Pith. sign in

REVIEW 3 major objections 6 minor 39 references

Planning-Query-Guided Model Generation for Model-Based Deformable Object Manipulation

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read A generative model learns query-specific model resolutions, doubling planning speed for deformable-object manipulation.

desk verdict A genuinely useful query-conditioned resolution-selection idea with a plausible speedup on a tree task, but the label-optimization details are too inconsistent to fully trust the mechanism yet. read the letter →

arxiv 2508.19199 v1 pith:VS5SQSPF submitted 2025-08-26 cs.RO cs.LG

classification cs.ROcs.LG
keywords deformableobjectmanipulationmodel-basedplanninggraphneuralnetworksdiffusionmodeladaptiveresolutionpredictivecontroltree
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 tries to establish that a learned, query-conditioned model generator can replace a fixed full-resolution dynamics model for model-based deformable-object manipulation, preserving closed-loop task performance while sharply reducing planning time. The central proposal is to predict, from the start and goal pointclouds of a planning query, which spatial regions of a deformable object should be modeled at high resolution and which can be collapsed to a single node in a graph neural network dynamics model. To train this predictor without running expensive closed-loop rollouts for every candidate resolution, the authors construct a dataset through a two-stage optimization: first minimize a dynamics-accuracy objective in the simplified graph space, then refine the resolution vector against actual MPC closed-loop cost within a tolerance. On a simulated tree-manipulation task, the method achieves a 2.1x speedup in planning time over full-resolution models at a small, quantified cost increase. If the approach generalizes, it offers a path from planning data to task-specific, computationally efficient models for new manipulation tasks.

What carries the argument

The load-bearing machinery is a diffusion-based model generator $p_\theta(\omega \mid z_{1-G})$ trained on a dataset of planning queries paired with optimized binary resolution vectors. The query is encoded as a single graph $z_{1-G}$ built from full-resolution encodings of start and goal states, with edges connecting corresponding particles; the diffusion model denoises binary resolution vectors conditioned on this graph. The other half of the machinery is the two-stage dataset-construction optimization: Eq. 4 initializes $\omega$ by minimizing the plan cost in the simplified graph space against the observed final state plus an $\ell^1$ penalty on high-resolution regions, and Algorithm 2 refines $\omega$ by running actual MPC and accepting simplifications whose closed-loop task cost stays within $\epsilon_{\mathrm{tol}}$ of the best cost seen. The GNN dynamics model itself, with complexity scaling quadratically in graph size, is what makes resolution selection consequential.

What would settle it

On a held-out set of planning queries, compute the correlation between the stage-1 objective (Eq. 4) and the closed-loop task cost achieved by MPC with the corresponding resolution vector; if the correlation is near zero or negative, the prior is not carrying the argument. Alternatively, run Algorithm 2 from a random initialization instead of the stage-1 prior and compare final resolutions and closed-loop costs: if random initialization finds substantially better or cheaper solutions, the prior is not a reliable starting point.

Watch

Extended reading notes

Core claim

The central claim is that region-specific model resolution can be learned as a function of the planning query: given a start state, a goal state, and a segmentation of the object pointcloud, a conditional diffusion model outputs a binary resolution vector $\omega$ that determines which segments are represented at high resolution and which are collapsed to a single vertex. The dataset for this mapping is generated by a chained optimization in which a dynamics-accuracy-constrained simplification (Eq. 4) produces an initialization, and a closed-loop task-performance-constrained optimization (Eq. 5) then simplifies further while keeping MPC cost within $\epsilon_{\mathrm{tol}}$ of the best observed cost. The experiments on a tree-manipulation task report that the learned generator matches the distribution of optimized resolutions across query classes and yields a 2.1x planning-time speedup over full-resolution models with a 0.006 increase in task cost, corresponding to roughly 1 cm extra average distance for moved particles.

Load-bearing premise

The first optimization stage assumes that a cost computed in the simplified graph space against the observed final state is a reliable proxy for how accurately the simplified model predicts motion; if that proxy is misleading, the learned resolution labels start from a biased initialization and may be suboptimal.

Editorial extensions

If this is right

  • If the method is correct, a model-based planner can use a query-specific simplified graph instead of a full-resolution graph, reducing planning time without retraining the planner or the dynamics model.
  • The two-stage optimization offers a recipe for building training data for task-informed model simplification without paying full closed-loop MPC cost for every candidate resolution.
  • The observed couplings between segments (for example, the top of the tree often staying high-resolution when a nearby branch moves) imply that the learned resolutions capture message-passing dependencies, not just geometric motion.
  • Because the reported speedup grows with graph size, applying the approach to larger pointclouds than the 1317-particle tree should yield larger planning-time savings at similar performance tolerance.
  • The fact that a query-independent mode baseline is both slower and worse than the query-guided generator supports the conclusion that conditioning on the start and goal states is doing the work.

Reading between the lines

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

  • The same query-conditioned resolution idea could be applied to other high-dimensional model classes, such as particle-based fluid or cloth simulators, by treating the segmentation as an additional input rather than a given.
  • A natural extension, flagged by the paper as future work, is to learn the regions themselves from raw pointclouds instead of relying on a provided segmentation; that would make the generator applicable to new object shapes without manual region definitions.
  • The closed-loop refinement stage could be made more sample-efficient by reusing the planner's own trajectory data across queries, potentially replacing the per-query MPC evaluations with a shared value model.
  • An untested implication is that the learned resolution distribution should transfer to unseen tree geometries if the GNN dynamics model generalizes across graph structures; if transfer fails, the pointcloud conditioning alone may be insufficient.
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 / 6 minor

Summary. The paper proposes a method for automatically generating spatially adaptive dynamics models for deformable-object manipulation. Given a planning query, defined by start and goal point clouds, a diffusion-based model generator outputs per-region resolution parameters for a graph neural network dynamics model used in MPPI planning. To obtain training labels, the authors design a two-stage optimization that first minimizes a dynamics-accuracy cost (Eq. 4) and then refines the resolution vector under a closed-loop task-performance constraint (Algorithm 2). The method is evaluated on a simulated tree-manipulation task, where it achieves a 2.1x reduction in planning time over a full-resolution model with a reported cost increase of 0.006. The paper also analyzes the distribution of predicted resolutions and compares against full, minimal, and mode baselines.

Significance. The central idea is timely and relevant: rather than hand-specifying model complexity, the method learns task-specific resolutions from planning-query data, with a practical two-stage procedure that limits the number of costly closed-loop evaluations. The reported speedup on a non-trivial tree-manipulation task is a concrete falsifiable result, and the use of a diffusion model to capture multimodal resolution distributions is appropriate. The paper explicitly provides architecture details and MPPI hyperparameters, which aid reproducibility. However, the validity of the central claim depends heavily on the correctness of the label-generation pipeline, and several presentation and specification issues currently weaken confidence in that pipeline.

major comments (3)
  1. [IV-D.1, Eq. (4)] The text states that BUILDGRAPH(sT,ω) "computes a rollout," but BUILDGRAPH is defined in Section IV-A as the function that encodes a state into a graph at resolution ω; it does not perform a rollout. As written, Eq. (4) minimizes cplan(BUILDGRAPH(sT,ω), zG) + wdyn|ω|1, which is the planning cost between the simplified encoding of the observed final state and the goal graph plus a sparsity penalty, not a prediction error. If the implementation follows Eq. (4), the first stage is not a dynamics-accuracy prior and the initialization for Algorithm 2 may be biased; if the implementation actually performs a rollout, the equation and the text must be corrected. Please clarify the intended objective and, if necessary, re-evaluate the impact on the learned resolutions.
  2. [IV-D.2, Algorithm 2] The termination condition in line 5, "while ω* unchanged ≥ Igrace iterations or Σω^t = 0 do," is ambiguous and appears to contain a misprint. It is unclear whether the loop continues while either condition holds or stops when either holds, and the expression "Σω^t" is not defined. This condition determines when the resolution-label search terminates and therefore directly affects the quality of the training data; please restate it as a precise Boolean expression.
  3. [IV-B, IV-D.2, V-A] The dataset size is reported inconsistently: Section IV-B says "We train on a dataset of 6075 samples holding out 10% for validation," the paragraph after Algorithm 2 says "dataset of 11K s1,sG,ω* tuples," and Section V-A says "5500 (s1,sG,ω*) tuples were in the dataset." These numbers must be reconciled, since the training-set size is important for assessing generalization claims.
minor comments (6)
  1. [IV-D, first paragraph] The sentence "This section outlines our approach for generating a resolution ω* which in order to construct a dataset mapping planning queries as (s1,sG) pairs to an optimized ω" is grammatically malformed and should be reworded.
  2. [V-C] The text references Figure 7, but the actual box plot is not present in the manuscript text I reviewed; the figure should be included to support the reported means and standard deviations.
  3. [IV-A] The sentence "The complexity of the graph, and thus the computation time for computing scales quadratically with the number of vertices and edges in z_t" is incomplete; clarify the intended statement.
  4. [IV-C] The chamfer-distance cost cplan has unclear notation: the bounds "i=1^P" and "j'<P" are not consistent, and the mask m is used without a clear definition in this context.
  5. [V-C] Please specify how the "mode" baseline is computed: whether it is the single most frequent full resolution vector in the training set or per-segment modes.
  6. [IV-D.2] The phrase "The optimization process (Algorithm. 2)" contains a typo; "Algorithm. 2" should be "Algorithm 2."

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central speedup claim is evaluated on held-out queries against independent baselines, and the only close call (Eq. 4 wording) is a correctness/proxy inconsistency rather than a definitional circle.

full rationale

I walked the derivation chain: dataset construction via chained optimization (Sec. IV-D, Eq. 4, Eq. 5, Algorithm 2), diffusion generator training (Sec. IV-B), and end-to-end evaluation (Sec. V-C). The central empirical claim, a 2.1x planning-time speedup at comparable task cost on 100 held-out test queries, is grounded in direct comparison against full-resolution, minimal-resolution, and query-independent mode baselines. The resolution labels used for training are optimized with the same dynamics model, MPPI planner, and SoftGym simulator in which the method is later evaluated; this is self-referential tuning, but it does not make the evaluation definitional. The test queries are held out, the mode baseline removes query conditioning, and the reported difference is an empirical outcome rather than a quantity forced by construction. No fitted parameter is renamed as a prediction, and no load-bearing premise is imported from a self-citation. The closest issue is in Sec. IV-D.1: the text after Eq. 4 claims that BUILDGRAPH(sT,ω) 'computes a rollout' and that cplan measures 'the distance between the predicted and observed trajectories,' but the equation actually evaluates a planning cost of the simplified final-state graph against zG, not a rollout prediction error. This is an internal inconsistency and a label-quality/proxy concern, not circularity, because the final labels are still refined by closed-loop task performance in Eq. 5 and the evaluation uses independent held-out queries. References to prior work by the authors, such as [17] and [28], are contextual and not used to justify the method's validity or to forbid alternatives. The paper's own stated limitation of a noisy relationship between closed-loop performance and resolution further indicates the authors do not treat the optimization as definitionally guaranteed. I therefore find no step where a claimed prediction reduces to its inputs by construction.

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

The method depends on a learned GNN dynamics model, a hand-chosen penalty wdyn, an unreported tolerance epsilon_tol, an unreported movement threshold delta, and several task-specific assumptions (segmentation, full observability, simulator fidelity). No new physical entities are introduced.

free parameters (6)
  • wdyn = 0.005
    Hand-chosen penalty on the number of high-resolution regions in the dynamics-accuracy stage (Eq. 4); balances simplification against the task-cost term. Not derived from data.
  • epsilon_tol = not reported
    Performance tolerance in the closed-loop optimization (Eq. 5, Algorithm 2) that defines how much task-cost increase is allowed during simplification. No value is given, so the label-generation threshold is unspecified.
  • delta = not reported
    Movement threshold in the task cost mask (Eq. 1); determines which particles count as 'moved' and therefore which regions matter for the cost. No value is given.
  • CMA-ES initialization = mean 0.7, population 20
    Optimizer settings for the first-stage simplification (Section IV-D.1); chosen by hand.
  • full-resolution graph vertex count = 227
    The 'full' model still downsamples 1317 particles to 227 graph vertices (Section V-C), so the reported speedup is relative to a reduced model, not the raw particle simulation. This design choice affects the baseline and the observed speedup.
  • number of regions K = 8
    The tree point cloud is segmented into 8 regions (Fig. 2). The resolution vector length and the optimization space depend on this segmentation, which is assumed given.
assumptions (6)
  • domain assumption The GNN dynamics model f_hat trained on random omega generalizes to arbitrary omega at test time.
    The model is trained on 600K interactions with randomly selected omega (Section IV-A); at test time it must predict dynamics for omega sampled by the generator.
  • domain assumption A semantic segmentation of the object point cloud into K regions is available.
    Stated in Section III: 'we assume access to a segmentation of the point cloud in st'.
  • domain assumption Full state observability.
    Stated in Section III: 'We assume full state observability.'
  • domain assumption The SoftGym/Flex simulated tree is a valid proxy for real tree dynamics.
    All experiments are in a custom SoftGym environment (Section V-A); the contribution is evaluated only in simulation.
  • ad hoc to paper The two-stage optimization (Eq 4 then Algorithm 2) finds omega* that is near-optimal for closed-loop performance.
    Stage 2 performs greedy coordinate descent by randomly zeroing K indices; no guarantee of optimality, and the validity of the stage-1 prior depends on Eq 4's relationship to true dynamics accuracy.
  • domain assumption The cost function in Eq 2 accurately reflects task success.
    Task performance is defined by average distance of particles that moved more than delta; this choice is task-specific.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Planning-Query-Guided Model Generation for Model-Based Deformable Object Manipulation." pith.science (2026). https://pith.science/paper/VS5SQSPF

@misc{pith2026250819199,
  author       = {Pith},
  title        = {Pith review of: Planning-Query-Guided Model Generation for Model-Based Deformable Object Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VS5SQSPF}},
  note         = {Machine review of arXiv:2508.19199}
}
read the original abstract

Efficient planning in high-dimensional spaces, such as those involving deformable objects, requires computationally tractable yet sufficiently expressive dynamics models. This paper introduces a method that automatically generates task-specific, spatially adaptive dynamics models by learning which regions of the object require high-resolution modeling to achieve good task performance for a given planning query. Task performance depends on the complex interplay between the dynamics model, world dynamics, control, and task requirements. Our proposed diffusion-based model generator predicts per-region model resolutions based on start and goal pointclouds that define the planning query. To efficiently collect the data for learning this mapping, a two-stage process optimizes resolution using predictive dynamics as a prior before directly optimizing using closed-loop performance. On a tree-manipulation task, our method doubles planning speed with only a small decrease in task performance over using a full-resolution model. This approach informs a path towards using previous planning and control data to generate computationally efficient yet sufficiently expressive dynamics models for new tasks.

Figures

Figures reproduced from arXiv: 2508.19199 by the authors.

Figure 1
Figure 1. Given a planning query and a segmented pointcloud, the model [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Examples of z constructed from BUILDGRAPH from ⃗ω and sobj. a) shows the input pointcloud with segments labeled, and b-d show meshes with segments simplified with different ⃗ω MPC start state: s1 ω⃗ BUILDGRAPH(s1 ,ω⃗) BUILDGRAPH(sG ,ω⃗) simplified problem (z1-G) planner goal state: sG Model generator [1,0,1,0,0] ω⃗example BUILDGRAPH(s1 ,ω⃗example) Multi-resolution GNN dynamics model: f̂ world observed s1:T +1 replan… view at source ↗
Figure 3
Figure 3. Overview of how a model generator is used. Starting at the left, start and goal states with the object represented as pointclouds are inputted to the [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Data flow diagram of chained optimization algorithm used to generate [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Frames showing an example of the free-floating capsule end effector [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Distribution of selected model resolutions for three classes of [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Mean (standard deviation) for final cost and average planning time [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 33 canonical work pages

  1. [1]

    Adaptigraph: Material-adaptive graph-based neural dynamics for robotic manipulation,

    K. Zhang, B. Li, K. Hauser, and Y . Li, “Adaptigraph: Material-adaptive graph-based neural dynamics for robotic manipulation,” arXiv preprint arXiv:2407.07889, 2024

  2. [2]

    Differentiable particles for general-purpose deformable object manipulation,

    S. Chen, Y . Xu, C. Yu, L. Li, and D. Hsu, “Differentiable particles for general-purpose deformable object manipulation,” arXiv preprint arXiv:2405.01044, 2024

  3. [3]

    Mesh-based dynamics with occlusion reasoning for cloth manipulation,

    Z. Huang, X. Lin, and D. Held, “Mesh-based dynamics with occlusion reasoning for cloth manipulation,” arXiv preprint arXiv:2206.02881 , 2022

  4. [4]

    Learning rigid dynamics with face interaction graph networks,

    K. R. Allen, Y . Rubanova, T. Lopez-Guevara, W. Whitney, A. Sanchez- Gonzalez, P. Battaglia, and T. Pfaff, “Learning rigid dynamics with face interaction graph networks,” 2022

  5. [5]

    Learning to simulate complex physics with graph networks,

    A. Sanchez-Gonzalez, J. Godwin, T. Pfaff, R. Ying, J. Leskovec, and P. Battaglia, “Learning to simulate complex physics with graph networks,” in International conference on machine learning, pp. 8459– 8468, PMLR, 2020

  6. [6]

    Learn- ing mesh-based simulation with graph networks,

    T. Pfaff, M. Fortunato, A. Sanchez-Gonzalez, and P. Battaglia, “Learn- ing mesh-based simulation with graph networks,” in International Conference on Learning Representations , 2020

  7. [7]

    Multiscale meshgraphnets,

    M. Fortunato, T. Pfaff, P. Wirnsberger, A. Pritzel, and P. Battaglia, “Multiscale meshgraphnets,” arXiv preprint arXiv:2210.00612 , 2022

  8. [8]

    Gfn: A graph feed- forward network for resolution-invariant reduced operator learning in multifidelity applications,

    O. M. Morrison, F. Pichi, and J. S. Hesthaven, “Gfn: A graph feed- forward network for resolution-invariant reduced operator learning in multifidelity applications,” Computer Methods in Applied Mechanics and Engineering, vol. 432, p. 117458, 2024

Show all 39 references
  1. [9]

    The surprising effectiveness of linear models for visual foresight in object pile manipulation,

    H. T. Suh and R. Tedrake, “The surprising effectiveness of linear models for visual foresight in object pile manipulation,” in Algorithmic Foundations of Robotics XIV: Proceedings of the Fourteenth Workshop on the Algorithmic Foundations of Robotics 14, pp. 347–363, Springer, 2021

  2. [10]

    Linear vs nonlinear mpc for trajectory tracking applied to rotary wing micro aerial vehicles,

    M. Kamel, M. Burri, and R. Siegwart, “Linear vs nonlinear mpc for trajectory tracking applied to rotary wing micro aerial vehicles,” IFAC- PapersOnLine, vol. 50, no. 1, pp. 3463–3469, 2017

  3. [11]

    Plan-time multi-model switching for motion planning,

    B. Styler and R. Simmons, “Plan-time multi-model switching for motion planning,” in Proceedings of the International Conference on Automated Planning and Scheduling , vol. 27, 2017

  4. [12]

    Motion planning for an underwater mobile manipulator by exploiting loose coupling,

    D. Youakim, A. Dornbush, M. Likhachev, and P. Ridao, “Motion planning for an underwater mobile manipulator by exploiting loose coupling,” in 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 7164–7171, IEEE, 2018

  5. [13]

    Planning with selective physics- based simulation for manipulation among movable objects,

    M. S. Saleem and M. Likhachev, “Planning with selective physics- based simulation for manipulation among movable objects,” in 2020 IEEE International Conference on Robotics and Automation (ICRA) , pp. 6752–6758, IEEE, 2020

  6. [14]

    Learning state-specific action masks for reinforcement learning,

    Z. Wang, X. Li, L. Sun, H. Zhang, H. Liu, and J. Wang, “Learning state-specific action masks for reinforcement learning,” Algorithms, vol. 17, no. 2, p. 60, 2024

  7. [15]

    Online state vector reduc- tion during model predictive control with gradient-based trajectory optimisation,

    D. Russell, R. Papallas, and M. Dogar, “Online state vector reduc- tion during model predictive control with gradient-based trajectory optimisation,” in Springer Proceedings in Advanced Robotics (SPAR) , Springer, 2024

  8. [16]

    An efficient approach to closed-loop shape control of deformable objects using finite element models,

    A. Koessler, N. R. Filella, B.-C. Bouzgarrou, L. Lequi `evre, and J.- A. C. Ramon, “An efficient approach to closed-loop shape control of deformable objects using finite element models,” in 2021 IEEE Inter- national conference on robotics and automation (ICRA) , pp. 1637– 1643...

  9. [17]

    Estimating model utility for de- formable object manipulation using multiarmed bandit methods,

    D. Mcconachie and D. Berenson, “Estimating model utility for de- formable object manipulation using multiarmed bandit methods,”IEEE Transactions on Automation Science and Engineering , vol. 15, no. 3, pp. 967–979, 2018

  10. [18]

    Dynamic- resolution model learning for object pile manipulation,

    Y . Wang, Y . Li, K. Driggs-Campbell, L. Fei-Fei, and J. Wu, “Dynamic- resolution model learning for object pile manipulation,” arXiv preprint arXiv:2306.16700, 2023

  11. [19]

    Neural field dynamics model for granular object piles manipulation,

    S. Xue, S. Cheng, P. Kachana, and D. Xu, “Neural field dynamics model for granular object piles manipulation,” in Conference on Robot Learning, pp. 2821–2837, PMLR, 2023

  12. [20]

    Goal-conditioned action space reduction for deformable object manipulation,

    S. Wang, R. Papallas, M. Leouctti, and M. Dogar, “Goal-conditioned action space reduction for deformable object manipulation,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) , pp. 3623–3630, IEEE, 2023

  13. [21]

    A comparison of mesh simplification algorithms,

    P. Cignoni, C. Montani, and R. Scopigno, “A comparison of mesh simplification algorithms,” Computers & Graphics , vol. 22, no. 1, pp. 37–54, 1998

  14. [22]

    Fast fluid simulation via dynamic multi-scale gridding,

    J. Liu, Y . Chen, B. Ni, W. Ren, Z. Yu, and X. Huang, “Fast fluid simulation via dynamic multi-scale gridding,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 37, pp. 1675–1682, 2023

  15. [23]

    Learn- ing controllable adaptive simulation for multi-resolution physics,

    T. Wu, T. Maruyama, Q. Zhao, G. Wetzstein, and J. Leskovec, “Learn- ing controllable adaptive simulation for multi-resolution physics,” in International Conference on Learning Representations (ICLR) , NSF- PAR, 2023

  16. [24]

    Efficient learning of mesh- based physical simulation with bi-stride multi-scale graph neural network,

    Y . Cao, M. Chai, M. Li, and C. Jiang, “Efficient learning of mesh- based physical simulation with bi-stride multi-scale graph neural network,” in International conference on machine learning , pp. 3541– 3558, PMLR, 2023

  17. [25]

    Eagle: Large-scale learning of turbulent fluid dynamics with mesh transformers,

    S. Janny, A. Beneteau, M. Nadri, J. Digne, N. Thome, and C. Wolf, “Eagle: Large-scale learning of turbulent fluid dynamics with mesh transformers,” in International Conference on Learning Representa- tions, 2023

  18. [26]

    Multiscale graph neural networks with adaptive mesh refinement for accelerating mesh-based simulations,

    R. Perera and V . Agrawal, “Multiscale graph neural networks with adaptive mesh refinement for accelerating mesh-based simulations,” Computer Methods in Applied Mechanics and Engineering , vol. 429, p. 117152, 2024

  19. [27]

    Hood: Hierarchical graphs for generalized modelling of clothing dynamics,

    A. Grigorev, M. J. Black, and O. Hilliges, “Hood: Hierarchical graphs for generalized modelling of clothing dynamics,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 16965–16974, 2023

  20. [28]

    Towards robotic tree manipulation: Leveraging graph representations,

    C. H. Kim, M. Lee, O. Kroemer, and G. Kantor, “Towards robotic tree manipulation: Leveraging graph representations,” in 2024 IEEE Inter- national Conference on Robotics and Automation (ICRA) , pp. 11884– 11890, IEEE, 2024

  21. [29]

    Gentle manipulation of tree branches: A contact-aware policy learn- ing approach,

    J. Jacob, S. Cai, P. V . K. Borges, T. Bandyopadhyay, and F. Ramos, “Gentle manipulation of tree branches: A contact-aware policy learn- ing approach,” in 8th Annual Conference on Robot Learning , PMLR, 2024

  22. [30]

    Learning to simulate tree-branch dynamics for manipulation,

    J. Jacob, T. Bandyopadhyay, J. Williams, P. Borges, and F. Ramos, “Learning to simulate tree-branch dynamics for manipulation,” IEEE Robotics and Automation Letters , 2024

  23. [31]

    3d graph neural networks for rgbd semantic segmentation,

    X. Qi, R. Liao, J. Jia, S. Fidler, and R. Urtasun, “3d graph neural networks for rgbd semantic segmentation,” in Proceedings of the IEEE international conference on computer vision , pp. 5199–5208, 2017

  24. [32]

    Point-gnn: Graph neural network for 3d object detection in a point cloud,

    W. Shi and R. Rajkumar, “Point-gnn: Graph neural network for 3d object detection in a point cloud,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. 1711– 1719, 2020

  25. [33]

    Learning particle dynamics for manipulating rigid bodies, deformable objects, and fluids,

    Y . Li, J. Wu, R. Tedrake, J. B. Tenenbaum, and A. Torralba, “Learning particle dynamics for manipulating rigid bodies, deformable objects, and fluids,” in International Conference on Learning Representations , 2018

  26. [34]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in neural information processing systems , vol. 33, pp. 6840–6851, 2020

  27. [35]

    A point set generation network for 3d object reconstruction from a single image,

    H. Fan, H. Su, and L. J. Guibas, “A point set generation network for 3d object reconstruction from a single image,” in Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 605–613, 2017

  28. [36]

    Sonicboom: Contact localization using array of microphones,

    M. Lee, U. Yoo, J. Oh, J. Ichnowski, G. Kantor, and O. Kroemer, “Sonicboom: Contact localization using array of microphones,” arXiv preprint arXiv:2412.09878, 2024

  29. [37]

    Softgym: Benchmarking deep reinforcement learning for deformable object manipulation,

    X. Lin, Y . Wang, J. Olkin, and D. Held, “Softgym: Benchmarking deep reinforcement learning for deformable object manipulation,” in Conference on Robot Learning , pp. 432–448, PMLR, 2021

  30. [38]

    The grove 3d,

    The Grove, “The grove 3d,” 2024. Accessed: 2024-04

  31. [39]

    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), pp. 1433–1440, IEEE, 2016

Pith tools

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