Pith. sign in

REVIEW 4 major objections 5 minor 47 references

Constrained Stein Variational Gradient Descent for Robot Perception, Planning, and Identification

T0 review · 4 major / 5 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 This paper proposes two methods, the Q method and the p method, that make Stein variational gradient descent approximate distributions whose samples all respect arbitrary constraints.

desk verdict A useful unification and a genuinely new p-method, but a sign error in Eq. 17 and unproven exact-feasibility claims make this a conditional accept. read the letter →

arxiv 2506.00589 v1 pith:SS42L2DT submitted 2025-05-31 cs.RO cs.LG

Griffin Tabor , Tucker Hermans This is my paper
classification cs.ROcs.LG
keywords SteinvariationalgradientdescentconstrainedinferencerobotmotionplanninginversekinematicspointcloudposeestimationaugmentedLagrangianpenaltymethodsparticle-based
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

Stein variational gradient descent (SVGD) is a particle-based approximate inference method that is useful in robotics precisely because it returns many diverse solutions, but it has no built-in way to enforce constraints. The paper argues that constraints can be added generically in one of two places: inside the variational family, by allowing only feasible particle positions (the Q method), or inside the target distribution, by reshaping it so infeasible points are penalized (the p method). Both recipes reuse standard constrained-optimization machinery—augmented Lagrangian, quadratic penalty, and log-barrier terms—and both are demonstrated on robot motion planning, inverse kinematics with exact placement constraints, and point-cloud object pose estimation. The result that matters for robotics is that particles can be kept feasible while preserving the diversity SVGD is valued for, with the Q method converging faster and handling equality constraints that the p method struggles with.

What carries the argument

The central object is Stein variational gradient descent, which represents an approximate distribution as a set of particles and updates each particle with a functional gradient combining a kernel-weighted pull toward high target density and a repulsive term for diversity. The machinery introduced here is the constraint wrapper: in the Q method, a soft-constraint cost $L_\theta$ from the constrained-optimization literature contributes its negative gradient to each particle's SVGD update, box constraints are enforced by projection, and the constraint parameters $\theta$ are updated in an outer loop once the inner SVGD loop converges; in the p method, the same $L_\theta$ is folded into the target density as $\hat p(x)\propto e^{-\alpha f(x)}e^{-L_\theta(x)}$, so constraint information flows between particles through the kernel. The outer-inner loop structure, where an unconstrained SVGD solve alternates with an update of penalty weights or Lagrange multipliers, is what lets the soft constraints grow strong enough to approximate hard feasibility.

What would settle it

A concrete check is to take an equality-constrained problem whose constrained distribution is known in closed form (for instance, a Gaussian restricted to a circle), run the Q method with a fixed finite inner-loop budget, and measure the maximum constraint violation and the Wasserstein distance to the true constrained distribution as the penalty multiplier grows; the central claim fails if the violation or the distance does not decrease to zero. A second check would test the p method's equality-constraint failure on a one-particle system, where kernel averaging is absent; if it still fails, the difficulty is in the shared Lagrange multipliers rather than the kernel.

Watch

Extended reading notes

Core claim

The paper's central claim is that constrained optimization principles can be applied to SVGD in two general ways without changing SVGD's particle-based core. The Q method restricts the search to particle configurations in which every particle satisfies the constraints, then solves the constrained KL-minimization by alternating SVGD updates with soft-constraint gradient terms and a projection step for box constraints; the p method instead builds a sequence of smoothed target distributions of the form $\hat p(x) \propto e^{-\alpha f(x)}\prod_i e^{-L_{\theta,i}(x)}$ and runs ordinary SVGD against the smoothed target, updating shared constraint parameters between inner loops. The same soft-constraint formulations (augmented Lagrangian, quadratic penalty, relaxed log barrier) serve both methods. In experiments, both approaches produce collision-free trajectory distributions, arm joint-angle distributions satisfying exact table-placement equality constraints, and table-constrained object pose distributions from point clouds, and the Q method reaches feasibility in substantially fewer gradient steps. The paper reports that the Q method preserves the underlying distribution while the p method's shared constraint gradients lose their way when the feasible set is a lower-dimensional manifold.

Load-bearing premise

The load-bearing premise is that running the inner SVGD loop to convergence and updating the penalty or Lagrange multipliers in the outer loop drives the particles to the exact feasible distribution, since no theorem or error bound is given for that convergence; the abstract's promise of exact collision avoidance and exact placement constraints depends on it.

Editorial extensions

If this is right

  • Existing constrained-SVGD algorithms, including the augmented-Lagrangian form of [31], fit inside the Q method, so they can be viewed as instances of a single recipe rather than separate inventions.
  • The Q method can turn a deterministic constrained trajectory optimizer into a generator of diverse, collision-free plans, since independent runs of the optimizer tend to produce duplicate trajectories.
  • The p method can under-approximate target distributions more faithfully than Q on inequality-constrained problems, as measured by Earth mover distance on the 2D benchmark, but at the cost of more gradient steps.
  • On equality constraints whose feasible sets are lower-dimensional manifolds, the Q method remains practical while the p method's shared constraint gradients can fail to converge within reasonable budgets.
  • Both methods can be used as anytime algorithms: with many particles initialized, the number of feasible particles grows with compute, giving an anytime trade-off between diversity and feasibility.

Reading between the lines

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

  • A natural extension the authors do not test is applying the same Q-method wrapper to other particle-based samplers (e.g., stochastic-gradient Langevin dynamics), since the recipe only assumes a gradient-based particle update and a soft-constraint cost.
  • The p method's failure mode suggests a diagnostic: when the constrained feasible set has codimension greater than zero, kernel-averaged soft-constraint gradients dilute the direction toward feasibility; one could predict this failure from the kernel's gradient norm restricted to the constraint surface.
  • If a certified guarantee is desired, the missing piece is an error bound between the constrained target $p$ of Eq. 14 and the particles after a finite number of outer iterations; without it, 'exact' feasibility rests on the empirical convergence of the inner SVGD loop.
  • A testable design choice is replacing box projection with a manifold projection or a kernel defined on the feasible set, which could let the Q method handle arbitrary equality constraints without relying on growing penalty weights.
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.

Referee Report

4 major / 5 minor

Summary. This paper introduces two ways to incorporate arbitrary equality and inequality constraints into Stein variational gradient descent (SVGD). The 'Q method' restricts the variational family to particle configurations whose particles are feasible and adds a soft-constraint gradient (augmented Lagrangian, quadratic penalty, or log barrier) to the SVGD update, with projection used only for box bounds. The 'p method' redefines the target as a soft-constrained density proportional to e^{-alpha f(x)} times products of constraint terms and runs ordinary SVGD on that density, updating the constraint parameters in an outer loop. The authors illustrate both methods on a 2D problem and evaluate them on 2D trajectory optimization with collision constraints, a robot-arm inverse-kinematics/placement problem with SE(3) equality constraints, and Stein-ICP pose estimation with table constraints. They report that the Q method converges faster and that the p method can fail on low-dimensional equality constraints (Section V-B), and they claim that the resulting samples satisfy constraints exactly.

Significance. The proposed unification is potentially useful: it places the augmented-Lagrangian SVGD method of [31] and CVSTO [19] in a common family and gives practitioners a menu of soft-constraint mechanisms. The empirical comparison of Q versus p on three robotic tasks is informative, particularly the observation that sharing constraint gradients through the kernel (the p method) degrades on equality-constrained manifolds; this is a falsifiable, reproducible finding in principle. I did not find any fitted-constant circularity: the method is evaluated directly against the intended constrained distribution. However, the contribution is currently undermined by the sign error in Eq. (17) and by the absence of any guarantee or tolerance analysis for exact feasibility, so the advertised 'exactly avoid collisions' claim is not supported by the presented analysis.

major comments (4)
  1. [IV-A, Eq. (17)] The displayed steepest-descent term has the opposite sign from Eq. (8). For p(x) = e^{-alpha f(x)}, we have grad log p = -alpha grad f, so the SVGD functional gradient is phi* = (1/n) sum_j [ -alpha k grad f + grad k ], not (1/n) sum_j [ alpha k grad f - grad k ]. As written, Eq. (17) updates particles in the direction of increasing f and replaces the repulsive term with an attractive one. This is a load-bearing error because Eq. (17) is the paper's only explicit instantiation of the Q method for equality constraints; please correct it and state whether the experiments use the corrected sign.
  2. [IV-A, Algorithm 1 and Eq. (16)] The definition of the Q method in Eq. (16) requires every particle to be exactly feasible, and the abstract promises collision-free and placement-constrained samples. Algorithm 1 nevertheless enforces only the box constraints by projection (lines 3 and 6); the equality and inequality constraints enter only through the gradient of the soft-constraint term L_theta. For finite penalty and multiplier values, the resulting particles are not exactly feasible, and no theorem or error bound is given for the outer loop. This gap is load-bearing because the paper's two main frameworks are characterized as producing distributions 'without violating constraints.' The authors should either prove a convergence or tolerance statement for Algorithm 1 or explicitly soften the exactness claims.
  3. [IV-B, Eq. (18) and Algorithm 2] For the p method, Eq. (18) defines p_hat proportional to e^{-alpha f(x)} times product_i e^{-mu h_i(x)^2}; for any finite mu this density assigns positive probability to infeasible points and therefore does not equal the target p in Eq. (14), whose support is exactly the feasible set. The outer-loop multiplier updates are intended to close this gap, but the manuscript provides no convergence argument for the sequence of soft targets, and the min-pooling rule for shared multipliers is introduced without analysis. Since Algorithm 2 is claimed to approximate p under constraints, the limit of the soft-target sequence and the effect of the pooling choice should be analyzed, or the claims should be restricted to a soft-constraint approximation.
  4. [V-B] The paper reports that, on the SE(3) equality-constrained inverse-kinematics problem, the p method with 50 particles failed to converge in 25 times the number of steps needed by the Q method, and that with 10 particles it converged only after 100,000 iterations. This is not merely a performance difference: it shows that the general claim of learning distributions 'without violating constraints' does not hold for a core problem class studied in the paper. The manuscript should either characterize when each method is guaranteed to work or remove the universal statement and present the p method as an approximate, task-dependent alternative.
minor comments (5)
  1. [III-B, Eq. (13)] As typeset, the relaxed log-barrier branches do not match at g = -delta; the quadratic branch should be continuous with the log branch, for example 1/2 ((g+2 delta)/delta)^2 - 1/2 - log(delta). Please fix the typo.
  2. [IV-A and IV-B] The backtracking line search minimizes sum_j (-log p(z_j) + L_theta(z_j)) and omits the SVGD repulsive term. The authors acknowledge this, but the adjacent claim in Section IV-B that diversity is still improved during optimization should be supported empirically, since a line search that ignores repulsion can select step sizes that reduce the repulsion-driven spread.
  3. [V-A, Fig. 3] The figure reports total gradient steps over 50 trials, but no dispersion measure or statistical test is provided; please add medians and interquartile ranges or individual trial points so the reader can assess the claimed speed advantage.
  4. [V-C] The statement that 'the rotation about y and z are exactly zero for all particles' is inconsistent with the stated constraint that the object z-axis is vertical and with Fig. 5, which shows a wide distribution about z; the sentence should say x and y, or the constraint should be restated.
  5. [Title and Section V] The title and introduction promise applications to system identification, but Section V contains no system identification experiment; either add one or adjust the title and abstract to match the presented experiments.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the Q and p methods are algorithmic extensions of SVGD evaluated empirically, with no fitted quantity or self-citation standing in for the claimed result.

full rationale

The claimed derivation chain is: Eq. 14 defines the constrained target p as zero outside the feasible set, Eq. 15-16 define the Q method's feasible variational family, Eq. 18 defines the p method's reshaped target, and Algorithms 1 and 2 apply the standard SVGD update from Eq. 8 to these objects. None of these steps is obtained by fitting a parameter to the evaluation metric, and the empirical comparisons (EMD, gradient-step counts) are not forced by construction: the SE(3) experiment in Section V-B shows a genuine failure mode ("the p method failed to converge in 25 times that many steps"), which would not occur if success were built into the update. The only self-citations ([2], and [4] with one co-author) are contextual motivation for multi-modal SVGD and are not load-bearing premises. The paper's rigor gaps are real but non-circular: exact feasibility is asserted rather than proven, since the inner loop is run "until converged" with no convergence theorem for the penalty/multiplier outer loop, and Eq. 17 has an apparent sign discrepancy with Eq. 8. These are correctness concerns, not a reduction of the output to the input. There is no self-definitional step, no fitted constant renamed as a prediction, no uniqueness theorem imported from the authors, and no ansatz smuggled in via citation.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The method rests on standard SVGD theory plus several unproven algorithmic choices: convergence of the inner/outer loop, min-pooling, and line-search. No new physical entities are introduced. The free parameters are all algorithm hyperparameters, none derived from first principles.

free parameters (5)
  • alpha (temperature in p(x) = 1/Z e^{-alpha f(x)}) = hand-tuned per problem (not reported numerically)
    Controls the sharpness of the target distribution; appears in Eq. 14 and is used in all experiments.
  • penalty weights (mu, c, d) and Lagrange multipliers (lambda, gamma) = updated by outer loop; schedules not fully specified
    The augmented Lagrangian and quadratic penalty terms in Eq. 10 are central to the Q and p methods; their schedules determine whether particles become feasible.
  • kernel bandwidth h = median heuristic or hand-tuned, e.g., h based on median heuristic in ICP section
    The RBF kernel bandwidth strongly affects SVGD particle diversity and convergence; in the ICP experiment they say 'hand tuned value h based on the median heuristic.'
  • step size epsilon = selected by backtracking line search; schedule not specified
    Step length is needed in Alg. 1 and Alg. 2; the line search objective omits the repulsive term, which may alter convergence.
  • number of particles n = 50, 10, 100, 300, 2000 depending on experiment
    Particle count is chosen per experiment and affects distribution quality.
assumptions (6)
  • standard math SVGD functional gradient is optimal in an RKHS with kernel k (Liu et al. [39]).
    The method builds on the standard SVGD update in Eq. 6-8.
  • standard math Stein's identity holds for the target distribution p and the kernel.
    Background of SVGD, Section III.A.
  • domain assumption The constrained set is non-empty and constraints are differentiable so penalty gradients are defined.
    Section IV assumes equality and inequality constraints h and g are smooth enough for gradient computation.
  • ad hoc to paper The inner-loop SVGD optimization converges to its stationary distribution for the augmented cost.
    No proof is given; the paper runs 'until converged' (Alg. 1, Alg. 2). This is a load-bearing but unproven premise.
  • ad hoc to paper Min-pooling of constraint violations for the p method's Lagrange multiplier update is a valid choice.
    Section IV.B: they choose min pooling over max/mean/median with only 'limited testing'; this choice affects the p method's behavior.
  • ad hoc to paper Backtracking line search with the negative log p (without repulsive term) provides a suitable shared step size.
    Section IV.A describes this heuristic; the repulsive term is excluded from the line search objective, which could bias updates.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Constrained Stein Variational Gradient Descent for Robot Perception, Planning, and Identification." pith.science (2026). https://pith.science/paper/SS42L2DT

@misc{pith2026250600589,
  author       = {Pith},
  title        = {Pith review of: Constrained Stein Variational Gradient Descent for Robot Perception, Planning, and Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SS42L2DT}},
  note         = {Machine review of arXiv:2506.00589}
}
read the original abstract

Many core problems in robotics can be framed as constrained optimization problems. Often on these problems, the robotic system has uncertainty, or it would be advantageous to identify multiple high quality feasible solutions. To enable this, we present two novel frameworks for applying principles of constrained optimization to the new variational inference algorithm Stein variational gradient descent. Our general framework supports multiple types of constrained optimizers and can handle arbitrary constraints. We demonstrate on a variety of problems that we are able to learn to approximate distributions without violating constraints. Specifically, we show that we can build distributions of: robot motion plans that exactly avoid collisions, robot arm joint angles on the SE(3) manifold with exact table placement constraints, and object poses from point clouds with table placement constraints.

Figures

Figures reproduced from arXiv: 2506.00589 by the authors.

Figure 1
Figure 1. Our methods shown on f(x) = x1 + x2 with the inequality constraint 2 − x 2 1 − x 2 2 ≥ 0. To ensure it’s finite, the total probability mass, Z, for each plot is only computed within the bound constraints. After some amount of unconstrained optimizer steps, we can update the parameters of each of the constraints based on the particle that violates it the most or independently for each particle. When applicable, for f… view at source ↗
Figure 3
Figure 3. Total gradient steps until convergence under different [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 2
Figure 2. Trajectory optimization problem initialized identically [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Constrained Stein variational gradient descent using [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: Identical problem to Fig. 5 but without the constraints. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 5
Figure 5. Figure 5: 300 particles approximating the distribution created [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 41 canonical work pages

  1. [31]

    Probabilistic Inference of Simulation Parameters via Parallel Differentiable Simulation

    E. Heiden, C. E. Denniston, D. Millard, F. Ramos, and G. S. Sukhatme, “Probabilistic inference of simulation parameters via parallel differentiable simulation,” CoRR, vol. abs/2109.08815, 2021. [Online]. Available: https://arxiv.org/abs/2109.08815

    work page Pith review arXiv 2021

  2. [19]

    Constrained stein variational trajectory optimization,

    T. Power and D. Berenson, “Constrained stein variational trajectory optimization,” 2023

    work page 2023

  3. [1]

    Thrun, W

    S. Thrun, W. Burgard, and D. Fox, Probabilistic robotics, ser. Intelli- gent robotics and autonomous agents. MIT Press, 2005

    work page 2005

  4. [2]

    Adaptive magnetic control using stein variational gradient descent computed distribution of object parameters,

    G. Tabor and T. Hermans, “Adaptive magnetic control using stein variational gradient descent computed distribution of object parameters,” in IROS 2023 Workshop on Differentiable Probabilistic Robotics: Emerging Perspectives on Robot Learning , 2023. [Online]. Available: https://openreview.net/forum?id=dWE6dGTiTS

    work page 2023

  5. [3]

    Robot motion planning in dynamic, uncertain environments,

    N. E. Du Toit and J. W. Burdick, “Robot motion planning in dynamic, uncertain environments,” IEEE Transactions on Robotics , vol. 28, no. 1, pp. 101–115, 2012

    work page 2012

  6. [4]

    Planning multi-fingered grasps as probabilistic inference in a learned deep network,

    Q. Lu, K. Chenna, B. Sundaralingam, and T. Hermans, “Planning multi-fingered grasps as probabilistic inference in a learned deep network,” in Robotics Research, N. M. Amato, G. Hager, S. Thomas, and M. Torres-Torriti, Eds. Cham: Springer International Publishing, 2020, pp. 455–472

    work page 2020

  7. [5]

    Variational inference MPC using normalizing flows and out-of-distribution projection,

    T. Power and D. Berenson, “Variational inference MPC using normalizing flows and out-of-distribution projection,” in Robotics: Science and Systems XVIII, New York City, NY, USA, June 27 - July 1, 2022 , K. Hauser, D. A. Shell, and S. Huang, Eds., 2022. [Online]. Available: https://doi.org/10.15607/RSS.2022.XVIII.027

    work page doi:10.15607/rss.2022.xviii.027 2022

  8. [6]

    A gaussian variational inference approach to motion planning,

    H. Yu and Y . Chen, “A gaussian variational inference approach to motion planning,” IEEE Robotics and Automation Letters , vol. 8, no. 5, p. 2518–2525, May 2023. [Online]. Available: http://dx.doi.org/10.1109/LRA.2023.3256134

    work page arXiv 2023

Show all 47 references
  1. [7]

    Nlp sampling: Combining mcmc and nlp methods for diverse constrained sampling,

    M. Toussaint, C. V . Braun, and J. Ortiz-Haro, “Nlp sampling: Combining mcmc and nlp methods for diverse constrained sampling,”

  2. [8]

    Variational inference: A review for statisticians,

    D. M. Blei, A. Kucukelbir, and J. D. McAuliffe, “Variational inference: A review for statisticians,” Journal of the American Statistical Association, vol. 112, no. 518, pp. 859–877, apr 2017. [Online]. Available: https://doi.org/10.1080%2F01621459.2017.1285773

    2017

  3. [9]

    On information and sufficiency,

    S. Kullback and R. A. Leibler, “On information and sufficiency,” The annals of mathematical statistics , vol. 22, no. 1, pp. 79–86, 1951

    1951

  4. [10]

    Sliced wasserstein variational inference,

    M. Yi and S. Liu, “Sliced wasserstein variational inference,” 2022

    2022

  5. [11]

    Universal boosting variational inference,

    T. Campbell and X. Li, “Universal boosting variational inference,” 2019

    2019

  6. [12]

    Variational inference MPC using tsallis divergence,

    Z. Wang, O. So, J. Gibson, B. I. Vlahov, M. Gandhi, G. Liu, and E. A. Theodorou, “Variational inference MPC using tsallis divergence,” in Robotics: Science and Systems XVII, Virtual Event, July 12-16, 2021 , D. A. Shell, M. Toussaint, and M. A. Hsieh, Eds., 2021. [Online]. Ava...

    2021 doi

  7. [13]

    f-divergence variational inference,

    N. Wan, D. Li, and N. HOV AKIMY AN, “f-divergence variational inference,” in Advances in Neural Information Processing Systems , H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 17 370–17 379. [Online]. Available: ...

    2020

  8. [14]

    Chapter 57 - markov chain monte carlo methods: Computation and inference,

    S. Chib, “Chapter 57 - markov chain monte carlo methods: Computation and inference,” ser. Handbook of Econometrics, J. J. Heckman and E. Leamer, Eds. Elsevier, 2001, vol. 5, pp. 3569–3649. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S1573441201050103

    2001

  9. [15]

    Integrating bottom-up/top-down for object recognition by data driven markov chain monte carlo,

    S.-C. Zhu, R. Zhang, and Z. Tu, “Integrating bottom-up/top-down for object recognition by data driven markov chain monte carlo,” in Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No.PR00662) , vol. 1, 2000, pp. 738– 745 vol.1

    2000

  10. [16]

    Dual online stein variational inference for control and dynamics,

    L. Barcelos, A. Lambert, R. Oliveira, P. Borges, B. Boots, and F. Ramos, “Dual online stein variational inference for control and dynamics,” 2021

    2021

  11. [17]

    Stein variational model predictive control,

    A. Lambert, A. Fishman, D. Fox, B. Boots, and F. Ramos, “Stein variational model predictive control,” 2021

    2021

  12. [18]

    Stein variational guided model predictive path integral control: Proposal and experiments with fast maneuvering vehicles,

    K. Honda, N. Akai, K. Suzuki, M. Aoki, H. Hosogaya, H. Okuda, and T. Suzuki, “Stein variational guided model predictive path integral control: Proposal and experiments with fast maneuvering vehicles,” 2023

    2023

  13. [20]

    Task-space kernels for diverse stein variational MPC,

    M. S. Sharma, T. Power, and D. Berenson, “Task-space kernels for diverse stein variational MPC,” in IROS 2023 Workshop on Differentiable Probabilistic Robotics: Emerging Perspectives on Robot Learning, 2023. [Online]. Available: https://openreview.net/forum?id= IeG4J5TF79

    2023

  14. [21]

    Stamp: Differentiable task and motion planning via stein variational gradient descent,

    Y . Lee, P. Huang, K. M. Jatavallabhula, A. Z. Li, F. Damken, E. Heiden, K. Smith, D. Nowrouzezahrai, F. Ramos, and F. Shkurti, “Stamp: Differentiable task and motion planning via stein variational gradient descent,” 2023

    2023

  15. [23]

    Stein particle filtering,

    J. Fan, A. Taghvaei, and Y . Chen, “Stein particle filtering,” 2021

    2021

  16. [24]

    Stein variational belief propagation for multi-robot coordination,

    J. Pavlasek, J. J. Z. Mah, R. Xu, O. C. Jenkins, and F. Ramos, “Stein variational belief propagation for multi-robot coordination,” IEEE Robotics and Automation Letters , vol. 9, no. 5, pp. 4194–4201, 2024

    2024

  17. [25]

    Altro: A fast solver for constrained trajectory optimization,

    T. A. Howell, B. E. Jackson, and Z. Manchester, “Altro: A fast solver for constrained trajectory optimization,” in 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2019, pp. 7674–7679

    2019

  18. [26]

    Finding locally optimal, collision-free trajectories with sequential convex optimization,

    J. Schulman, J. Ho, A. X. Lee, I. Awwal, H. Bradlow, and P. Abbeel, “Finding locally optimal, collision-free trajectories with sequential convex optimization,” Robotics: Science and Systems IX , 2013. [Online]. Available: https://api.semanticscholar.org/CorpusID:2393365

    2013

  19. [27]

    In-hand object-dynamics inference using tactile fingertips,

    B. Sundaralingam and T. Hermans, “In-hand object-dynamics inference using tactile fingertips,” CoRR, vol. abs/2003.13165, 2020. [Online]. Available: https://arxiv.org/abs/2003.13165

    2003 arXiv

  20. [28]

    Linear matrix inequalities for physically consistent inertial parameter identification: A statistical perspective on the mass distribution,

    P. M. Wensing, S. Kim, and J.-J. E. Slotine, “Linear matrix inequalities for physically consistent inertial parameter identification: A statistical perspective on the mass distribution,” IEEE Robotics and Automation Letters, vol. 3, no. 1, pp. 60–67, 2018

    2018

  21. [29]

    Robust iterative closest point algorithm with bounded rotation angle for 2d registration,

    C. Zhang, S. Du, J. Liu, Y . Li, J. Xue, and Y . Liu, “Robust iterative closest point algorithm with bounded rotation angle for 2d registration,” Neurocomputing, vol. 195, pp. 172–180, 2016, learning for Medical Imaging. [Online]. Available: https: //www.sciencedirect.com/scie...

    2016

  22. [30]

    Kinematic-icp: Enhancing lidar odometry with kinematic constraints for wheeled mobile robots moving on planar surfaces,

    T. Guadagnino, B. Mersch, I. Vizzo, S. Gupta, M. V . R. Malladi, L. Lobefaro, G. Doisy, and C. Stachniss, “Kinematic-icp: Enhancing lidar odometry with kinematic constraints for wheeled mobile robots moving on planar surfaces,” 2024. [Online]. Available: https://arxiv.org/abs/...

    2024 arXiv

  23. [32]

    Sampling in constrained domains with orthogonal-space variational gradient descent,

    R. Zhang, Q. Liu, and X. T. Tong, “Sampling in constrained domains with orthogonal-space variational gradient descent,” 2022

    2022

  24. [33]

    Invariant EKF design for scan matching-aided localization,

    M. Barczyk, S. Bonnabel, J. Deschaud, and F. Goulette, “Invariant EKF design for scan matching-aided localization,” CoRR, vol. abs/1503.01407, 2015. [Online]. Available: http://arxiv.org/abs/1503. 01407

    2015 arXiv

  25. [34]

    Stein particle filter for nonlinear, non-gaussian state estimation,

    F. A. Maken, F. Ramos, and L. Ott, “Stein particle filter for nonlinear, non-gaussian state estimation,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 5421–5428, 2022

    2022

  26. [35]

    Dual online stein variational inference for control and dynamics,

    L. Barcelos, A. Lambert, R. Oliveira, P. Borges, B. Boots, and F. Ramos, “Dual online stein variational inference for control and dynamics,” CoRR, vol. abs/2103.12890, 2021. [Online]. Available: https://arxiv.org/abs/2103.12890

    2021 arXiv

  27. [36]

    Adaptive unscented kalman filter for online state, parameter, and process covariance estimation,

    M. H. Riva, M. Dagen, and T. Ortmaier, “Adaptive unscented kalman filter for online state, parameter, and process covariance estimation,” in 2016 American Control Conference (ACC) , 2016, pp. 4513–4519

    2016

  28. [37]

    Stochastic motion planning as gaussian variational inference: Theory and algorithms,

    H. Yu and Y . Chen, “Stochastic motion planning as gaussian variational inference: Theory and algorithms,” 2024. [Online]. Available: https://arxiv.org/abs/2308.14985

    2024 arXiv

  29. [38]

    Stein variational gradient descent: A general purpose bayesian inference algorithm,

    Q. Liu and D. Wang, “Stein variational gradient descent: A general purpose bayesian inference algorithm,” 2019

    2019

  30. [39]

    A kernelized stein discrepancy for goodness-of-fit tests and model evaluation,

    Q. Liu, J. D. Lee, and M. I. Jordan, “A kernelized stein discrepancy for goodness-of-fit tests and model evaluation,” 2016

    2016

  31. [40]

    Nocedal and S

    J. Nocedal and S. J. Wright, Numerical Optimization , 2nd ed. New York, NY , USA: Springer, 2006

    2006

  32. [41]

    A novel augmented lagrangian approach for inequalities and convergent any-time non-central updates,

    M. Toussaint, “A novel augmented lagrangian approach for inequalities and convergent any-time non-central updates,” 2014

    2014

  33. [42]

    Feedback mpc for torque-controlled legged robots,

    R. Grandia, F. Farshidian, R. Ranftl, and M. Hutter, “Feedback mpc for torque-controlled legged robots,” in 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2019, pp. 4730– 4737

    2019

  34. [43]

    A barrier function method for the optimiza- tion of trajectory functionals with constraints,

    J. Hauser and A. Saccon, “A barrier function method for the optimiza- tion of trajectory functionals with constraints,” in Proceedings of the 45th IEEE Conference on Decision and Control , 2006, pp. 864–869

    2006

  35. [44]

    Projected stein variational gradient descent,

    P. Chen and O. Ghattas, “Projected stein variational gradient descent,” in Proceedings of the 34th International Conference on Neural Infor- mation Processing Systems , ser. NIPS ’20. Red Hook, NY , USA: Curran Associates Inc., 2020

    2020

  36. [45]

    Motion planning as probabilistic inference using gaussian processes and factor graphs,

    J. Dong, M. Mukadam, F. Dellaert, and B. Boots, “Motion planning as probabilistic inference using gaussian processes and factor graphs,” in Robotics: Science and Systems , 2016. [Online]. Available: https://api.semanticscholar.org/CorpusID:17831189

    2016

  37. [46]

    Efficient reinforcement learning using gaussian processes,

    M. P. Deisenroth, “Efficient reinforcement learning using gaussian processes,” Ph.D. dissertation, 2010

    2010

  38. [2021]

    Available: https://arxiv.org/abs/2106.03287

    [Online]. Available: https://arxiv.org/abs/2106.03287

    arXiv

  39. [2024]

    Available: https://arxiv.org/abs/2407.03035

    [Online]. Available: https://arxiv.org/abs/2407.03035

Pith tools

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