Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

Co-Design of Soft Gripper with Neural Physics

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read This paper claims that jointly optimizing a soft gripper's block-wise stiffness distribution and grasp pose through a neural surrogate trained on finite-element simulation yields grippers that significantly outperform rigid and overly…

desk verdict A genuinely integrated co-design pipeline with real hardware validation, but the missing per-block stiffness verification and lack of statistics leave the central claim about the optimized distribution under-supported. read the letter →

arxiv 2505.20404 v3 pith:IMXCP6DT submitted 2025-05-26 cs.RO

classification cs.RO
keywords softgripperdesignneuralphysicssurrogatestiffnessdistributionoptimizationtendon-drivenflexurefingergraspposeco-designsim-to-realtransferfiniteelementsimulation
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 tries to establish that a soft gripper's design and control can be co-optimized: by jointly choosing the spatial stiffness distribution across the finger blocks and the grasp pose, using a fast neural surrogate trained on finite-element simulation, one obtains a gripper that beats both rigid and overly soft designs. The authors derive a uniform-pressure tendon routing rule so that the finger bends evenly along its whole length instead of curling at the tip, then train a point-cloud-plus-MLP surrogate to predict grasp outcomes from object geometry, mass, and the 22-block stiffness vector. They embed this surrogate in a gradient-based outer-inner optimization loop and 3D-print the resulting design in TPU by varying infill density and wall loops. In simulation the jointly optimized gripper reaches 92.1% success on light objects and 89.5% on heavy objects with pose sampling; in hardware it succeeds in 68 of 100 trials, versus 50 for a rigid gripper and 42 for a soft gripper.

What carries the argument

The load-bearing machinery is twofold. First, the uniform-pressure tendon waypoint distribution $h_i = H(1 - l_i/L)^2$, derived from requiring constant bending moment along the finger, makes the finger conform to object geometry rather than curling at the tip; the tendon force is modeled as uniform tension projected onto surface tangent planes. Second, a neural physics surrogate, a point-cloud feature extractor feeding a multilayer perceptron, is trained on roughly 80,000 finite-element simulations with randomized stiffnesses and poses, and predicts the object's body force, pose displacement, and ground collision. The surrogate's smooth gradients make it possible to run gradient descent on the 22-dimensional stiffness vector inside an outer-inner loop that also samples candidate grasp poses.

What would settle it

Print the optimized gripper, actuate the tendon under a known tension, and compare the measured finger curvature and contact pattern against the point-force finite-element prediction; if the deformation differs substantially, for instance by showing tip curling or uneven bending, then the optimized stiffness distribution was tuned against the wrong physics. A complementary check is to compare hardware grasp success between the optimized gripper and the same gripper with uniform tendon waypoint spacing.

Watch

Extended reading notes

Core claim

The central claim is that co-designing the stiffness pattern together with the grasp pose, mediated by a differentiable neural physics model, yields a soft gripper that generalizes better than either extreme of the stiffness spectrum. The optimized designs are not uniform: they have asymmetric block stiffnesses and at least one rigid fingertip, an emergent structure the authors compare to primate nails. The same surrogate used for optimization serves at deployment as a pose selector given a fixed design, and sim-to-real transfer is achieved by mapping simulated Young's moduli to printer parameters such as infill for segment blocks and wall loops for flexure blocks. The paper reports that joint optimization across objects both avoids local minima and transfers to out-of-domain objects, with the co-designed gripper outperforming soft, semi-rigid, rigid, and per-object-optimized baselines in both simulation and hardware.

Load-bearing premise

The load-bearing assumption is that the simplified tendon model, which replaces the tendon's kinematic constraint with uniform tension applied as point forces at surface waypoints, predicts the real flexure finger's deformation accurately enough for the stiffness pattern optimized under it to transfer to the 3D-printed gripper.

Editorial extensions

If this is right

  • If the central claim is correct, co-design of soft end-effectors can move from manual stiffness tuning to data-driven optimization, with the same surrogate acting as optimizer and deployment-time pose selector.
  • The reported speedup of about three orders of magnitude over differentiable simulation gradients makes design search over many objects tractable on a single GPU.
  • Sim-to-real transfer of optimized stiffness is achievable with a single TPU material by varying infill and wall loops, avoiding multimaterial fabrication.
  • Joint optimization across objects yields designs that generalize to out-of-domain objects in simulation, suggesting the discovered stiffness patterns capture object-independent grasping principles.
  • Pose sampling through the neural surrogate improves success rates for every baseline, making it a plug-in improvement independent of design.

Reading between the lines

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

  • The uniform-pressure tendon rule could be tested independently of the co-design loop: if it is the true cause of whole-finger conforming, then a gripper printed with uniform waypoint spacing should show tip curling and lower success on the same objects.
  • The emergent rigid-tip pattern suggests a general design principle for compliant grippers, soft body for conforming and stiff tip for load transfer, which could guide hand design even without simulation.
  • Because the framework optimizes over a fixed geometry and stiffness only, extending it to topology or to time-varying stiffness would test whether the discovered asymmetric patterns persist in richer design spaces.
  • The surrogate's dependence on pose-sampler diversity implies that improving the candidate pose generator may yield larger gains than improving the network, a testable prediction.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. This paper proposes a co-design framework for a two-finger soft gripper in which a 22-block stiffness distribution and a grasp pose are jointly optimized using a neural surrogate trained on finite-element simulation data. The simulation uses a simplified tendon model with a uniform-pressure waypoint distribution. The optimized gripper is fabricated by varying infill and shell parameters in 3D printing. The paper reports that the jointly optimized gripper achieves higher success rates than uniform soft and semi-rigid baselines in simulation (Table 1) and in hardware trials (Table 2, aggregate 68/100 vs. 50/100 and 42/100), and that the surrogate-based optimization is about three orders of magnitude faster than using differentiable simulation gradients (Figure 4).

Significance. If the central claims hold, the paper makes a useful contribution by demonstrating an end-to-end co-design pipeline for soft grippers, combining a learned physics surrogate with hardware fabrication. The hardware experiments are a notable strength because they provide an external evaluation beyond the simulation. The paper also yields an interesting design insight (asymmetric stiffness with relatively rigid fingertips). However, the strength of the claims is limited by the absence of statistical rigor and by several validation gaps, which are discussed below.

major comments (5)
  1. [Section 4.3, Table 2] The abstract claims that the co-designed gripper 'significantly outperform[s] baseline designs,' but no statistical test or confidence interval is provided. The aggregate hardware counts are 68/100 (optimized), 50/100 (rigid), and 42/100 (soft), yet per-object results vary widely; for example, on the first two objects the optimized gripper achieves 6/10 and 7/10, while the rigid gripper achieves 8/10 and 8/10. Without a paired test across the ten objects or at least confidence intervals on the aggregate proportions, these differences could be within sampling noise for individual objects. This is load-bearing because the central claim is the significance of the improvement.
  2. [Section 4.2 and Appendix 6.3] The paper does not validate that the 3D-printed gripper realizes the optimized 22-dimensional stiffness vector. Figure 5 reports Young's moduli for individual flexure and segment blocks under a few printer settings, and Appendix 6.3 states that the simulated stiffness values are 'mapped onto the printed fingers by linearly scaling them to the empirical modulus,' but no per-block comparison (e.g., measured versus intended stiffness for each of the 22 blocks, or a finger-level deformation matching study) is presented. Consequently, the hardware improvement over the baselines cannot be attributed to the specific co-optimized stiffness distribution rather than to generic compliance of a non-uniform gripper. This gap is not acknowledged in the Limitations section.
  3. [Section 3.1] The simplified tendon model, which replaces the inextensible tendon kinematic constraint with uniform tension applied as point forces at surface waypoints, is a key assumption for the correctness of the optimized design. The paper provides no experimental validation that this model predicts the deformation of the physical tendon-driven flexure finger, such as a comparison of simulated and real finger curvature or contact shape under actuation. Furthermore, the simulation applies a fixed tendon force (Section 3.3.1) while the hardware uses position-controlled tendon displacement (Section 4.3), a control-mode mismatch that can affect sim-to-real transfer. If the tendon model is inaccurate, the optimized stiffness distribution is tuned to the wrong physics.
  4. [Section 3.3.2 and Table 1] The simulation evaluation is not independent of the surrogate model, because the surrogate is trained on the same FEM simulator used to compute the success rates in Table 1. The simulation results therefore demonstrate that the surrogate can be optimized within the simulator, but they do not provide an independent test of the co-design. Since the hardware experiments are the only independent evaluation, the validation gaps in the two comments above become critical for the paper's central claim.
  5. [Section 4.1 and Appendix 6.4] There is an inconsistency in object density units: Section 4.1 states 'density = 2 kg/ m3' and '8 kg/ m3,' while Appendix 6.4 uses 'density of 8.0 ± 0.1 g/cm3.' These differ by three orders of magnitude, and it is unclear which values were actually used in the simulation experiments. This ambiguity undermines the reproducibility of the 'light' and 'heavy' categories in Table 1 and should be corrected.
minor comments (6)
  1. [Section 3.2] The phrase 'tenden waypoint' appears in the text; it should be 'tendon waypoint.' The word 'tendon' is misspelled in a few other places in the manuscript.
  2. [Figure 4d] The claim of 'approximately three orders of magnitude' speedup needs a precise statement of what is being compared (e.g., wall-clock time per gradient step, total optimization time, or including data generation and training time) and whether the comparison is normalized appropriately.
  3. [References] References [20] and [30] are duplicates of the same paper (Allen et al., 'Inverse design for fluid-structure interactions using graph network simulators'); one duplicate should be removed and citations updated accordingly.
  4. [Section 6.2 vs. Section 4.1] The appendix says data generation used 'six 3090 GPUs,' while Section 4.1 says 'four 3090 GPUs.' The discrepancy should be resolved.
  5. [Section 3.3.1 and Eq. (5)] The neural surrogate architecture is described only briefly; specifying the number of hidden units, activation functions, training epochs, and the values of the loss weights w1 and w2 would improve reproducibility.
  6. [Section 4.2 and Table 4] It is unclear how the chosen wall-loop and infill settings map to the continuous stiffness values listed in Table 4. A table that lists the specific printer settings for each stiffness level would clarify the sim-to-real transfer.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: optimized stiffness is a free optimization output, and hardware trials provide an independent external benchmark.

full rationale

The paper's central claim—that a jointly optimized block-wise stiffness distribution and grasp pose outperform rigid and soft baselines—is not definitional. The 22D stiffness vector is the output of a gradient-based optimization loop over a neural surrogate trained on FEM simulation; it is not a parameter fitted to the evaluation metric. Simulation success rates are computed from the full soft-body simulator over 850 frames at 4000 fps, not from the surrogate's own loss, so the surrogate functions as an approximation tool rather than as the evaluator of the reported success rates. Hardware experiments (Table 2) are an independent external benchmark: printed fingers are calibrated to measured moduli via linear scaling (Appendix 6.3), and success is judged by physical lift-and-translate trials, so the central result does not reduce to the simulation inputs. The one author-overlapping citation, [43], merely notes prior use of the uniform-pressure model in a pneumatic soft robot; the waypoint formula Eq. (3) is derived in-text from Eq. (2) and attributed to the independent source [42], so the self-citation is not load-bearing. The sim-to-real transfer of the exact per-block stiffness is a validation or correctness concern rather than a circularity: the printed realization of the optimized vector is assumed rather than per-block verified, but that is an empirical gap, not an identity between input and output. No step in the derivation chain is, by the paper's own equations, equivalent to its own inputs.

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

The central optimization depends on the simulated physics being correct and the surrogate generalizing. The most significant assumptions are the simplified tendon model, the uniform-pressure waypoint rule, and the fidelity of the Warp FEM simulator. The linear stiffness mapping and loss weights are hand-chosen elements that affect the final design.

free parameters (5)
  • Loss weights w1, w2, wd, wp
    Eqs. (4) and (5) use manually chosen weights that shape the optimized stiffness and pose selection; numeric values are not reported.
  • Sim-to-real stiffness linear scaling coefficient
    Appendix 6.3 says simulated log(E) values are linearly scaled to measured printed modulus, but the scaling is not specified or validated.
  • Object density categories = 2 and 8 kg/m3
    Light and heavy density values were chosen experimentally for stable simulation and directly affect the reported success rates.
  • Stiffness domain bounds = log E in [13.5, 17.0]
    The Young's modulus range 0.7 to 24 MPa was chosen for simulation stability and constrains the optimized design space.
  • Algorithm 1 hyperparameters = B = 5; Tp not stated
    The top-B pose selection and patience threshold in joint optimization are algorithm choices with no sensitivity analysis.
assumptions (5)
  • domain assumption A tendon can be modeled as uniform tension applied as point forces at surface waypoints.
    Section 3.1 simplifies the inextensible tendon by enforcing uniform tension only; this is asserted to preserve the mechanical effect but is not validated against a full tendon model.
  • domain assumption The bending moment along the finger is proportional to hi times Ti.
    Section 3.2 uses this linear relation to derive the quadratic waypoint height rule in Eq. (3), inheriting assumptions from [42,43].
  • domain assumption The NVIDIA Warp FEM simulation is an accurate ground truth for soft gripper grasping.
    All neural surrogate training labels come from this simulator, and the simulation success rates in Table 1 are measured with the same physics.
  • domain assumption The neural surrogate generalizes from 80k in-domain YCB samples to out-of-domain objects and real camera point clouds.
    The paper trains on in-domain YCB objects and evaluates on KIT/EGAD and real hardware; the Limitations section notes the training dataset is limited.
  • domain assumption Anygrasp poses augmented by rigid-gripper optimization cover the pose space relevant for the soft gripper.
    Section 3.3.1 relies on this pose sampler, and the Limitations section admits most poses are downward-facing and many are kinematically infeasible.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Co-Design of Soft Gripper with Neural Physics." pith.science (2026). https://pith.science/paper/IMXCP6DT

@misc{pith2026250520404,
  author       = {Pith},
  title        = {Pith review of: Co-Design of Soft Gripper with Neural Physics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IMXCP6DT}},
  note         = {Machine review of arXiv:2505.20404}
}
read the original abstract

For robot manipulation, both the controller and end-effector design are crucial. Soft grippers are generalizable by deforming to different geometries, but designing such a gripper and finding its grasp pose remains challenging. In this paper, we propose a co-design framework that generates an optimized soft gripper's block-wise stiffness distribution and its grasping pose, using a neural physics model trained in simulation. We derived a uniform-pressure tendon model for a flexure-based soft finger, then generated a diverse dataset by randomizing both gripper pose and design parameters. A neural network is trained to approximate this forward simulation, yielding a fast, differentiable surrogate. We embed that surrogate in an end-to-end optimization loop to optimize the ideal stiffness configuration and best grasp pose. Finally, we 3D-print the optimized grippers of various stiffness by changing the structural parameters. We demonstrate that our co-designed grippers significantly outperform baseline designs in both simulation and hardware experiments. More info: http://yswhynot.github.io/codesign-soft/

Figures

Figures reproduced from arXiv: 2505.20404 by the authors.

Figure 1
Figure 1. We introduce a co-design framework that jointly optimizes the spatial stiffness distribution and grasp [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Tendon along a soft finger. The side view shows the dimensions of the tendon waypoints going [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. System architecture. We first generate data by sampling different stiffness design parameters, different [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Numerical results of gradient-based optimization in simulations. (a) Differentiable simulation loss. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: 3D printing different stiffnesses. (a) Top: different infills for the segment blocks; Bottom: different [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Automated Synthesis of Facial Mechanisms for Conversational Animatronic Robots

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A parametric linkage face template plus hierarchical collision-driven optimization synthesizes manufacturable facial mechanisms from 2D portraits and runs them with dual-identity conversational motion.

Reference graph

Works this paper leans on

50 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    Rus and M

    D. Rus and M. T. Tolley. Design, fabrication and control of soft robots. Nature, 521(7553): 467–475, 2015

  2. [2]

    Shintake, V

    J. Shintake, V . Cacucciolo, D. Floreano, and H. Shea. Soft robotic grippers. Advanced materi- als, 30(29):1707035, 2018

  3. [3]

    Lipson and J

    H. Lipson and J. B. Pollack. Automatic design and manufacture of robotic lifeforms. Nature, 406(6799):974–978, 2000

  4. [4]

    J. Xu, T. Chen, L. Zlokapa, M. Foshey, W. Matusik, S. Sueda, and P. Agrawal. An end-to-end differentiable framework for contact-aware robot design. arXiv preprint arXiv:2107.07501 , 2021

  5. [5]

    T. Chen, Z. He, and M. Ciocarlie. Hardware as policy: Mechanical and computational co- optimization using deep reinforcement learning. arXiv preprint arXiv:2008.04460, 2020

  6. [6]

    Schaff, A

    C. Schaff, A. Sedal, S. Ni, and M. R. Walter. Sim-to-real transfer of co-optimized soft robot crawlers. Autonomous Robots, 47(8):1195–1211, 2023

  7. [7]

    He and M

    Z. He and M. Ciocarlie. Morph: Design co-optimization with reinforcement learning via a differentiable hardware model proxy. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 7764–7771. IEEE, 2024

  8. [8]

    A. Zhao, J. Xu, M. Konakovi ´c-Lukovi´c, J. Hughes, A. Spielberg, D. Rus, and W. Matusik. Robogrammar: graph grammar for terrain-optimized robot design. ACM Transactions on Graphics (TOG), 39(6):1–16, 2020

Show all 50 references
  1. [9]

    W. G. Bircher, A. S. Morgan, and A. M. Dollar. Complex manipulation with a simple robotic hand through contact breaking and caging. Science Robotics, 6(54):eabd2666, 2021

  2. [10]

    L. U. Odhner, L. P. Jentoft, M. R. Claffee, N. Corson, Y . Tenzer, R. R. Ma, M. Buehler, R. Ko- hout, R. D. Howe, and A. M. Dollar. A compliant, underactuated hand for robust manipulation. The International Journal of Robotics Research, 33(5):736–752, 2014

  3. [11]

    J. Gao, M. Y . Michelis, A. Spielberg, and R. K. Katzschmann. Sim-to-real of soft robots with learned residual physics. IEEE Robotics and Automation Letters, 2024

  4. [12]

    F. Chen, W. Xu, H. Zhang, Y . Wang, J. Cao, M. Y . Wang, H. Ren, J. Zhu, and Y . Zhang. Topology optimized design, fabrication, and characterization of a soft cable-driven gripper. IEEE Robotics and Automation Letters, 3(3):2463–2470, 2018

  5. [13]

    Chen and M

    F. Chen and M. Y . Wang. Design optimization of soft robots: A review of the state of the art. IEEE Robotics & Automation Magazine, 27(4):27–43, 2020

  6. [14]

    Buckley, B

    H. Buckley, B. Zhou, and D. Zingg. Airfoil optimization using practical aerodynamic design requirements. Journal of Aircraft, 47, 09 2010. doi:10.2514/1.C000256

  7. [15]

    P. He, C. Mader, J. Martins, and K. Maki. Dafoam: An open-source adjoint framework for multidisciplinary design optimization with openfoam. AIAA Journal, page 2019, 10 2019. doi:10.2514/1.j058853. 10

  8. [16]

    Feshbach, W.-H

    D. Feshbach, W.-H. Chen, L. Xu, E. Schaumburg, I. Huang, and C. Sung. Algorithmic design of kinematic trees based on csc dubins planning for link shapes. In International Workshop on the Algorithmic Foundations of Robotics (WAFR), 2024

  9. [17]

    Kodnongbua, I

    M. Kodnongbua, I. G. Y . Lou, J. Lipton, and A. Schulz. Computational design of passive grippers. arXiv preprint arXiv:2306.03174, 2023

  10. [18]

    M. Macklin. Warp: A high-performance python framework for gpu simulation and graphics. https://github.com/nvidia/warp, March 2022. NVIDIA GPU Technology Conference (GTC)

  11. [19]

    Y . Hu, L. Anderson, T.-M. Li, Q. Sun, N. Carr, J. Ragan-Kelley, and F. Durand. Difftaichi: Differentiable programming for physical simulation. ICLR, 2020

  12. [20]

    Allen, T

    K. Allen, T. Lopez-Guevara, K. L. Stachenfeld, A. Sanchez Gonzalez, P. Battaglia, J. B. Ham- rick, and T. Pfaff. Inverse design for fluid-structure interactions using graph network simula- tors. Advances in Neural Information Processing Systems, 35:13759–13774, 2022

  13. [21]

    Georgiev, V

    I. Georgiev, V . Giridhar, N. Hansen, and A. Garg. Pwm: Policy learning with multi-task world models. In The Thirteenth International Conference on Learning Representations

  14. [22]

    Calandra, A

    R. Calandra, A. Seyfarth, J. Peters, and M. P. Deisenroth. Bayesian optimization for learning gaits under uncertainty: An experimental comparison on a dynamic bipedal walker. Annals of Mathematics and Artificial Intelligence, 76:5–23, 2016

  15. [23]

    Snoek, H

    J. Snoek, H. Larochelle, and R. P. Adams. Practical bayesian optimization of machine learning algorithms, 2012. URL https://arxiv.org/abs/1206.2944

  16. [24]

    Hansen, S

    N. Hansen, S. D. M ¨uller, and P. Koumoutsakos. Reducing the time complexity of the deran- domized evolution strategy with covariance matrix adaptation (cma-es). Evolutionary compu- tation, 11(1):1–18, 2003

  17. [25]

    K ¨ulz and M

    J. K ¨ulz and M. Althoff. Optimizing modular robot composition: A lexicographic genetic algorithm approach. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 16752–16758. IEEE, 2024

  18. [26]

    H. Ha, S. Agrawal, and S. Song. Fit2form: 3d generative model for robot gripper form design. In Conference on Robot Learning, pages 176–187. PMLR, 2021

  19. [27]

    X. Xu, H. Ha, and S. Song. Dynamics-guided diffusion model for robot manipulator design. arXiv preprint arXiv:2402.15038, 2024

  20. [28]

    Pfaff, M

    T. Pfaff, M. Fortunato, A. Sanchez-Gonzalez, and P. Battaglia. Learning mesh-based simu- lation with graph networks. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=roNqYL0_XP

  21. [29]

    de Avila Belbute-Peres, T

    F. de Avila Belbute-Peres, T. D. Economon, and J. Z. Kolter. Combining differentiable PDE solvers and graph neural networks for fluid flow prediction. CoRR, abs/2007.04439, 2020. URL https://arxiv.org/abs/2007.04439

  22. [30]

    K. R. Allen, T. Lopez-Guavara, K. Stachenfeld, A. Sanchez-Gonzalez, P. Battaglia, J. B. Ham- rick, and T. Pfaff. Inverse design for fluid-structure interactions using graph network simula- tors. In A. H. Oh, A. Agarwal, D. Belgrave, and K. Cho, editors, Advances in Neural Info...

  23. [31]

    Piazza, G

    C. Piazza, G. Grioli, M. G. Catalano, and A. Bicchi. A century of robotic hands. Annual Review of Control, Robotics, and Autonomous Systems, 2(1):1–32, 2019. 11

  24. [32]

    R. A. Bilodeau, E. L. White, and R. K. Kramer. Monolithic fabrication of sensors and actuators in a soft robotic gripper. In 2015 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 2324–2329. IEEE, 2015

  25. [33]

    Rothemund, A

    P. Rothemund, A. Ainla, L. Belding, D. J. Preston, S. Kurihara, Z. Suo, and G. M. White- sides. A soft, bistable valve for autonomous control of soft actuators. Science Robotics, 3(16): eaar7986, 2018

  26. [34]

    Y . Wei, Y . Chen, T. Ren, Q. Chen, C. Yan, Y . Yang, and Y . Li. A novel, variable stiffness robotic gripper based on integrated soft actuating and particle jamming. Soft Robotics, 3(3): 134–143, 2016

  27. [35]

    J. D. Hubbard, R. Acevedo, K. M. Edwards, A. T. Alsharhan, Z. Wen, J. Landry, K. Wang, S. Schaffer, and R. D. Sochol. Fully 3d-printed soft robots with integrated fluidic circuitry. Science Advances, 7(29):eabe5257, 2021

  28. [36]

    Y . Zhai, A. De Boer, J. Yan, B. Shih, M. Faber, J. Speros, R. Gupta, and M. T. Tolley. Desktop fabrication of monolithic soft robotic devices with embedded fluidic control circuits. Science robotics, 8(79):eadg3792, 2023

  29. [37]

    Kim, W.-Y

    J. Kim, W.-Y . Choi, S. Kang, C. Kim, and K.-J. Cho. Continuously variable stiffness mech- anism using nonuniform patterns on coaxial tubes for continuum microsurgical robot. IEEE Transactions on Robotics, 35(6):1475–1487, 2019

  30. [38]

    Y . S. Narang, A. Degirmenci, J. J. Vlassak, and R. D. Howe. Transforming the dynamic response of robotic structures and systems through laminar jamming. IEEE Robotics and Automation Letters, 3(2):688–695, 2017

  31. [39]

    L. L. Howell. Introduction to compliant mechanisms. Handbook of Compliant Mechanisms, pages 1–13, 2013

  32. [40]

    A. M. Dollar and R. D. Howe. The highly adaptive sdm hand: Design and performance evalu- ation. The international journal of robotics research, 29(5):585–597, 2010

  33. [41]

    Faure, C

    F. Faure, C. Duriez, H. Delingette, J. Allard, B. Gilles, S. Marchesseau, H. Talbot, H. Courte- cuisse, G. Bousquet, I. Peterlik, et al. Sofa: A multi-model framework for interactive physical simulation. Soft tissue biomechanical modeling for computer assisted surgery, pages 2...

  34. [42]

    Hirose and Y

    S. Hirose and Y . Umetani. The development of soft gripper for the versatile robot hand.Mech- anism and machine theory, 13(3):351–359, 1978

  35. [43]

    Glick, S

    P. Glick, S. A. Suresh, D. Ruffatto, M. Cutkosky, M. T. Tolley, and A. Parness. A soft robotic gripper with gecko-inspired adhesive. IEEE Robotics and Automation Letters, 3(2):903–910, 2018

  36. [44]

    C ¸ alli, A

    B. C ¸ alli, A. Walsman, A. Singh, S. S. Srinivasa, P. Abbeel, and A. M. Dollar. Benchmarking in manipulation research: The YCB object and model set and benchmarking protocols. CoRR, abs/1502.03143, 2015. URL http://arxiv.org/abs/1502.03143

  37. [45]

    Son and B

    D. Son and B. Kim. Local object crop collision network for efficient simulation of non-convex objects in gpu-based simulators. arXiv preprint arXiv:2304.09439, 2023

  38. [46]

    H.-S. Fang, C. Wang, H. Fang, M. Gou, J. Liu, H. Yan, W. Liu, Y . Xie, and C. Lu. Anygrasp: Robust and efficient grasp perception in spatial and temporal domains. IEEE Transactions on Robotics, 39(5):3929–3945, 2023

  39. [47]

    C. R. Qi, H. Su, K. Mo, and L. J. Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660, 2017. 12

  40. [48]

    Kasper, Z

    A. Kasper, Z. Xue, and R. Dillmann. The KIT object models database: An object model database for object recognition, localization and manipulation in service robotics. Int. J. Robotics Res., 31(8):927–934, 2012. doi:10.1177/0278364912445831. URL https://doi. org/10.1177/027836...

  41. [49]

    Morrison, P

    D. Morrison, P. Corke, and J. Leitner. Egad! an evolved grasping analysis dataset for diversity and reproducibility in robotic manipulation. CoRR, abs/2003.01314, 2020. URL https: //arxiv.org/abs/2003.01314

  42. [50]

    R. R. Craig Jr and E. M. Taleff. Mechanics of materials. John Wiley & Sons, 2020. 13 Appendix 6.1 Experiments with Disturbance We provide more justifications for larger impulses in simulation, shown in Table 3(a). Here, 0.1 N.s is under the fixed disturbance force of 500 N for...

Pith tools

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