pith. sign in

arxiv: 2511.01770 · v2 · submitted 2025-11-03 · 💻 cs.RO

Lightweight Learning from Actuation-Space Demonstrations via Flow Matching for Whole-Body Soft Robotic Grasping

Liudi Yang , Yang Bai , Yuhao Wang , Ibrahim Alsarraj , Gitta Kutyniok , Zhanchi Wang , Ke Wu This is my paper

Pith reviewed 2026-05-18 01:23 UTC · model grok-4.3

classification 💻 cs.RO
keywords soft roboticsrobotic graspingflow matchingimitation learningactuation spacewhole-body controlgrasping under uncertainty
0
0 comments X

The pith

A flow matching model trained on 30 actuation demonstrations enables a soft robot to grasp objects at 97.5 percent success across its full workspace.

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

Soft robots possess passive flexibility that helps them manage uncertain contacts during grasping, yet most control methods still demand heavy sensing and feedback loops. This paper shows that a lightweight framework can learn effective control policies directly in actuation space by applying a flow matching model to a small set of deterministic demonstrations. Training on only 30 examples that cover less than 8 percent of the reachable workspace produces a policy that succeeds in 97.5 percent of trials over the entire space. The same policy also handles object size changes of plus or minus 33 percent and remains stable when execution time is scaled from 20 to 200 percent of the original duration. A reader would care because the result suggests that the robot's own mechanical compliance can replace much of the computational burden usually placed on central controllers.

Core claim

The authors establish that a rectified flow model, trained solely on deterministic actuation-space demonstrations, infers the distributional control representations needed for whole-body soft robotic grasping. With only 30 such demonstrations covering less than 8 percent of the workspace, the resulting policy achieves 97.5 percent grasp success across the full workspace, generalizes to grasped-object size variations of plus or minus 33 percent, and maintains performance when execution time is scaled between 20 and 200 percent of nominal speed. The method operates without dense sensing or closed-loop feedback by converting the soft body's passive redundant degrees of freedom and flexibility直接

What carries the argument

Rectified Flow model that converts deterministic actuation-space demonstrations into distributional control policies for whole-body soft robot grasping

If this is right

  • Grasp success remains high even though training data covers less than 8 percent of the reachable workspace.
  • The policy adapts to grasped objects that are up to 33 percent larger or smaller without retraining.
  • Performance stays stable when the robot executes the same motions at speeds ranging from 20 to 200 percent of the training speed.
  • The approach reduces the need for dense sensing and continuous feedback by relying on the soft body's inherent compliance.

Where Pith is reading between the lines

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

  • The same actuation-space flow matching technique could be applied to other contact-rich soft robot tasks such as in-hand manipulation or locomotion.
  • Collecting a minimal set of demonstrations in actuation space might allow quick deployment of soft robots to new workspaces with little additional data.
  • Treating the robot's mechanical properties as the primary source of robustness could shift design priorities away from complex sensing hardware toward simpler learning pipelines.

Load-bearing premise

Deterministic demonstrations from a tiny fraction of the workspace suffice for the flow matching model to infer the full range of control distributions required for robust grasping under uncertainty without dense sensing or closed-loop feedback.

What would settle it

Running the learned policy on objects whose sizes fall outside the plus or minus 33 percent range and recording whether success rate drops sharply below 80 percent would directly test whether the claimed generalization holds.

Figures

Figures reproduced from arXiv: 2511.01770 by Gitta Kutyniok, Ibrahim Alsarraj, Ke Wu, Liudi Yang, Yang Bai, Yuhao Wang, Zhanchi Wang.

Figure 1
Figure 1. Figure 1: Learning from Actuation-Space Demonstration for Grasping. A. SpiRob. B. Distinction in LfD schemes between rigid and soft robots. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the proposed framework. A. Overview of the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example of an expert grasping demo in the simulation. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Workspace and training region configuration for data generation. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: Experimental results of workspace generalization from sparse demonstrations and geometric adaptability to object size variations [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distributional sampling of control sequences. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Whole-body grasping of a SpiRob in uncertain environments. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
read the original abstract

Robotic grasping under uncertainty remains a fundamental challenge due to its uncertain and contact-rich nature. Traditional rigid robotic hands, with limited degrees of freedom and compliance, rely on complex model-based and heavy feedback controllers to manage such interactions. Soft robots, by contrast, exhibit embodied mechanical intelligence: their underactuated structures and passive flexibility of their whole body, naturally accommodate uncertain contacts and enable adaptive behaviors. To harness this capability, we propose a lightweight actuation-space learning framework that infers distributional control representations for whole-body soft robotic grasping, directly from deterministic demonstrations using a flow matching model (Rectified Flow),without requiring dense sensing or heavy control loops. Using only 30 demonstrations (less than 8% of the reachable workspace), the learned policy achieves a 97.5% grasp success rate across the whole workspace, generalizes to grasped-object size variations of +-33%, and maintains stable performance when the robot's dynamic response is directly adjusted by scaling the execution time from 20% to 200%. These results demonstrate that actuation-space learning, by leveraging its passive redundant DOFs and flexibility, converts the body's mechanics into functional control intelligence and substantially reduces the burden on central controllers for this uncertain-rich task.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a lightweight actuation-space learning framework for whole-body soft robotic grasping. It uses a Rectified Flow (flow matching) model trained directly on 30 deterministic demonstrations covering less than 8% of the reachable workspace. The central claims are that the resulting open-loop policy achieves a 97.5% grasp success rate across the entire workspace, generalizes to object-size variations of ±33%, and remains stable under execution-time scaling from 20% to 200% of nominal, all without dense sensing or closed-loop feedback, by exploiting the robot's passive compliance and redundant DOFs.

Significance. If the reported generalization and robustness hold under rigorous testing, the result would be significant for soft robotics and imitation learning. It would demonstrate that sparse actuation-space data combined with embodied mechanical intelligence can yield distributional control policies for contact-rich tasks, substantially lowering data and sensing requirements. The approach aligns with trends in generative modeling for robotics but would need to clearly separate learned policy effects from passive hardware properties to be fully convincing.

major comments (3)
  1. [Abstract / Experiments] Abstract and Experiments section: The quantitative claims (97.5% success across the full workspace, ±33% size generalization, and 20%-200% timing robustness) are presented without any description of the evaluation protocol, number of trials, workspace sampling strategy, statistical measures, or failure cases. This prevents assessment of whether the Rectified Flow model truly infers robust distributional behaviors or merely interpolates the 30 deterministic trajectories.
  2. [Method] Method section: The framework learns from deterministic actuation-space trajectories yet claims to produce policies robust to unmodeled contact uncertainties without feedback. No details are given on how the flow-matching vector field captures distributional contact-rich behaviors, nor on any regularization or augmentation that would enable extrapolation beyond the demonstrated <8% workspace region.
  3. [Experiments] Experiments section: The manuscript attributes robustness to the combination of learned policy and passive compliance but provides no ablation or quantitative separation of these contributions. Without such analysis, it is impossible to determine whether the high success rates would persist if the mechanical compliance were reduced or if the policy were transferred to a different soft robot.
minor comments (2)
  1. [Abstract] The abstract would benefit from a short description of the specific soft robot platform (number of actuators, material properties) to contextualize the 'whole-body' aspect for readers outside soft robotics.
  2. [Method] Notation for the flow-matching objective and the mapping from learned vector field to actuation commands should be introduced more explicitly, perhaps with a simple equation in the method section.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We sincerely thank the referee for the constructive and detailed feedback on our manuscript. The comments have identified key areas where additional clarity and rigor are needed. We have revised the manuscript to incorporate more detailed descriptions of the evaluation protocol, expanded explanations in the Method section, and added analysis to better separate the contributions of the learned policy and passive compliance. Our point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract / Experiments] Abstract and Experiments section: The quantitative claims (97.5% success across the full workspace, ±33% size generalization, and 20%-200% timing robustness) are presented without any description of the evaluation protocol, number of trials, workspace sampling strategy, statistical measures, or failure cases. This prevents assessment of whether the Rectified Flow model truly infers robust distributional behaviors or merely interpolates the 30 deterministic trajectories.

    Authors: We agree that the original manuscript did not provide sufficient details on the evaluation protocol, which limits the ability to fully assess the results. In the revised version, we have added a new 'Evaluation Protocol' subsection in the Experiments section. This includes the total number of trials (500 trials across conditions with 20 repetitions per sampled configuration), the workspace sampling strategy (uniform discretization of the reachable workspace into 25 regions with random perturbations), statistical reporting (mean success rate of 97.5% ± 1.8% standard deviation), and a summary of failure cases (primarily occurring at workspace boundaries with sparse demonstration coverage, accounting for the 2.5% failure rate). These additions clarify that the observed performance reflects generalization enabled by the flow model's generative sampling rather than pure interpolation of the 30 trajectories. revision: yes

  2. Referee: [Method] Method section: The framework learns from deterministic actuation-space trajectories yet claims to produce policies robust to unmodeled contact uncertainties without feedback. No details are given on how the flow-matching vector field captures distributional contact-rich behaviors, nor on any regularization or augmentation that would enable extrapolation beyond the demonstrated <8% workspace region.

    Authors: We appreciate this observation and have revised the Method section accordingly. The Rectified Flow model learns a continuous vector field that defines probability paths from a base noise distribution to the distribution of the demonstrated actuation trajectories. At inference time, the generative sampling process introduces controlled variations around the deterministic demonstrations, which, when executed on the compliant robot, accommodate unmodeled contacts without feedback. We have added details on the training procedure, including implicit regularization from the flow-matching objective (encouraging straight trajectories) and data augmentation via small temporal shifts and actuation noise to support extrapolation. This enables the policy to cover the full workspace by leveraging the smoothness of the learned vector field. We note that explicit contact modeling is absent, and robustness emerges from the interplay with the robot's passive properties. revision: yes

  3. Referee: [Experiments] Experiments section: The manuscript attributes robustness to the combination of learned policy and passive compliance but provides no ablation or quantitative separation of these contributions. Without such analysis, it is impossible to determine whether the high success rates would persist if the mechanical compliance were reduced or if the policy were transferred to a different soft robot.

    Authors: This is a fair critique. We have added an 'Analysis of Contributions' subsection to the Experiments section. Because the soft robot's compliance is an inherent hardware property, a direct physical ablation is not feasible without redesigning the system. Instead, we include a simulation-based comparison using a reduced-compliance model, showing success rates dropping to approximately 68% without compliance effects. We also quantify relative contributions based on trajectory deviation measurements during real experiments (policy providing nominal sequences accounting for the majority of performance, with compliance handling residual uncertainties). For transferability, we discuss that the actuation-space formulation is modular and could be adapted to other soft robots with similar redundancy. A dedicated limitations paragraph has been added to address these points openly. revision: partial

Circularity Check

0 steps flagged

No significant circularity: standard flow matching trained on external demonstrations with empirical validation

full rationale

The paper applies a standard Rectified Flow model to learn from 30 deterministic actuation-space demonstration trajectories collected from a small fraction of the workspace. Reported performance metrics (97.5% success rate, generalization to object size and timing variations) are obtained via physical robot experiments rather than by algebraic reduction to fitted parameters or self-referential definitions within the paper. No equations, self-citations, or ansatzes are presented that would make the central claims equivalent to the inputs by construction. The derivation chain relies on an external generative modeling technique and independent experimental evaluation, rendering the result self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the assumption that limited deterministic demonstrations suffice to capture the control distribution required for the task; no free parameters or new entities are explicitly introduced beyond the standard flow-matching generative process.

axioms (1)
  • domain assumption Deterministic demonstrations from a small workspace subset contain the distributional information needed for successful grasping under uncertainty.
    Invoked in the description of the lightweight learning framework that infers control representations directly from demonstrations.

pith-pipeline@v0.9.0 · 5766 in / 1215 out tokens · 40062 ms · 2026-05-18T01:23:11.548269+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages · 5 internal anchors

  1. [1]

    Siciliano, O

    B. Siciliano, O. Khatib, and T. Kröger,Springer handbook of robotics. Springer, 2008, vol. 200

    work page 2008

  2. [2]

    Toward robotic manipulation,

    M. T. Mason, “Toward robotic manipulation,”Annual Review of Control, Robotics, and Autonomous Systems, vol. 1, no. 1, pp. 1–28, 2018

    work page 2018

  3. [3]

    A concise guide to modelling the physics of embodied intelligence in soft robotics,

    G. Mengaldo, F. Renda, S. L. Brunton, M. Bächer, M. Calisti, C. Duriez, G. S. Chirikjian, and C. Laschi, “A concise guide to modelling the physics of embodied intelligence in soft robotics,” Nature Reviews Physics, vol. 4, no. 9, pp. 595–610, 2022

    work page 2022

  4. [4]

    Pfeifer and J

    R. Pfeifer and J. Bongard,How the body shapes the way we think: a new view of intelligence. MIT press, 2006

    work page 2006

  5. [5]

    Model-based control of soft robots: A survey of the state of the art and open challenges,

    C. Della Santina, C. Duriez, and D. Rus, “Model-based control of soft robots: A survey of the state of the art and open challenges,”IEEE Control Systems Magazine, vol. 43, no. 3, pp. 30–65, 2023

    work page 2023

  6. [6]

    Exploiting frictional effects to reproduce octopus-like reaching movements with a cable-driven spiral robot,

    Z. Wang and N. M. Freris, “Exploiting frictional effects to reproduce octopus-like reaching movements with a cable-driven spiral robot,” in 2024 IEEE 7th International Conference on Soft Robotics (RoboSoft). IEEE, 2024, pp. 537–542

    work page 2024

  7. [7]

    A hybrid hinge-beam continuum robot with passive safety capping for real-time fatigue awareness,

    T. Chen, Z. Sun, Y . Sun, Y . Wang, D. Song, and K. Wu, “A hybrid hinge-beam continuum robot with passive safety capping for real-time fatigue awareness,”arXiv preprint arXiv:2509.09404, 2025

    work page arXiv 2025

  8. [8]

    Spirobs: Logarithmic spiral- shaped robots for versatile grasping across scales,

    Z. Wang, N. M. Freris, and X. Wei, “Spirobs: Logarithmic spiral- shaped robots for versatile grasping across scales,”Device, vol. 3, no. 4, 2025

    work page 2025

  9. [9]

    Data-driven methods ap- plied to soft robot modeling and control: A review,

    Z. Chen, F. Renda, A. Le Gall, L. Mocellin, M. Bernabei, T. Dangel, G. Ciuti, M. Cianchetti, and C. Stefanini, “Data-driven methods ap- plied to soft robot modeling and control: A review,”IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 2241–2256, 2024

    work page 2024

  10. [10]

    Theory and applications of hyper-redundant robotic manipulators,

    G. S. Chirikjian, “Theory and applications of hyper-redundant robotic manipulators,” Ph.D. dissertation, California Institute of Technology, 1992

    work page 1992

  11. [11]

    Elastic stability of cosserat rods and parallel continuum robots,

    J. Till and D. C. Rucker, “Elastic stability of cosserat rods and parallel continuum robots,”IEEE Transactions on Robotics, vol. 33, no. 3, pp. 718–733, 2017

    work page 2017

  12. [12]

    Cosserat rod modeling of continuum robots from new- tonian and lagrangian perspectives,

    M. Tummers, V . Lebastard, F. Boyer, J. Troccaz, B. Rosa, and M. T. Chikhaoui, “Cosserat rod modeling of continuum robots from new- tonian and lagrangian perspectives,”IEEE Transactions on Robotics, vol. 39, no. 3, pp. 2360–2378, 2023

    work page 2023

  13. [13]

    Control of elastic soft robots based on real-time finite element method,

    C. Duriez, “Control of elastic soft robots based on real-time finite element method,” in2013 IEEE international conference on robotics and automation. IEEE, 2013, pp. 3982–3987

    work page 2013

  14. [14]

    Model-based disturbance estimation for a fiber-reinforced soft manipulator using orientation sensing,

    B. G. Cangan, S. E. Navarro, B. Yang, Y . Zhang, C. Duriez, and R. K. Katzschmann, “Model-based disturbance estimation for a fiber-reinforced soft manipulator using orientation sensing,” in2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2022, pp. 9424–9430

    work page 2022

  15. [15]

    Design and kinematic modeling of constant curvature continuum robots: A review,

    R. J. Webster III and B. A. Jones, “Design and kinematic modeling of constant curvature continuum robots: A review,”The International Journal of Robotics Research, vol. 29, no. 13, pp. 1661–1683, 2010

    work page 2010

  16. [16]

    Conformational modeling of continuum structures in robotics and structural biology: A review,

    G. S. Chirikjian, “Conformational modeling of continuum structures in robotics and structural biology: A review,”Advanced Robotics, vol. 29, no. 13, pp. 817–829, 2015

    work page 2015

  17. [17]

    Control strategies for soft robotic manipulators: A survey,

    T. George Thuruthel, Y . Ansari, E. Falotico, and C. Laschi, “Control strategies for soft robotic manipulators: A survey,”Soft robotics, vol. 5, no. 2, pp. 149–163, 2018

    work page 2018

  18. [18]

    Neural network and jacobian method for solving the inverse statics of a cable-driven soft arm with nonconstant curvature,

    M. Giorelli, F. Renda, M. Calisti, A. Arienti, G. Ferri, and C. Laschi, “Neural network and jacobian method for solving the inverse statics of a cable-driven soft arm with nonconstant curvature,”IEEE Trans- actions on Robotics, vol. 31, no. 4, pp. 823–834, 2015

    work page 2015

  19. [19]

    Model-based reinforcement learning for closed-loop dynamic control of soft robotic manipulators,

    T. G. Thuruthel, E. Falotico, F. Renda, and C. Laschi, “Model-based reinforcement learning for closed-loop dynamic control of soft robotic manipulators,”IEEE Transactions on Robotics, vol. 35, no. 1, pp. 124– 134, 2018

    work page 2018

  20. [20]

    Learning dexterous manipulation for a soft robotic hand from human demonstrations,

    A. Gupta, C. Eppner, S. Levine, and P. Abbeel, “Learning dexterous manipulation for a soft robotic hand from human demonstrations,” in 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2016, pp. 3786–3793

    work page 2016

  21. [21]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,”nature, vol. 521, no. 7553, pp. 436–444, 2015

    work page 2015

  22. [22]

    Bridge Data: Boosting Generalization of Robotic Skills with Cross-Domain Datasets

    F. Ebert, Y . Yang, K. Schmeckpeper, B. Bucher, G. Georgakis, K. Daniilidis, C. Finn, and S. Levine, “Bridge data: Boosting gener- alization of robotic skills with cross-domain datasets,”arXiv preprint arXiv:2109.13396, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  23. [23]

    An algorithmic perspective on imitation learning,

    T. Osa, J. Pajarinen, G. Neumann, J. A. Bagnell, P. Abbeel, J. Peters et al., “An algorithmic perspective on imitation learning,”Foundations and Trends® in Robotics, vol. 7, no. 1-2, pp. 1–179, 2018

    work page 2018

  24. [24]

    Reinforcement learning in robotics: A survey,

    J. Kober, J. A. Bagnell, and J. Peters, “Reinforcement learning in robotics: A survey,”The International Journal of Robotics Research, vol. 32, no. 11, pp. 1238–1274, 2013

    work page 2013

  25. [25]

    Reinforcement learning of cpg- regulated locomotion controller for a soft snake robot,

    X. Liu, C. D. Onal, and J. Fu, “Reinforcement learning of cpg- regulated locomotion controller for a soft snake robot,”IEEE Trans- actions on Robotics, vol. 39, no. 5, pp. 3382–3401, 2023

    work page 2023

  26. [26]

    Open loop position control of soft continuum arm using deep reinforcement learning,

    S. Satheeshbabu, N. K. Uppalapati, G. Chowdhary, and G. Krishnan, “Open loop position control of soft continuum arm using deep reinforcement learning,” in2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 5133–5139

    work page 2019

  27. [27]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song, “Diffusion policy: Visuomotor policy learning via action diffusion,”The International Journal of Robotics Research, vol. 44, no. 10-11, pp. 1684–1704, 2025

    work page 2025

  28. [28]

    On-device diffusion transformer policy for efficient robot manipulation,

    Y . Wu, H. Wang, Z. Chen, J. Pang, and D. Xu, “On-device diffusion transformer policy for efficient robot manipulation,”arXiv preprint arXiv:2508.00697, 2025

    work page arXiv 2025

  29. [29]

    Hierarchical diffu- sion policy for kinematics-aware multi-task robotic manipulation,

    X. Ma, S. Patidar, I. Haughton, and S. James, “Hierarchical diffu- sion policy for kinematics-aware multi-task robotic manipulation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 18 081–18 090

    work page 2024

  30. [30]

    3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations

    Y . Ze, G. Zhang, K. Zhang, C. Hu, M. Wang, and H. Xu, “3d diffusion policy: Generalizable visuomotor policy learning via simple 3d representations,”arXiv preprint arXiv:2403.03954, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  31. [31]

    Learning from demonstration,

    S. Schaal, “Learning from demonstration,”Advances in neural infor- mation processing systems, vol. 9, 1996

    work page 1996

  32. [32]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    X. Liu, C. Gong, and Q. Liu, “Flow straight and fast: Learning to generate and transfer data with rectified flow,”arXiv preprint arXiv:2209.03003, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  33. [33]

    Flow Matching for Generative Modeling

    Y . Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,”arXiv preprint arXiv:2210.02747, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  34. [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

    work page 2020

  35. [35]

    Flowpolicy: Enabling fast and robust 3d flow-based policy via consistency flow matching for robot manipulation,

    Q. Zhang, Z. Liu, H. Fan, G. Liu, B. Zeng, and S. Liu, “Flowpolicy: Enabling fast and robust 3d flow-based policy via consistency flow matching for robot manipulation,” inProceedings of the AAAI Con- ference on Artificial Intelligence, vol. 39, no. 14, 2025, pp. 14 754– 14 762

    work page 2025

  36. [36]

    A loop-closure theory for the analysis and synthesis of compliant mechanisms,

    L. L. Howell and A. Midha, “A loop-closure theory for the analysis and synthesis of compliant mechanisms,”Journal of Mechanical Design, vol. 118, no. 1, pp. 121–125, 1996

    work page 1996

  37. [37]

    Scalable diffusion models with transformers,

    W. Peebles and S. Xie, “Scalable diffusion models with transformers,” 2023

    work page 2023

  38. [38]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  39. [39]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma, “Adam: A method for stochastic optimization,”arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  40. [40]

    J. J. Craig,Introduction to robotics: mechanics and control, 3/E. Pearson Education India, 2009

    work page 2009