general lesson: in sampling-based control of well-modeled mechanical systems, the choice of integrator deserves as much attention as the sampling scheme.

What carries the argument

The load-bearing object is the discrete Euler–Lagrange residual in momentum form, paired with a midpoint discrete Lagrangian $L_d(q_n,q_{n+1}) = L((q_n+q_{n+1})/2, (q_{n+1}-q_n)/\Delta t)\Delta t$. Instead of solving the implicit update by rootfinding, the integrator takes an explicit Euler guess and applies one correction $\Delta \hat q_i = A^{-1}(p_k - \hat p_k)$, where $p_k$ is the momentum from the known velocity, $\hat p_k$ is the momentum read off the discrete Lagrangian, and $A$ is the Jacobian of the discrete Lagrangian gradient. This single correction is what keeps the rollout accurate and effectively symplectic at $\Delta t = 0.02$ s, and it is what converts a larger timestep into a longer planning horizon for the same number of rollouts.

What would settle it

Run VIMPPI's 0.02 s rollout against a high-accuracy reference (e.g., RK4 or a variational integrator converged to tight tolerance at $\Delta t = 10^{-4}$ s) over the full 20-step horizon and measure state error and energy drift. If the single-correction rollout drifts faster than an explicit Euler rollout at 0.001 s given the same wall-clock compute budget, or if reducing the sample count to offset the larger timestep loses the reported uptime margins, the central 'longer horizon at no extra cost' claim would be refuted.

Watch

Extended reading notes

Core claim

VIMPPI's central claim is that a variational integrator—one that discretizes Hamilton's principle rather than the equations of motion—can serve as the rollout engine of MPPI, making the planning horizon 4–20 times longer without extra computational cost. Concretely, the paper replaces explicit Euler integration with a discrete Euler–Lagrange solve in momentum form: after an Euler guess for the next configuration, a single Newton-type correction based on the mismatch between true momentum and the momentum implied by the discrete Lagrangian brings the rollout onto the symplectic flow. Because energy and momentum drift are controlled at large timesteps, the cost-to-go computed over the horizon reflects true dynamics rather than numerical artifact. The paper reports that with a 0.02 s rollout timestep and a 20-step horizon at 500–700 Hz, VIMPPI maintains the unstable equilibrium for $48.03 \pm 0.9$ s on the pendubot and $38.6 \pm 2.51$ s on the acrobot, beating the next-best controllers by roughly 8 and 11 s respectively.

Load-bearing premise

The claim rests on the assumption that one Newton-style correction in the discrete Euler–Lagrange solve keeps the 0.02 s rollout accurate enough that the cost-to-go faithfully represents the true dynamics over the planning horizon; the paper reports this by observation but gives no error analysis, and if the single correction is inaccurate at large timesteps the horizon advantage over smaller-timestep integrators disappears.

Editorial extensions

If this is right

  • If the integrator is the bottleneck, then any sampling-based predictive controller for a well-modeled mechanical system can gain a 4–20x horizon extension by swapping its rollout integrator, with no change to the sampling or cost structure.
  • At equal horizon, VIMPPI's larger timestep can lower the number of rollout steps per plan, reducing per-iteration compute and allowing higher control rates (500–700 Hz here) on the same hardware.
  • The reported uptime margins (about 8 s on the pendubot, 11 s on the acrobot over the next-best controller) follow from the longer lookahead, so the benefit should grow for systems with limited control authority, where prediction quality matters more.
  • Because the variational integrator preserves conservative structure, the approach should transfer to other underactuated platforms whose dynamics are Lagrangian, such as multi-link manipulators, quadrupeds, and humanoids.

Reading between the lines

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

  • A testable extension is to run the same VIMPPI update on a damped or actuated system with non-conservative terms; variational integrators do not preserve energy there, so the claimed 4–20x advantage would likely shrink unless dissipation is folded into the discrete Lagrangian.
  • The one-step Newton correction is presented as empirically sufficient; measuring the contraction rate of the correction map as a function of $\Delta t$ would tell whether the method's accuracy guarantee extends beyond the two benchmark systems.
  • Since MPPI's importance weights depend on rollout costs, a more accurate rollout integrator may also reduce estimator variance for a fixed sample count, which could let practitioners cut the 4096 samples and recover some of the computational budget.
  • If the horizon extension is as large as claimed, VIMPPI-type rollouts could serve as a cheap forward model for model-based reinforcement learning, replacing learned dynamics for systems whose physics are known.
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 / 4 minor

Summary. The paper presents VIMPPI, a Model Predictive Path Integral (MPPI) controller for underactuated double pendulum systems (pendubot and acrobot) developed for the AI Olympics competition. The core idea is to replace the standard explicit Euler integrator in MPPI rollouts with a variational integrator based on a discrete Euler-Lagrange momentum formulation, allowing much larger rollout timesteps (0.02 s versus 0.001-0.005 s) and hence longer planning horizons 'without additional computational cost.' Additional practical components include linear control interpolation, a disturbance detection and warm-start mechanism, and a JAX implementation. Experiments on an NVIDIA RTX 4070 Ti Super show uptime improvements over several baselines from previous competitions and over MPPI variants using other integrators, with the largest gains on the acrobot.

Significance. If the central claims hold, the paper offers a simple, model-based way to extend the effective planning horizon of sampling-based MPC for underactuated systems, with potential applicability to resource-constrained platforms. The empirical comparison against multiple external baselines (evolsac, AR-EAPO, mcpilco, TVLQR) and alternative integrators is a strength, and the competition setting provides a concrete, reproducible benchmark. The use of JAX for parallel rollouts is also appropriate. However, the theoretical justification of the efficient variational-integrator solve is incomplete: the Jacobian used in the one-step correction appears incorrect, and the 'without additional computational cost' claim is not backed by any timing measurements. These issues directly affect the central contribution and must be addressed before the paper's conclusions can be accepted.

major comments (3)
  1. [Section II-B, Eqs. (10)-(13)] The one-step configuration correction is not a Newton step for the momentum residual it claims to solve. The momentum residual is r(q_{k+1}) = p_k - \hat{p}_k = p_k + \nabla_{q_{k+1}} L_d(q_k, q_{k+1}), so its derivative with respect to q_{k+1} is the pure second derivative D_2D_2 L_d. Equation (11) instead sets A = \nabla_{q_{k+1}}(\nabla_{q_k} L_d) = D_1D_2 L_d, the mixed partial. For the midpoint discrete Lagrangian (7), D_2D_2L_d = (1/\Delta t)M + (\Delta t/4)\nabla^2 V and D_1D_2L_d = -(1/\Delta t)M + (\Delta t/4)\nabla^2 V, which differ by 2M/\Delta t. At \Delta t = 0.02 s this difference is large, so the update (12)-(13) does not in general drive the momentum error to zero. The claim in Section II-B that 'a single error correction iteration is typically sufficient' is therefore unsupported by the equations. Please either use the correct Jacobian D_2D_2L_d or provide a numerical error analysis demonstrating that the iteration converges with the current A at the operating timestep.
  2. [Sections III-A and V] The central claim that the 4-20x increase in effective planning horizon is achieved 'without additional computational cost' is not substantiated by any computational cost measurement. The paper reports an operating frequency of 500-700 Hz for the full controller, but it does not report wall-clock time, FLOPs, or the number of integration steps for the variational integrator versus explicit Euler at a matched planning horizon and sample count. The variational integrator uses an extra correction step per rollout step, which has a nonzero cost; whether the larger timestep fully offsets this cost is exactly what needs to be measured. Please provide a timing or complexity comparison for VIMPPI against an Euler-based MPPI with equivalent planning horizon and sample size.
  3. [Section IV] The comparison against other integrators under 'the same setup horizon conditions' is not well defined. If all integrators are run with the same rollout timestep of 0.02 s, explicit Euler is likely unstable or highly inaccurate, making the comparison trivially favorable to the variational integrator. If each integrator instead uses a different stable timestep, then the 4-20x horizon increase is conflated with the choice of timestep, and the computational cost of each integrator for the same planning horizon is not controlled. Please specify the exact conditions (timestep, horizon, number of samples) used for each integrator and report the resulting planning accuracy and wall-clock time.
minor comments (4)
  1. [Equation (4) and Table I] The stage cost uses \Sigma_u^{-1}, while Table I lists \sigma = 0.2 I_2; please clarify the relation between \sigma and \Sigma_u and specify the units of the noise variance.
  2. [Abstract and Introduction] The phrase 'first application of such an approach' in the Introduction is already qualified with 'To our knowledge,' but the abstract states it more categorically; please align the wording in both places.
  3. [Tables II and III] The tables report 'Swingups' with mean and standard deviation, but the text emphasizes uptime; please clarify whether swingups count successful transitions or include failed attempts, and consider adding a statistical significance test between the top two controllers.
  4. [Section II-B, Eq. (10)] The notation for the momentum residual uses both q_n and q_{n+1} in Eq. (10) and q_k and q_{k+1} in Eqs. (11)-(13); please unify the index notation to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: performance claims are externally benchmarked; the variational integrator derivation, while containing an unproven one-step correction claim, does not reduce to its own inputs.

full rationale

The paper's central claim, that replacing the Euler integrator with a variational integrator in MPPI rollouts yields longer effective planning horizons, is not circular. The variational integrator equations (7)-(13) are standard midpoint discrete Lagrangian constructions; the 4-20x horizon claim is validated by head-to-head experiments with alternative integrators (implicit, implicit-fast, explicit) under identical hyperparameters (Tables II-III) and against external AI Olympics baselines. There are no self-citations: references [1]-[12] are all external works, and no uniqueness theorem or ansatz from the authors' prior work is load-bearing. The cost-to-go function and the MPPI update law are imported from external references as standard building blocks, not fitted to the target result. Hyperparameters are tuned, which is standard practice, and the improvement claims are measured on uptime and swing-up counts, not derived from the method's own equations. The only notable concern is a correctness risk rather than circularity: Eq. (11) defines A as the mixed partial D1D2L_d, while a Newton step for the momentum residual p_k + D2 L_d = 0 would require D2D2L_d; consequently, the claim that a single error correction iteration is sufficient (Sec. II-B) is not established by the paper's equations. This unsupported assertion matters for the accuracy-at-0.02s claim, but it is an omitted-proof/typo concern, not an equivalence of conclusions and inputs. The paper is self-contained against external benchmarks for its performance claims, so no significant circularity is present.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim depends on the MPPI update law, the accuracy of the single-iteration variational integrator, and a set of hand-tuned hyperparameters. No new physical entities are introduced.

free parameters (8)
  • Rollout timestep dt = 0.02 s
    Core to the claim of 4-20x longer horizon; chosen for the task, not derived.
  • Inverse temperature lambda = 50.0
    Controls softmax weighting in MPPI update; tuned.
  • Noise covariance sigma = 0.2 * I2
    Sampling noise, tuned.
  • State cost weights Q = diag(10.0, 1.0, 0.10, 0.10)
    Shapes cost-to-go; tuned.
  • Input cost weight R = diag(0.10, 0.10)
    Tuned.
  • Terminal cost weight P = 10^6 * diag(5.0, 5.0, 2.0, 2.0)
    Tuned terminal penalty.
  • Outlier penalty gamma = not reported
    Appears in Eq. (4) but no value given, essential for cost computation.
  • Number of Newton correction iterations = 1
    Design choice asserted sufficient without error analysis.
assumptions (5)
  • standard math Standard MPPI update law (Eq. 5) is valid.
    Taken from Williams et al. [10] without proof.
  • domain assumption Midpoint discrete Lagrangian approximation accurately represents the action integral.
    Sec. II-B, Eq. (7): the approximation is assumed accurate for the pendulum dynamics.
  • ad hoc to paper A single Newton iteration is sufficient for the discrete Euler-Lagrange solve.
    Sec. II-B states this without convergence or symplecticity analysis; central to the low-cost claim.
  • domain assumption Control perturbations are zero-mean Gaussian.
    Eq. (1) models delta_u ~ N(0, Sigma_u); no justification for this noise model.
  • domain assumption Simulation dynamics model matches the actual system.
    No system identification or model-error analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VIMPPI: Enhancing Model Predictive Path Integral Control with Variational Integration for Underactuated Systems." pith.science (2026). https://pith.science/paper/JK7MGVSK

@misc{pith2026250505507,
  author       = {Pith},
  title        = {Pith review of: VIMPPI: Enhancing Model Predictive Path Integral Control with Variational Integration for Underactuated Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JK7MGVSK}},
  note         = {Machine review of arXiv:2505.05507}
}
read the original abstract

This paper presents VIMPPI, a novel control approach for underactuated double pendulum systems developed for the AI Olympics competition. We enhance the Model Predictive Path Integral framework by incorporating variational integration techniques, enabling longer planning horizons without additional computational cost. Operating at 500-700 Hz with control interpolation and disturbance detection mechanisms, VIMPPI substantially outperforms both baseline methods and alternative MPPI implementations

Figures

Figures reproduced from arXiv: 2505.05507 by the authors.

Figure 1
Figure 1. Time series results for VIMPPI-AR-EAPO controller [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 4 canonical work pages

  1. [1]

    Realaigym: Education and research platform for studying athletic intelligence,

    F. Wiebe, S. Vyas, L. J. Maywald, S. Kumar, and F. Kirchner, “Realaigym: Education and research platform for studying athletic intelligence,” in Proceedings of Robotics Science and Systems Work- shop Mind the Gap: Opportunities and Challenges in the Transition Between Research and Industry, New York , 2022

    work page 2022

  2. [2]

    Open source dual-purpose acrobot and pendubot platform: Bench- marking control algorithms for underactuated robotics,

    F. Wiebe, S. Kumar, L. J. Shala, S. Vyas, M. Javadi, and F. Kirchner, “Open source dual-purpose acrobot and pendubot platform: Bench- marking control algorithms for underactuated robotics,” IEEE Robotics & Automation Magazine , vol. 31, no. 2, pp. 113–124, 2024

    work page 2024

  3. [3]

    Reinforcement learning for athletic intelligence: lessons from the 1st

    F. Wiebe, N. Turcato, A. D. Libera, C. Zhang, T. Vincent, S. Vyas, G. Giacomuzzo, R. Carli, D. Romeres, A. Sathuluri, M. Zimmermann, B. Belousov, J. Peters, F. Kirchner, and S. Kumar, “Reinforcement learning for athletic intelligence: lessons from the 1st ”ai olympics with realaigym” competition,” in Proceedings of the Thirty-Third International Joint Con...

    doi:10.24963/ijcai.2024/1043 2024

  4. [4]

    Reinforcement learning for robust athletic intelligence: Lessons from the 2nd ’ai olympics with realaigym’ competition,

    F. Wiebe, N. Turcato, A. D. Libera, J. S. B. Choe, B. Choi, T. L. Faust, H. Maraqten, E. Aghadavoodi, M. Cali, A. Sinigaglia, G. Giacomuzzo, D. Romeres, J. kook Kim, G. A. Susto, S. Vyas, D. Mronga, B. Belousov, J. Peters, F. Kirchner, and S. Kumar, “Reinforcement learning for robust athletic intelligence: Lessons from the 2nd ’ai olympics with realaigym’...

    arXiv 2025

  5. [5]

    Average-Reward Maximum Entropy Reinforcement Learning for Underactuated Double Pendulum Tasks

    J. S. B. Choe, B. Choi, and J. kook Kim, “Average-reward maximum entropy reinforcement learning for underactuated double pendulum tasks,” 2024. [Online]. Available: https://arxiv.org/abs/2409.08938

    work page Pith review arXiv 2024

  6. [6]

    Learning control of underactuated double pendulum with model-based reinforcement learning,

    N. Turcato, A. D. Libera, G. Giacomuzzo, R. Carli, and D. Romeres, “Learning control of underactuated double pendulum with model-based reinforcement learning,” 2024. [Online]. Available: https://arxiv.org/abs/2409.05811

    arXiv 2024

  7. [7]

    Ai olympics challenge with evolutionary soft actor critic,

    M. Cal `ı, A. Sinigaglia, N. Turcato, R. Carli, and G. A. Susto, “Ai olympics challenge with evolutionary soft actor critic,” 2024. [Online]. Available: https://arxiv.org/abs/2409.01104

    arXiv 2024

  8. [8]

    Sampling-based motion planning: A comparative review,

    A. Orthey, C. Chamzas, and L. E. Kavraki, “Sampling-based motion planning: A comparative review,” 2023. [Online]. Available: https://arxiv.org/abs/2309.13119

    arXiv 2023

Show all 12 references
  1. [9]

    Aggressive driving with model predictive path integral control,

    G. Williams, P. Drews, B. Goldfain, J. M. Rehg, and E. A. Theodorou, “Aggressive driving with model predictive path integral control,” in 2016 IEEE international conference on robotics and automation (ICRA). IEEE, 2016, pp. 1433–1440

    2016

  2. [10]

    Model Predictive Path Integral Control: From Theory to Parallel Computation,

    G. Williams, A. Aldrich, and E. A. Theodorou, “Model Predictive Path Integral Control: From Theory to Parallel Computation,” Journal of Guidance, Control, and Dynamics, vol. 40, no. 2, pp. 344–357, Feb. 2017, publisher: American Institute of Aeronautics and Astronautics. [Onli...

    2017 doi

  3. [11]

    Variational integrators,

    M. West, “Variational integrators,” Ph.D. dissertation, California Institute of Technology, 2003. [Online]. Available: https://thesis. library.caltech.edu/2492/

    2003

  4. [12]

    JAX: composable transformations of Python+NumPy programs,

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman- Milne, and Q. Zhang, “JAX: composable transformations of Python+NumPy programs,” 2018. [Online]. Available: http://github. com/jax-ml/jax

    2018

Pith tools

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