Pith. sign in

REVIEW 3 major objections 5 minor 4 cited by

Diffusion Predictive Control with Constraints

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Diffusion policies trained offline can be made to satisfy novel test-time constraints by inserting a dynamics-aware projection into every denoising step and tightening constraints by the known model-error bound.

desk verdict DPCC is a practical, well-evaluated method for constraining diffusion policies at test time, but its headline guarantee rests on an unproven recursive-feasibility assumption. read the letter →

arxiv 2412.09342 v2 pith:GKNOCBOL submitted 2024-12-12 cs.RO cs.LGcs.SYeess.SY

Ralf Römer , Alexander von Rohr , Angela P. Schoellig This is my paper · ORCID
classification cs.ROcs.LGcs.SYeess.SY
keywords diffusionpoliciestrajectorymodelpredictivecontrolconstraintsatisfactionimitationlearningrecedinghorizontighteningrobotmanipulation
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

The paper claims that an offline-trained trajectory diffusion policy can be turned into a controller that respects novel state and action constraints absent from its training data, without retraining. The device is to modify each backward-diffusion step so the sampled trajectory is projected onto a set of dynamically feasible, constraint-satisfying trajectories, and to tighten the constraints by a margin that absorbs the bounded mismatch between the model used for projection and the true system. In the robot-manipulator simulations, this combination raises the rate of reaching the goal while satisfying constraints from below 50% (projection without tightening) to approximately 98%, with essentially zero constraint violations. If the claim is right, diffusion-based imitation learning gains the online constraint-handling ability of model predictive control while keeping the expressiveness of generative models.

What carries the argument

The operative object is the constraint- and model-based projection $\Pi_{Z_f}$, defined as the nearest trajectory (in $\ell^2$) that satisfies the state and action constraints and the dynamics equality constraints. It is applied to the noisy denoising output at each of the $K$ diffusion steps, which makes the final trajectory feasible by construction. The second piece is the tightening operation $\tilde S_{t+1} = S_{t+1} \ominus B_\gamma$, which inflates the safety margin by the worst-case disturbance $\gamma$; this absorbs model mismatch so that predicted feasibility implies actual feasibility under the true dynamics. Together they convert a learned generative prior into a receding-horizon controller whose closed-loop trajectories are dynamically feasible and constraint-satisfying.

What would settle it

Run the closed-loop experiment under the assumed model-mismatch bound $\gamma$ and, at every replanning step, record whether the current state lies in the tightened set $\tilde S_t$. If a state in $S_t$ but outside $\tilde S_t$ is ever used as the start of a plan, the premise of Theorem 2 is void; and if constraints are violated while $\|w_t\|_2 \le \gamma$ holds, the claimed guarantee fails. A single such episode would settle the matter.

Watch

Extended reading notes

Core claim

The central claim is that constraint satisfaction can be guaranteed, not merely encouraged, by embedding model-based projections into the denoising process. The paper defines the feasible trajectory set $Z_f$ of trajectories whose states and actions satisfy the constraints and whose dynamics agree with the model $f$, and replaces the plain denoising step with the model-informed step $\tau^{k-1} = \Pi_{Z_f}(\mu_\theta(\tau^k,k,c) + \sigma_k \epsilon_k)$ (Eq. 16). The projection is applied at every diffusion iteration, so the final sample lies in $Z_f$ for any noise schedule. To handle the discrepancy between $f$ and the true dynamics $s_{t+1}=f(s_t,a_t)+w_t$ with $\|w_t\|_2 \le \gamma$, the state constraints are tightened to $\tilde S_{t+1} = S_{t+1} \ominus B_\gamma$; Theorem 2 then asserts that if each replanning samples from the tightened feasible set, all future actual states stay inside the original constraints. The paper also supplies two trajectory-selection criteria to avoid mode-switching and to preserve the learned task behavior, and reports that the best variant, DPCC-C, reaches the goal with constraint satisfaction in 98% of episodes while taking fewer timesteps than the baselines.

Load-bearing premise

The guarantee depends on being able to keep sampling trajectories that satisfy the stricter, safety-margin constraints at every replanning step, which requires the robot to remain inside that stricter region; the proof only shows it remains inside the original, looser region, so that requirement is assumed rather than proven.

Editorial extensions

If this is right

  • An offline-trained diffusion policy can be deployed in environments with constraints never seen in training, simply by changing the projection sets at test time.
  • Because the projection enforces the dynamics model at every denoising step, the generated trajectories remain executable by the low-level controller, unlike constraints that are only projected once after sampling.
  • With a known upper bound on model mismatch, the constraint-tightening argument turns predicted feasibility into actual feasibility, which is why adding tightening raises the combined goal-and-constraint success rate from below 50% to about 98%.
  • The two selection criteria, temporal consistency and cumulative projection cost, reduce mode-switching and preserve goal-reaching performance, with the cumulative-cost criterion also reaching the goal fastest.
  • Each control action costs about 80 ms of computation, so DPCC is usable at moderate replanning rates on a workstation GPU but not at high control frequencies.

Reading between the lines

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

  • Because the projection is redefined at each replanning step, the same algorithm should extend directly to time-varying constraints such as moving obstacles; the paper notes this but does not test it, and a moving-obstacle benchmark would be a natural stress test.
  • The experiments solve non-convex projected sets with a nonlinear optimizer even though Theorem 1 assumes convexity, so the convexity assumption appears conservative; analyzing the non-convex projection's effect on the sampling distribution would either justify or refine the method.
  • The cumulative projection-cost selection criterion can be read as a proxy for staying close to the learned data distribution, hinting at a principled link between projection-guided sampling and likelihood-preserving guidance.
  • In practice the guarantee rests on knowing a valid mismatch bound $\gamma$; where that bound is uncertain, an online estimator or an outer safety filter would be needed to preserve the near-100% constraint satisfaction rate.
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 / 5 minor

Summary. The paper proposes DPCC, an inference-time algorithm that combines an offline-trained trajectory diffusion policy with receding-horizon control and model-based projections onto state, action, and dynamics constraint sets. During each denoising step, candidate trajectories are projected into a feasible set, and the constraints are tightened by a model-mismatch bound. Two batch-selection criteria (temporal consistency and cumulative projection cost) are introduced, and the method is evaluated in a simulated 2D robot manipulator task with novel circular and halfspace obstacles. The central claims are that DPCC generates constraint-satisfying, dynamically feasible, goal-reaching trajectories, and that the constraint-tightening mechanism provides a closed-loop guarantee against model mismatch.

Significance. If the formal claims can be made correct, DPCC is a practically useful contribution: it addresses an important problem (enforcing novel test-time constraints for offline diffusion policies without retraining), the idea of iterating model-based projections inside the backward diffusion process is elegant, and the experimental comparison against guidance, post-processing, and model-free projection baselines is informative. The paper ships code and reports computation times and hyperparameters, which strengthens reproducibility. The relative empirical conclusions are credible; the main weakness is that the theoretical guarantee is stronger than what is proven.

major comments (3)
  1. [Section 5.3, Theorem 2] The closed-loop guarantee is conditional on an unproven recursive-feasibility assumption. The theorem's premise states that at each timestep t one can sample a trajectory from p_theta(·|c, tilde Z_f), which requires the current state to lie in the tightened set tilde S_t, since the first predicted state st|t is constrained by tilde Z_f. The induction step, however, only proves st+1 in S_{t+1} from st in S_t; it never proves st+1 in tilde S_{t+1}, and Eq. (17) defines tilde S_{t+1} but does not specify the base case tilde S_0. Consequently, for a state in S_t \ tilde S_t the problem may be infeasible and the premise can fail. A recursive-feasibility or terminal-invariance argument is needed, or the word "guaranteed" in the contributions and abstract should be softened.
  2. [Section 6.1 and Appendix B] The deterministic constraint-satisfaction claim in Eq. (16) and Theorem 2 presupposes that the projection onto Z_f (resp. tilde Z_f) is computed exactly. In the experiments, however, the state constraint sets are explicitly non-convex (St = {As <= b, ||s-p||_2^2 >= r^2}) and the projections are solved with the local SLSQP solver. A local solution of a non-convex QCQP need not be the global projection, so the iterates tau^{k-1} in Algorithm 1 are not guaranteed to lie in tilde Z_f. This is not a fatal flaw for the empirical comparison, but it invalidates the formal guarantee in the experimental setting; the gap should be acknowledged and the theoretical statements qualified.
  3. [Section 5.2, Theorem 1 and Eq. (16)] Eq. (16) is not the sampling update derived in Theorem 1. Theorem 1 justifies the update tau^{k-1} ~ N(Pi_Zf(mu_theta), sigma_k^2 I), whereas Eq. (16) projects mu_theta + sigma_k epsilon_k after adding noise. These two operations produce different distributions, and the paper does not prove that iterating Eq. (16) approximates the conditional distribution p_theta(tau | O=1). Since the task-performance argument relies on preserving the learned trajectory distribution, this discrepancy needs either a formal justification or an explicit statement that Eq. (16) is a heuristic modification whose distributional properties are not guaranteed.
minor comments (5)
  1. [Section 6.1] The sentence defining the action constraints is duplicated: "The action constraints are defined as At = A, where A is the smallest bounding box containing all actions from the demonstration dataset (2)" appears twice in the same paragraph.
  2. [Section 5.2] There is a capitalization and wording error after Eq. (16): "By using (16), We denote the samples projected distribution as tau^0 ~ p_theta(·|c, Z_f)" should be "we denote" and the syntax should be corrected.
  3. [Section 5.3 and Algorithm 1] The notation for tightened feasible sets is inconsistent: Eq. (16) uses Z_f, while Algorithm 1 and Theorem 2 use tilde Z_f. The relation between the untightened and tightened projections should be made explicit at the point where the algorithm first introduces tilde Z_f.
  4. [Section 6.1] The model-mismatch bound gamma is estimated from 100 unconstrained rollouts of the same policy and then used to tighten the constraints. As reported, this is a heuristic estimate rather than a certified bound; Theorem 2 relies on the assumption ||w_t||_2 <= gamma, so the practical safety claim inherits the uncertainty of this estimate. The authors should state this limitation explicitly.
  5. [Section 7] The conclusion states that DPCC can handle time-varying constraints directly, but the experiments do not evaluate time-varying constraints and no formal argument is given. Either add an experiment or soften the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DPCC's constraint guarantee is a constructive projection plus a set-inclusion tightening argument, not a fitted quantity renamed as a prediction.

full rationale

The derivation chain is self-contained. Theorem 1 follows the standard classifier-guidance first-order expansion (Eqs. 13-15) and introduces no fitted parameter; Eq. (16) enforces τ^{k-1} ∈ Z_f by applying the projection after adding noise, so feasibility of each generated trajectory follows from the definition of Π_{Z_f} rather than from an empirical fit. Theorem 2 is a set-inclusion consequence of Minkowski tightening: if the sampled trajectory lies in \tilde Z_f, then s_{t+1} ∈ \tilde S_{t+1} ⊕ B_γ = (S_{t+1} ⊖ B_γ) ⊕ B_γ ⊆ S_{t+1}. The theorem's conclusion is thus derived from its stated sampling premise; it does not secretly insert the desired outcome into the assumptions. The estimate γ = 0.025 from 100 unconstrained rollouts is an empirical calibration of the disturbance bound, not a predicted success metric, so it is not circular, although it is not a certified bound. The only self-citation, Romer et al. (2024), is used to contrast the model-free baseline and is not load-bearing for DPCC's guarantees. One non-circular correctness gap should be flagged per the review rule: Theorem 2's proof does not establish recursive feasibility of the tightened projection; its induction shows st+1 ∈ S_{t+1} but not st+1 ∈ \tilde S_{t+1}, so it assumes rather than proves that a sample from pθ(·|c, \tilde Z_f) exists at the next replanning step. This is an omitted proof/assumption affecting the strength of the word 'guaranteed', not an equation-level circularity.

Assumptions & free parameters 2 free parameters · 6 assumptions · 1 invented entities

The central claim depends on several unproved or data-fitted ingredients: the bounded-mismatch dynamics model with gamma estimated from rollouts, exact computability of the projection Pi_Zf (violated by the nonconvex SLSQP experiments), and an unproven recursive-feasibility assumption that the current state remains in the tightened set at every replanning step. These are structural assumptions of the safety guarantee rather than new physical postulates.

free parameters (2)
  • gamma (model mismatch bound) = 0.025
    Estimated from 100 policy rollouts without constraints and used in the constraint tightening of Theorem 2. It is data-fitted rather than certified, so the safety guarantee inherits its uncertainty.
  • Guidance gradient weights (baseline) = not reported
    Selected via an ablation study for the cost-guidance baseline. This affects the fairness of the baseline comparison but is not a parameter of the DPCC algorithm itself.
assumptions (6)
  • domain assumption Dynamics model f is known and mismatch w_t is bounded by gamma (Eq. (1))
    Assumed in Problem 1 and Theorem 2. In experiments f is Euler integration of desired velocities and gamma is estimated from rollouts, not certified.
  • ad hoc to paper Z_f is nonempty and the projection Pi_Zf is computed exactly (Eq. (9))
    Required for the deterministic feasibility claim of Eq. (16). Experiments use nonconvex Z_f solved by SLSQP, so the projection is approximate.
  • ad hoc to paper At every replanning step the current state lies in the tightened set \tilde S_t
    Needed for sampling from \tilde Z_f in Theorem 2, but the proof only establishes s_t in S_t; recursive feasibility is unproven.
  • ad hoc to paper Z_f is closed and convex for Theorem 1
    Theorem 1 relies on convexity for the uniqueness of the projection and the gradient formula. The experiments violate convexity, so the theorem is only a heuristic justification there.
  • domain assumption The learned distribution p_theta approximates the demonstration trajectory distribution q(tau|c) (Eq. (6))
    Standard imitation-learning assumption; the paper validates it indirectly through task success metrics.
  • standard math Standard results: projection theorem for closed convex sets and Minkowski property (S minus B_gamma) plus B_gamma is contained in S
    Used in the proofs of Theorem 1 and Theorem 2 respectively.
invented entities (1)
  • Feasibility indicator O in the control-as-inference formulation (Eq. (10))
    purpose: Latent binary variable used to derive the constrained denoising update (Eq. (12)) via classifier guidance.
    Mathematical modeling device, not a physical quantity. It has no falsifiable external handle and serves only as a tool in the derivation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diffusion Predictive Control with Constraints." pith.science (2026). https://pith.science/paper/GKNOCBOL

@misc{pith2026241209342,
  author       = {Pith},
  title        = {Pith review of: Diffusion Predictive Control with Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GKNOCBOL}},
  note         = {Machine review of arXiv:2412.09342}
}
read the original abstract

Diffusion models have become popular for policy learning in robotics due to their ability to capture high-dimensional and multimodal distributions. However, diffusion policies are stochastic and typically trained offline, limiting their ability to handle unseen and dynamic conditions where novel constraints not represented in the training data must be satisfied. To overcome this limitation, we propose diffusion predictive control with constraints (DPCC), an algorithm for diffusion-based control with explicit state and action constraints that can deviate from those in the training data. DPCC incorporates model-based projections into the denoising process of a trained trajectory diffusion model and uses constraint tightening to account for model mismatch. This allows us to generate constraint-satisfying, dynamically feasible, and goal-reaching trajectories for predictive control. We show through simulations of a robot manipulator that DPCC outperforms existing methods in satisfying novel test-time constraints while maintaining performance on the learned control task.

Figures

Figures reproduced from arXiv: 2412.09342 by the authors.

Figure 1
Figure 1. Experiments: (a) Simulation environment, where the objective is to reach the green line with the end-effector without collisions. (b) Multimodal trajectory distribution in the training dataset. (c) Novel test-time constraints (blue). 6.1. Setup We conduct our experiments1 in the Avoiding simulation environment (Jia et al., 2024) shown in [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Impact of our constraint tightening method and the trajectory selection criterion (DPCC-R: random, DPCC-T: temporal consistency, DPCC-C: cumulative projection cost) on the success rates and the number of timesteps needed. None of these prior works use constraint tightening, but we evaluate their performance with and without our constraint tightening method to ensure the comparison is fair. We use four evaluation met… view at source ↗
Figure 3
Figure 3. Closed-loop trajectories with DPCC for different trajectory selection criteria and five training seeds, which are indicated by the trajectories’ colors. The tightened constraints are visualized in light blue. With our proposed trajectory selection criteria (DPCC-T and DPCC-C), we obtain a smoother behavior and shorter time to reach the goal. Constraint tightening Timesteps Goal Constraints & goal # Constraint violat… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Motion Planning with Model-Based Diffusion via Constraint Optimization and Adaptive Scheduling

    cs.RO 2026-07 conditional novelty 6.0 of 10

    MD-COAS unifies inexact augmented-Lagrangian soft constraints with convex-feasible-set hard projection and adaptively schedules them during model-based diffusion, improving safe and successful planning in non-convex e...

  2. Diffusion-Residual Model Predictive Steering Control for Vehicle Stabilization at the Limit of Handling under Model Uncertainty

    cs.RO 2026-07 conditional novelty 6.0 of 10

    Command-conditioned diffusion residual moments resize the MPC yaw reference and chance-tighten the handling envelope, cutting peak side-slip and recovering low-μ stability in simulation at 100 Hz.

  3. Joint Model-based Model-free Diffusion for Planning with Constraints

    cs.RO 2025-09 conditional novelty 6.0 of 10

    JM2D samples diffusion plans and safety-filter corrections jointly using a single importance-sampling-guided diffusion process, improving task success and reducing safety-filter interventions.

  4. Constraint-Aware Diffusion Guidance for Robotics: Real-Time Obstacle Avoidance for Autonomous Racing

    cs.RO 2025-05 conditional novelty 4.0 of 10

    A diffusion trajectory planner with a barrier-function guidance term and warm starting avoids obstacles in real time on a miniature race car, with 100% success in the reported trials.

Reference graph

Works this paper leans on

31 extracted references · 25 canonical work pages · cited by 4 Pith papers

  1. [1]

    Is C onditional G enerative M odeling all you need for D ecision M aking? In International Conference on Learning Representations (ICLR), 2023

    Anurag Ajay, Yilun Du, Abhi Gupta, Joshua B Tenenbaum, Tommi S Jaakkola, and Pulkit Agrawal. Is C onditional G enerative M odeling all you need for D ecision M aking? In International Conference on Learning Representations (ICLR), 2023

    work page 2023

  2. [2]

    Physics- I nformed D iffusion M odels

    Jan-Hendrik Bastek, WaiChing Sun, and Dennis M Kochmann. Physics- I nformed D iffusion M odels. arXiv preprint arXiv:2403.14404, 2024

    arXiv 2024

  3. [3]

    Nonlinear P rogramming: T heory and A lgorithms

    Mokhtar S Bazaraa, Hanif D Sherali, and Chitharanjan M Shetty. Nonlinear P rogramming: T heory and A lgorithms . John Wiley & Sons, 2006

    work page 2006

  4. [4]

    Predictive C ontrol for L inear and H ybrid S ystems

    Francesco Borrelli, Alberto Bemporad, and Manfred Morari. Predictive C ontrol for L inear and H ybrid S ystems . Cambridge University Press, 2017

    work page 2017

  5. [5]

    Motion P lanning D iffusion: L earning and P lanning of R obot M otions with D iffusion M odels

    Joao Carvalho, An T Le, Mark Baierl, Dorothea Koert, and Jan Peters. Motion P lanning D iffusion: L earning and P lanning of R obot M otions with D iffusion M odels. In Proc. of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 1916--1923, 2023

    work page 1916

  6. [6]

    Play F usion: S kill A cquisition via D iffusion from L anguage- A nnotated P lay

    Lili Chen, Shikhar Bahl, and Deepak Pathak. Play F usion: S kill A cquisition via D iffusion from L anguage- A nnotated P lay. In Conference on Robot Learning (CoRL), pages 2012--2029, 2023

    work page 2012

  7. [7]

    Diffusion P olicy: V isuomotor P olicy L earning via A ction D iffusion

    Cheng Chi, Siyuan Feng, Yilun Du, Zhenjia Xu, Eric Cousineau, Benjamin Burchfiel, and Shuran Song. Diffusion P olicy: V isuomotor P olicy L earning via A ction D iffusion. In Robotics: Science and Systems (RSS), 2023

    work page 2023

  8. [8]

    C onstrained S ynthesis with P rojected D iffusion M odels

    Jacob K Christopher, Stephen Baek, and Ferdinando Fioretto. C onstrained S ynthesis with P rojected D iffusion M odels. In Advances in Neural Information Processing Systems (NeurIPS), 2024

    work page 2024

Show all 31 references
  1. [9]

    Diffusion M odels B eat GAN s on I mage S ynthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion M odels B eat GAN s on I mage S ynthesis. Advances in Neural Information Processing Systems (NeurIPS), 34: 0 8780--8794, 2021

    2021

  2. [10]

    Aligning O ptimization T rajectories with D iffusion M odels for C onstrained D esign G eneration

    Giorgio Giannone, Akash Srivastava, Ole Winther, and Faez Ahmed. Aligning O ptimization T rajectories with D iffusion M odels for C onstrained D esign G eneration. Advances in Neural Information Processing Systems, 36: 0 51830--51861, 2023

    2023

  3. [11]

    Classifier- F ree D iffusion G uidance

    Jonathan Ho and Tim Salimans. Classifier- F ree D iffusion G uidance. arXiv preprint arXiv:2207.12598, 2022

    2022 arXiv

  4. [12]

    Denoising D iffusion P robabilistic M odels

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising D iffusion P robabilistic M odels. Advances in Neural Information Processing Systems, 33: 0 6840--6851, 2020

    2020

  5. [13]

    Diffuse L oco: R eal- T ime L egged L ocomotion C ontrol with D iffusion from O ffline D atasets

    Xiaoyu Huang, Yufeng Chi, Ruofeng Wang, Zhongyu Li, Xue Bin Peng, Sophia Shao, Borivoje Nikolic, and Koushil Sreenath. Diffuse L oco: R eal- T ime L egged L ocomotion C ontrol with D iffusion from O ffline D atasets. In Conference on Robot Learning (CoRL), 2024

    2024

  6. [14]

    Planning with D iffusion for F lexible B ehavior S ynthesis

    Michael Janner, Yilun Du, Joshua Tenenbaum, and Sergey Levine. Planning with D iffusion for F lexible B ehavior S ynthesis. In Proc. of the International Conference on Machine Learning, volume 162 of PMLR, pages 9902--9915, 2022

    2022

  7. [15]

    Towards D iverse B ehaviors: A B enchmark for I mitation L earning with H uman D emonstrations

    Xiaogang Jia, Denis Blessing, Xinkai Jiang, Moritz Reuss, Atalay Donat, Rudolf Lioutikov, and Gerhard Neumann. Towards D iverse B ehaviors: A B enchmark for I mitation L earning with H uman D emonstrations. In International Conference on Learning Representations (ICLR), 2024

    2024

  8. [16]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  9. [17]

    Cgd: C onstraint- G uided D iffusion P olicies for UAV T rajectory P lanning

    Kota Kondo, Andrea Tagliabue, Xiaoyi Cai, Claudius Tewari, Olivia Garcia, Marcos Espitia-Alvarez, and Jonathan P How. Cgd: C onstraint- G uided D iffusion P olicies for UAV T rajectory P lanning. arXiv preprint arXiv:2405.01758, 2024

    2024 arXiv

  10. [18]

    Algorithm 733: TOMP -- F ortran modules for optimal control calculations

    Dieter Kraft. Algorithm 733: TOMP -- F ortran modules for optimal control calculations. ACM Transactions on Mathematical Software (TOMS), 20 0 (3): 0 262--281, 1994

    1994

  11. [19]

    Diffusion models beat GAN s on topology optimization

    Fran c ois Maz \'e and Faez Ahmed. Diffusion models beat GAN s on topology optimization. In Proc. of the AAAI Conference on Artificial Intelligence, volume 37, pages 9108--9116, 2023

    2023

  12. [20]

    Improved D enoising D iffusion P robabilistic M odels

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved D enoising D iffusion P robabilistic M odels. In Proc. of the International Conference on Machine Learning (ICML), pages 8162--8171. PMLR, 2021

    2021

  13. [21]

    Imitating H uman B ehaviour with D iffusion M odels

    Tim Pearce, Tabish Rashid, Anssi Kanervisto, Dave Bignell, Mingfei Sun, Raluca Georgescu, Sergio Valcarcel Macua, Shan Zheng Tan, Ida Momennejad, Katja Hofmann, et al. Imitating H uman B ehaviour with D iffusion M odels. In International Conference on Learning Representations ...

    2023

  14. [22]

    Sampling C onstrained T rajectories U sing C omposable D iffusion M odels

    Thomas Power, Rana Soltani-Zarrin, Soshi Iba, and Dmitry Berenson. Sampling C onstrained T rajectories U sing C omposable D iffusion M odels. In IROS 2023 Workshop on Differentiable Probabilistic Robotics: Emerging Perspectives on Robot Learning, 2023

    2023

  15. [23]

    Model P redictive C ontrol: T heory, C omputation, and D esign , volume 2

    James Blake Rawlings, David Q Mayne, Moritz Diehl, et al. Model P redictive C ontrol: T heory, C omputation, and D esign , volume 2. Nob Hill Publishing Madison, 2017

    2017

  16. [24]

    Goal- C onditioned I mitation L earning using S core-based D iffusion P olicies

    Moritz Reuss, Maximilian Li, Xiaogang Jia, and Rudolf Lioutikov. Goal- C onditioned I mitation L earning using S core-based D iffusion P olicies. In Robotics: Science and Systems (RSS), 2023

    2023

  17. [25]

    Safe O ffline R einforcement L earning using T rajectory- L evel D iffusion M odels

    Ralf R \"o mer, Lukas Brunke, Martin Schuck, and Angela P Schoellig. Safe O ffline R einforcement L earning using T rajectory- L evel D iffusion M odels. In ICRA 2024 Workshop Back to the Future: Robot Learning Going Probabilistic, 2024

    2024

  18. [26]

    Deep U nsupervised L earning using N onequilibrium T hermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep U nsupervised L earning using N onequilibrium T hermodynamics. In Proc. of the International Conference on Machine Learning (ICML), pages 2256--2265. PMLR, 2015

    2015

  19. [27]

    Robot T rajectory O ptimization using A pproximate I nference

    Marc Toussaint. Robot T rajectory O ptimization using A pproximate I nference. In Proc. of the International Conference on Machine Learning (ICML), pages 1049--1056, 2009

    2009

  20. [28]

    Deep G enerative M odels in R obotics: A S urvey on L earning from M ultimodal D emonstrations

    Julen Urain, Ajay Mandlekar, Yilun Du, Mahi Shafiullah, Danfei Xu, Katerina Fragkiadaki, Georgia Chalvatzaki, and Jan Peters. Deep G enerative M odels in R obotics: A S urvey on L earning from M ultimodal D emonstrations. arXiv preprint arXiv:2408.04380, 2024

    2024 arXiv

  21. [29]

    Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, St \'e fan J

    Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, St \'e fan J. van der Walt , Matthew Brett, Joshua Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew R. J. Nels...

    2020

  22. [30]

    Diffusers: State-of-the-art diffusion models , 2024

    Patrick von Platen, Suraj Patil, Anton Lozhkov, Pedro Cuenca, Nathan Lambert, Kashif Rasul, Mishig Davaadorj, Dhruv Nair, Sayak Paul, Steven Liu, William Berman, Yiyi Xu, and Thomas Wolf. Diffusers: State-of-the-art diffusion models , 2024. URL https://github.com/huggingface/diffusers

    2024

  23. [31]

    Diffusion M odel P redictive C ontrol

    Guangyao Zhou, Sivaramakrishnan Swaminathan, Rajkumar Vasudeva Raju, J Swaroop Guntupalli, Wolfgang Lehrach, Joseph Ortiz, Antoine Dedieu, Miguel L \'a zaro-Gredilla, and Kevin Murphy. Diffusion M odel P redictive C ontrol. arXiv preprint arXiv:2410.05364, 2024

    arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.