Pith. sign in

REVIEW 5 major objections 5 minor 2 cited by

TensorTouch: Calibration of Tactile Sensors for High Resolution Stress Tensor and Deformation for Dexterous Manipulation

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

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

Pith's one-line read TensorTouch claims that a soft optical tactile gel can be calibrated with finite-element simulation and a hierarchical vision transformer to output pixel-level stress tensors, deformation fields, and contact forces from a single image.

desk verdict A serious, well-built tactile-calibration pipeline with a real robot demo, but the central pixel-level stress-tensor claim rests on unvalidated FE targets and the abstract oversells the results; deserves revision, not rejection. read the letter →

arxiv 2506.08291 v1 pith:2AQ7HJLG submitted 2025-06-09 cs.RO

Won Kyung Do , Matthew Strong , Aiden Swann , Boshu Lei , Monroe Kennedy III This is my paper
classification cs.RO
keywords tactilesensingopticalsensorstresstensorestimationfiniteelementanalysishyperelasticmaterialvisiontransformerdexterousmanipulationsim-to-realcalibration
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

TensorTouch claims that a soft optical tactile sensor can be calibrated to output physically meaningful contact information at every pixel: the full six-component stress tensor, the three-axis displacement field, and contact normal and shear force distributions. The authors argue this works even for very compliant gels under large deformations, a regime where earlier calibration methods relied on small displacements or only a single six-axis force wrench. Their route is to simulate the gel with a nonlinear finite-element model driven by motion-captured real indentations with diverse 3D-printed contact geometries, convert the 3D stress and deformation fields into 2D image-like training targets, and train a lightweight hierarchical vision transformer to predict those targets from one tactile image. If correct, this would let robots use calibrated touch fields instead of raw sensor images, making tactile perception transferable across sensor hardware and enabling multi-contact tasks such as selectively grasping one of two strings.

What carries the argument

The load-bearing mechanism is a nonlinear finite-element model of the gel: a single homogeneous hyperelastic material described by a strain-energy function of the first strain invariant, a literature friction coefficient of 2.2 against the indenter, and a fully fixed bottom surface, solved with a general-purpose finite-element solver. To make the simulation match the real sensor, the authors collect synchronized motion-capture poses and force/torque readings during human-guided indentations with 12 diverse 3D-printed contact geometries, extract keypoints from each trajectory, simulate those keypoints in the finite-element model, and convert the 3D nodal stress and displacement results into 2D image channels through a Gaussian-process-based 2D-to-3D correspondence map. Simulated forces are then corrected component-wise using offset ratios and a random forest fitted to real force/torque measurements, and a dynamic baseline-update mechanism re-anchors the undeformed reference image to handle gel hysteresis. The network is a hierarchical vision transformer pretrained with masked autoencoding and paired with a Q-Upsampling decoder that fuses multi-resolution features before separate convolutional heads emit the five output modalities: displacement, normal stress, shear stress, contact normal force, and contact shear force.

What would settle it

Run the same network and clustering on indentations with indenter shapes never seen in training, while measuring ground-truth force with a high-accuracy sensor, and check the per-contact integrated force error after removing the random-forest correction; if errors grow well beyond the reported 0.24 N, the finite-element stress ground truth is not generalizing. A second check is to compare the finite-element-predicted gel surface deformation against an independent 3D scan of the gel during a 5 mm indentation; a mismatch of more than a fraction of a millimeter would mean the dense stress maps are not trustworthy even where forces are corrected.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the TensorTouch pipeline yields accurate pixel-level estimates of the contact stress tensor, deformation field, and contact forces from a single optical tactile image while accommodating the large deformations that soft gels undergo. Evaluated on a custom modular hemispherical sensor with two gel softness levels, the framework reports mean contact-position errors of 0.684 mm, 0.376 mm, and 1.292 mm along x, y, z and mean force errors of 0.106 N, 0.113 N, and 0.139 N per axis, with an overall force-magnitude error of 0.239 N, compared against motion capture and a commercial force/torque sensor. The authors further claim that these calibrated fields let a two-finger hand track multiple simultaneous contacts and selectively maintain grasp on a moving target object, with success rates from 67.7% for two identical cables up to 90% for two rigid objects.

Load-bearing premise

The whole calibration rests on the assumption that the finite-element model of the gel—one homogeneous hyperelastic material, a single literature friction coefficient, and a rigidly fixed bottom—predicts the real gel's internal stress fields well enough that a network trained on those simulations can estimate true forces and stresses from images, with the force-correction step able to fix only systematic magnitude errors, not fundamentally wrong stress distributions.

Editorial extensions

If this is right

  • Very soft gels become usable for tactile manipulation without giving up accurate measurement, because the finite-element pipeline supplies the large-deformation ground truth that small-deformation methods lacked.
  • Policies can consume calibrated force and stress fields instead of raw tactile images, which is the route the paper identifies toward sensor-agnostic manipulation.
  • Multiple simultaneous contacts can be separated, tracked, and attributed to individual objects, enabling tasks such as selective string grasping.
  • Recalibration after sensor aging or replacement is cheap, since the random-forest correction needs only about 5 to 7 measurement points.
  • The modular sensor design and parametric finite-element setup mean the same calibration framework can be applied to other flat or curved optical tactile sensors.

Reading between the lines

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

  • Beyond the paper, the pipeline should transfer to other optical sensor geometries, since the finite-element model and the correspondence mapping are parametric in sensor shape; the cost is recollecting calibration data and re-fitting the map, not redesigning the network.
  • The reported force errors sit close to the reference sensor's resolution (about 0.06 N), so part of the quoted error could be measurement noise; a higher-resolution reference would reveal the true floor of the model.
  • If the predicted stress fields are physically consistent, continuum equilibrium could be imposed as a self-supervised training signal, allowing the network to improve on unlabeled real images.
  • Selective grasping of identical cables succeeded at a lower rate than mixed object pairs, suggesting the practical limit for telling two similar objects apart is force-field resolution; finer discrimination should improve with a more accurate reference.
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 / 5 minor

Summary. The paper presents TensorTouch, a calibration framework for optical tactile sensors that combines Abaqus finite-element simulation of hyperelastic gels with a hierarchical vision transformer (Hiera with a Q-upsampling decoder) to predict pixel-level displacement fields, normal/shear stress tensors, and contact force distributions from raw tactile images. A motion-capture and ATI-based data collection system with diverse YCB-derived indenter shapes is used to collect real images and poses, and a sim-to-real correction step (component-wise offset ratios plus a random-forest regressor) is applied to the FE forces. The authors evaluate cluster-level position and force estimates against motion-capture and ATI measurements, and demonstrate a two-finger selective string-grasping task with success rates between 67.7% and 90% across object combinations.

Significance. If the dense stress-tensor and deformation outputs were validated against independent measurements, TensorTouch could be an important step toward standardized, physics-based tactile sensing for dexterous manipulation. Strengths of the paper include the large-scale paired dataset, the use of multiple indenter geometries from YCB objects to improve generalization, a systematic FE pipeline adaptable to different sensor shapes, and a careful network ablation showing the benefit of the proposed lightweight Hiera+Q-upsampling architecture. The reported sub-Newton force errors and the successful multi-object manipulation demo are encouraging, provided the evaluation protocol is made rigorous.

major comments (5)
  1. [V-A, III-I, III-J] The 60 evaluation points used for the force and position comparison are not described as held out from either the network training set or the ATI-based calibration data (offset ratios and random forest). Because the calibration and evaluation use the same ATI sensor, the reported force errors of 0.106 to 0.239 N could simply reflect fitted values rather than generalization. The paper must state how the 60 points were selected relative to training and calibration, and ideally evaluate on a strictly disjoint set.
  2. [V-A, III-I] The central claim of accurate pixel-level normal and shear stress distributions is not directly validated against any independent ground truth. The only quantitative checks are cluster-level force and position comparisons; the stress tensors are trained on FE targets whose fidelity is uncertain because the gel is modeled as a single homogeneous Yeoh material, the friction coefficient of 2.2 is taken from the literature [54], and the bottom boundary is fully encastre. The offset ratios in Eq. (14)-(15) are scalar multipliers applied to integrated forces and cannot correct spatial stress-distribution errors. The authors should either provide independent validation of the stress and displacement fields (for example, using stereo or structured-light measurement of the physical gel surface under the same indentations) or substantially weaken the stress-accuracy claim.
  3. [Abstract, Table II] The abstract states that experimental validation demonstrates 90% success in selectively grasping one of two strings, but Table II reports 90% only for the two-rigid-object control condition; the string-related conditions achieve 67.7%, 75%, and 85%. The headline claim should be corrected to match the data, or the abstract should clarify which condition the 90% figure refers to.
  4. [V-A, Abstract] The position errors are reported as 0.684, 0.376, and 1.292 mm in x, y, and z, so the z-axis error exceeds 1 mm and the abstract's 'sub-millimeter position accuracy' is not fully supported. The paper also notes that the cluster centroid does not coincide with the indenter origin due to geometry, so part of this error is a systematic offset rather than sensor localization error. Please separate the sensor localization component from the geometric offset and restate the accuracy claim accordingly.
  5. [III-I] The random-forest correction is described only briefly, as 'as few as 5-7 measurement points,' without specifying the input features, the training and validation split, or how the zero-constraint is enforced. Given that this correction is load-bearing for the force estimates, the lack of detail prevents reproducibility and makes the reported accuracies difficult to interpret. Please provide the full calibration protocol or clarify whether the random-forest correction is used in the final evaluation.
minor comments (5)
  1. [III-I (Eq. 14-15)] The ratio r_i is defined component-wise in the text but appears as a single scalar in Equations (14) and (15); please clarify whether r_i is a vector and how its index aligns with the normal and shear force components.
  2. [IV-B] The sentence 'Each dataset includes four sensors with the same gel softness' is ambiguous; please specify the number of physical sensors and gels per softness setting.
  3. [III-H, Table I] Because outputs are normalized using both global and local statistics, the PSNR/MSE/SSIM values in Table I depend on the normalization procedure; please state whether the metrics are computed on the denormalized physical values.
  4. [III-C] The baseline-update threshold is given as 'MSE threshold = 8' without units; please specify the metric (for example, per-pixel squared difference over the full image) so the threshold is interpretable.
  5. [Throughout] There are several typographical and formatting issues, including 'preseves' in Section III-J and corrupted equation formatting in Eq. (10) ('withWandHas'); the manuscript should be proofread.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: the FE-to-image supervised pipeline is self-contained; the ATI-based force correction is declared calibration rather than a hidden prediction, and the self-citations are not load-bearing.

full rationale

The derivation chain is: Abaqus FE simulation of the gel (Section III-E) -> projection of stress, displacement, and contact fields onto 2D images (Section III-H) -> supervised training of a Hiera-based network to map tactile images to those fields (Section III-J) -> contact clustering (Section III-K) -> external comparison with mocap and ATI measurements (Section V-A). The stress-tensor and displacement outputs are learned targets taken directly from the FE simulation, so Table I is a regression-accuracy report against those training labels; it is not an independent physical validation of stress, but it is also not a circular reduction because the network learns a genuine image-to-field map that is not algebraically identical to its input. The force correction in Section III-I, Eqs. (14)-(15), is explicitly a calibration: r_i is fit from ATI readings and multiplies the Abaqus force, so at the calibration points the corrected force equals the ATI measurement by construction. However, the paper presents this as calibration rather than as a first-principles force prediction, and it does not state that the 60 evaluation points in Section V-A were the same points used to fit r_i or the random-forest correction. Without that stated overlap, no forced reduction can be exhibited. The genuine weaknesses are validation gaps: pixel-level stress tensors are never compared with an independent ground truth, and the held-out status of the force-evaluation split is not documented. These are limitations on evidence quality, not circularity. Self-citations to the authors' earlier DenseTact work ([21], [25], [36]) appear in related work and sensor-design context and are not used to justify the FE calibration, the network outputs, or any uniqueness claim, so they are not load-bearing. Overall the central derivation is self-contained, warranting a low circularity score.

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

The central claim depends on material and calibration parameters fit to data (Yeoh coefficients, friction, offset ratios, random forest), on modeling assumptions about the gel and boundary conditions, and on the generalization of the FE simulation. The stress tensor output has no independent validation, so its faithfulness rests entirely on the FE model plus the force corrections.

free parameters (5)
  • Yeoh material parameters C1, C2, C3, D1 for each gel formulation = Not reported numerically; fit to uniaxial and biaxial test data
    The FE model's mechanical response depends on these; parameters are fit to material tests in Appendix B and then used as simulation inputs.
  • Friction coefficient between indenter and gel = 2.2
    Set based on reference [54], not measured for this gel; affects contact forces and stress distributions in the FE model.
  • Component-wise contact offset ratios r_i = Not reported numerically
    Fit to match Abaqus reaction forces to ATI sensor measurements (Eq. 14 and 15); these corrections are applied to all contact points and are a direct fit to the evaluation instrument.
  • Random forest force correction model = Trained on 5-7 paired measurements
    Learned mapping from model-predicted forces to ATI ground truth, used for ongoing sim-to-real correction; if evaluation shares these measurements, force estimates are fit to the evaluation target.
  • Data selection thresholds = F_contact=1.5N, F_detach=1.1N, MSE_threshold=8
    Hand-chosen thresholds that determine which contacts are included in the training dataset and when the baseline image resets; directly affect the training labels.
assumptions (6)
  • domain assumption The Yeoh hyperelastic constitutive model adequately represents the silicone gel's response over the deformation range.
    Adopted in Sections III-D and III-E; uniaxial and biaxial data is fitted to Yeoh, but the model form itself is assumed.
  • domain assumption The gel is modeled as a single homogeneous material; the thin reflective surface layer is mechanically negligible.
    Section III-E states this simplification explicitly.
  • domain assumption The bottom gel surface is fully constrained (encastre), mimicking full attachment to the camera and LED module.
    Section III-E boundary conditions; this anchoring is load-bearing for the simulated deformation fields.
  • domain assumption The friction coefficient from reference [54] applies to this indenter-gel interface.
    Section III-E sets the friction coefficient to 2.2 based on silicone rubber literature, without direct measurement of this sensor's gel.
  • domain assumption Abaqus FE with hybrid elements and penalty contact accurately simulates the multi-contact large-deformation scenarios.
    The entire training label generation depends on the FE solver being a valid surrogate for real physics; no independent validation of stress fields is provided.
  • domain assumption The GPR-based 2D-to-3D mapping trained on pokes generalizes across the whole gel surface.
    Section III-G; if this mapping is inaccurate, all pixel-level physical outputs are misaligned with the real gel.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TensorTouch: Calibration of Tactile Sensors for High Resolution Stress Tensor and Deformation for Dexterous Manipulation." pith.science (2026). https://pith.science/paper/2AQ7HJLG

@misc{pith2026250608291,
  author       = {Pith},
  title        = {Pith review of: TensorTouch: Calibration of Tactile Sensors for High Resolution Stress Tensor and Deformation for Dexterous Manipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2AQ7HJLG}},
  note         = {Machine review of arXiv:2506.08291}
}
read the original abstract

Advanced dexterous manipulation involving multiple simultaneous contacts across different surfaces, like pinching coins from ground or manipulating intertwined objects, remains challenging for robotic systems. Such tasks exceed the capabilities of vision and proprioception alone, requiring high-resolution tactile sensing with calibrated physical metrics. Raw optical tactile sensor images, while information-rich, lack interpretability and cross-sensor transferability, limiting their real-world utility. TensorTouch addresses this challenge by integrating finite element analysis with deep learning to extract comprehensive contact information from optical tactile sensors, including stress tensors, deformation fields, and force distributions at pixel-level resolution. The TensorTouch framework achieves sub-millimeter position accuracy and precise force estimation while supporting large sensor deformations crucial for manipulating soft objects. Experimental validation demonstrates 90% success in selectively grasping one of two strings based on detected motion, enabling new contact-rich manipulation capabilities previously inaccessible to robotic systems.

Figures

Figures reproduced from arXiv: 2506.08291 by the authors.

Figure 1
Figure 1. TensorTouch framework applied to multi-finger robotic manipulation [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of a pipeline. We capture real-world pose, force of indenter and sensor images and simulate gel deformation and sensor motion with FE [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Experimental setup for data collection with motion capture cameras [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Left: Collection of diverse 3D-printed indenter shapes derived from [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Keypoint extraction process showing the correspondence between [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: The application of frameworks on various sensors. The real-to-sim [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Conversion of 3D FE analysis results to 2D tactile image representa [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: 2D-3D correspondence mapping between image coordinates and [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: The top row illustrates our self-supervised pretraining: masked or partial tactile images are passed through the Hiera encoder and a ViT-style decoder [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: Proposed sensor design for TensorTouch. This modular layout shares [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Distribution of force and displacement measurements of the dataset. [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Visualization of multi-channel tactile sensing outputs generated by our calibrated model from a single tactile image (top right). The left panels show [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]
Figure 13
Figure 13. Figure 13: Evaluation result of the sensor. (a) Measured ATI force sensor output [PITH_FULL_IMAGE:figures/full_fig_p015_13.png]
Figure 14
Figure 14. Figure 14: Error distribution of the motion capture system across the three spatial [PITH_FULL_IMAGE:figures/full_fig_p017_14.png]
Figure 15
Figure 15. Figure 15: Position-dependent error analysis for the motion capture system. [PITH_FULL_IMAGE:figures/full_fig_p018_15.png]
Figure 17
Figure 17. Figure 17: Contact points (blue) in 3D space for a representative configuration [PITH_FULL_IMAGE:figures/full_fig_p018_17.png]
Figure 18
Figure 18. Figure 18: (Left) Mean manipulability p det(JJ⊤)  across different z-offsets (cm) and pitch angles (deg). (Right) The number of principal points (i.e., fingertip contact points inliers) for each (z, pitch) pair. thumb-index contacts and form our core dataset for each URDF varia…

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. 3D Cal: An Open-Source Software Library for Depth Reconstruction on Vision-Based Tactile Sensors

    cs.RO 2025-11 conditional novelty 6.0 of 10

    3D Cal repurposes a 3D printer as an automated calibration rig and trains a lightweight CNN, TouchNet, to reconstruct depth maps for DIGIT and GelSight Mini.

  2. Pose-Aware Modeling to Mitigate Pose-Related Artifacts in Tactile Gloves

    cs.RO 2026-07 conditional novelty 5.0 of 10

    Feeding index-finger joint angles into a tactile glove as a learned residual correction reduces pose-induced artifacts and lowers the minimum detectable force by 10-18%.

Reference graph

Works this paper leans on

67 extracted references · 37 canonical work pages · cited by 2 Pith papers

  1. [54]

    Influence of surface structure on friction and wear characteristics of silicone rubber for hydraulic rod seals,

    S.-J. Lee and C.-L. Kim, “Influence of surface structure on friction and wear characteristics of silicone rubber for hydraulic rod seals,”RSC advances, vol. 13, no. 48, pp. 33 595–33 602, 2023

    work page 2023

  2. [1]

    π 0: A vision–language–action flow model for general robot control,

    K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, L. Groom, K. Hausman, B. Ichteret al., “π 0: A vision–language–action flow model for general robot control,”arXiv preprint, 2024. [Online]. Available: https://arxiv.org/abs/2410.24164

    arXiv 2024

  3. [2]

    π 0.5: A vision–language–action model with open-world generalization,

    Physical Intelligence, K. Black, N. Brown, J. Darpinian, K. Dhabalia, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusaiet al., “π 0.5: A vision–language–action model with open-world generalization,”arXiv preprint, 2025. [Online]. Available: https://arxiv.org/abs/2504.16054

    arXiv 2025

  4. [3]

    Dexterous contact-rich ma- nipulation via the contact trust region,

    H. Suh, T. Pang, T. Zhao, and R. Tedrake, “Dexterous contact-rich ma- nipulation via the contact trust region,”arXiv preprint arXiv:2505.02291, 2025

    arXiv 2025

  5. [4]

    Pp-tac: Paper picking using tactile feedback in dexterous robotic hands,

    P. Lin, Y . Huang, W. Li, J. Ma, C. Xiao, and Z. Jiao, “Pp-tac: Paper picking using tactile feedback in dexterous robotic hands,”arXiv preprint arXiv:2504.16649, 2025

    arXiv 2025

  6. [5]

    Lessons from learning to spin “pens

    J. Wang, Y . Yuan, H. Che, H. Qi, Y . Ma, J. Malik, and X. Wang, “Lessons from learning to spin “pens”,” inCoRL, 2024

    2024

  7. [6]

    TacSL: A Library for Visuotactile Sensor Simulation and Learning,

    I. Akinola, J. Xu, J. Carius, D. Fox, and Y . Narang, “TacSL: A Library for Visuotactile Sensor Simulation and Learning,”IEEE Transactions on Robotics, pp. 1–17, 2025. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10912733

    work page arXiv 2025

  8. [7]

    PoCo: Policy Composition from and for Heterogeneous Robot Learning,

    L. Wang, J. Zhao, Y . Du, E. Adelson, and R. Tedrake, “PoCo: Policy Composition from and for Heterogeneous Robot Learning,” inRobotics: Science and Systems XX. Robotics: Science and Systems Foundation, Jul. 2024. [Online]. Available: http://www.roboticsproceedings.org/ rss20/p127.pdf

    work page 2024

Show all 67 references
  1. [8]

    General In-hand Object Rotation with Vision and Touch,

    H. Qi, B. Yi, S. Suresh, M. Lambeta, Y . Ma, R. Calandra, and J. Malik, “General In-hand Object Rotation with Vision and Touch,” inProceedings of The 7th Conference on Robot Learning. PMLR, Dec. 2023, pp. 2549–2564, iSSN: 2640-3498. [Online]. Available: https://proceedings.mlr...

    2023

  2. [9]

    Fine Manipulation Using a Tactile Skin: Learning in Simulation and Sim-to-Real Transfer,

    U. Kasolowsky and B. B ¨auml, “Fine Manipulation Using a Tactile Skin: Learning in Simulation and Sim-to-Real Transfer,” Sep. 2024, arXiv:2409.12735. [Online]. Available: http://arxiv.org/abs/2409.12735

    2024 arXiv

  3. [10]

    A Learning-based Controller for Multi-Contact Grasps on Unknown Objects with a Dexterous Hand,

    D. Winkelbauer, R. Triebel, and B. B ¨auml, “A Learning-based Controller for Multi-Contact Grasps on Unknown Objects with a Dexterous Hand,” Sep. 2024, arXiv:2409.12339. [Online]. Available: http://arxiv.org/abs/2409.12339

    2024 arXiv

  4. [11]

    Biomimetic tactile sensor array,

    N. Wettels, V . J. Santos, R. S. Johansson, and G. E. Loeb, “Biomimetic tactile sensor array,”Advanced robotics, vol. 22, no. 8, pp. 829–849, 2008

    2008

  5. [12]

    Active touch sensing,

    W. D. Hillis, “Active touch sensing,” Ph.D. dissertation, Massachusetts Institute of Technology, 1981

    1981

  6. [13]

    Location and Orientation Super- Resolution Sensing With a Cost-Efficient and Repairable Barometric Tactile Sensor,

    J. Hou, X. Zhou, and A. Spiers, “Location and Orientation Super- Resolution Sensing With a Cost-Efficient and Repairable Barometric Tactile Sensor,”IEEE Transactions on Robotics, vol. 41, pp. 729–741,

  7. [14]

    PX-6AX GEN2 Hall Effect Tactile Sen- sor,

    Paxini Technology Co., Ltd., “PX-6AX GEN2 Hall Effect Tactile Sen- sor,” https://paxini.com/ax/gen2, 2024, multi-dimensional tactile sensor based on hall effect technology capable of 15 types of tactile perceptions with 0.01N high-precision measurement

    2024

  8. [15]

    GelSight: High-Resolution Robot Tactile Sensors for Estimating Geometry and Force

    “GelSight: High-Resolution Robot Tactile Sensors for Estimating Geometry and Force.” [Online]. Available: https://www.mdpi.com/ 1424-8220/17/12/2762

  9. [16]

    GelSlim 3.0: High- Resolution Measurement of Shape, Force and Slip in a Compact Tactile-Sensing Finger,

    I. H. Taylor, S. Dong, and A. Rodriguez, “GelSlim 3.0: High- Resolution Measurement of Shape, Force and Slip in a Compact Tactile-Sensing Finger,” in2022 International Conference on Robotics and Automation (ICRA), May 2022, pp. 10 781–10 787. [Online]. Available: https://ieeex...

    2022

  10. [17]

    DIGIT: A Novel Design for a Low-Cost Compact High-Resolution Tactile Sensor With Application to In-Hand Manipulation,

    M. Lambeta, P.-W. Chou, S. Tian, B. Yang, B. Maloon, V . R. Most, D. Stroud, R. Santos, A. Byagowi, G. Kammerer, D. Jayaraman, and R. Calandra, “DIGIT: A Novel Design for a Low-Cost Compact High-Resolution Tactile Sensor With Application to In-Hand Manipulation,”IEEE Robotics ...

    2020

  11. [18]

    MagicTac: A Novel High-Resolution 3D Multi-layer Grid-Based Tactile Sensor,

    W. Fan, H. Li, and D. Zhang, “MagicTac: A Novel High-Resolution 3D Multi-layer Grid-Based Tactile Sensor,” in2024 IEEE International Conference on Robotics and Automation (ICRA), May 2024, pp. 388–

    2024

  12. [19]

    DelTact: A Vision-Based Tactile Sensor Using a Dense Color Pattern,

    G. Zhang, Y . Du, H. Yu, and M. Y . Wang, “DelTact: A Vision-Based Tactile Sensor Using a Dense Color Pattern,”IEEE Robotics and Automation Letters, vol. 7, no. 4, pp. 10 778–10 785, Oct. 2022. [Online]. Available: https://ieeexplore.ieee.org/document/9849124

    2022

  13. [20]

    Digitizing Touch with an Artificial Multimodal Fingertip,

    M. Lambeta, T. Wu, A. Sengul, V . R. Most, N. Black, K. Sawyer, R. Mercado, H. Qi, A. Sohn, B. Taylor, N. Tydingco, G. Kammerer, D. Stroud, J. Khatha, K. Jenkins, K. Most, N. Stein, R. Chavira, T. Craven-Bartle, E. Sanchez, Y . Ding, J. Malik, and R. Calandra, “Digitizing Touc...

    2024 arXiv

  14. [21]

    Densetact: Optical tactile sensor for dense shape reconstruction,

    W. K. Do and M. Kennedy, “Densetact: Optical tactile sensor for dense shape reconstruction,” in2022 International Conference on Robotics and Automation (ICRA). IEEE, 2022, pp. 6188–6194

    2022

  15. [22]

    The TacTip Family: Soft Optical Tactile Sensors with 3D-Printed Biomimetic Morphologies,

    B. Ward-Cherrier, N. Pestell, L. Cramphorn, B. Winstone, M. E. Giannaccini, J. Rossiter, and N. F. Lepora, “The TacTip Family: Soft Optical Tactile Sensors with 3D-Printed Biomimetic Morphologies,” Soft Robotics, vol. 5, no. 2, pp. 216–227, Apr. 2018, publisher: Mary Ann Liebe...

    2018

  16. [23]

    Omnitact: A multi-directional high-resolution touch sensor,

    A. Padmanabha, F. Ebert, S. Tian, R. Calandra, C. Finn, and S. Levine, “Omnitact: A multi-directional high-resolution touch sensor,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 618–624

    2020

  17. [24]

    GelSight Svelte: A Human Finger-Shaped Single-Camera Tactile Robot Finger with Large Sensing Coverage and Proprioceptive Sensing,

    J. Zhao and E. H. Adelson, “GelSight Svelte: A Human Finger-Shaped Single-Camera Tactile Robot Finger with Large Sensing Coverage and Proprioceptive Sensing,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Oct. 2023, pp. 8979– 8984, iSSN: 21...

    2023

  18. [25]

    Densetact-mini: An optical tactile sensor for grasping multi-scale objects from flat surfaces,

    W. K. Do, A. K. Dhawan, M. Kitzmann, and M. Kennedy III, “Densetact-mini: An optical tactile sensor for grasping multi-scale objects from flat surfaces,”arXiv preprint arXiv:2309.08860, 2023

    2023 arXiv

  19. [26]

    A soft thumb- sized vision-based sensor with accurate all-round force perception,

    H. Sun, K. J. Kuchenbecker, and G. Martius, “A soft thumb- sized vision-based sensor with accurate all-round force perception,” Nature Machine Intelligence, vol. 4, no. 2, pp. 135–145, Feb. 2022, publisher: Nature Publishing Group. [Online]. Available: https://www.nature.com/a...

    2022

  20. [27]

    Soft-bubble grippers for robust and perceptive manipulation,

    N. Kuppuswamy, A. Alspach, A. Uttamchandani, S. Creasey, T. Ikeda, and R. Tedrake, “Soft-bubble grippers for robust and perceptive manipulation,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Oct. 2020, pp. 9917– 9924, iSSN: 2153-0866. [Onl...

    2020

  21. [28]

    Texterity: Tactile extrinsic dexterity,

    A. Bronars, S. Kim, P. Patre, and A. Rodriguez, “Texterity: Tactile extrinsic dexterity,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 7976–7983

    2024

  22. [29]

    UniT: Data efficient tactile representation with generalization to unseen objects,

    Z. Xu, R. Uppuluri, X. Zhang, C. Fitch, P. G. Crandall, W. Shou, D. Wang, and Y . She, “UniT: Data efficient tactile representation with generalization to unseen objects,” 2025. [Online]. Available: https://arxiv.org/abs/2408.06481

    2025 arXiv

  23. [30]

    Self-supervised perception for tactile skin covered dexterous hands,

    A. Sharma, C. Higuera, C. K. Bodduluri, Z. Liu, T. Fan, T. Hellebrekers, M. Lambeta, B. Boots, M. Kaess, T. Wuet al., “Self-supervised perception for tactile skin covered dexterous hands,”arXiv preprint arXiv:2505.11420, 2025

    2025 arXiv

  24. [31]

    Diffusion policy: Visuomotor policy learning via ac- tion diffusion,

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song, “Diffusion policy: Visuomotor policy learning via ac- tion diffusion,”The International Journal of Robotics Research, p. 02783649241273668, 2023

    2023

  25. [32]

    Reactive diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation,

    H. Xue, J. Ren, W. Chen, G. Zhang, Y . Fang, G. Gu, H. Xu, and C. Lu, “Reactive diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation,”arXiv preprint arXiv:2503.02881, 2025

    2025 arXiv

  26. [33]

    Robot Tactile Sensing: Vision Based Tactile Sensor for Force Perception,

    T. Zhang, Y . Cong, X. Li, and Y . Peng, “Robot Tactile Sensing: Vision Based Tactile Sensor for Force Perception,” in2018 IEEE 8th Annual International Conference on CYBER Technology in Automation, Control, and Intelligent Systems (CYBER), Jul. 2018, pp. 1360– 1365, iSSN: 237...

    2018

  27. [34]

    In-Hand Object Localization Using a Novel High-Resolution Visuotactile Sensor,

    S. Cui, R. Wang, J. Hu, J. Wei, S. Wang, and Z. Lou, “In-Hand Object Localization Using a Novel High-Resolution Visuotactile Sensor,”IEEE Transactions on Industrial Electronics, vol. 69, no. 6, pp. 6015–6025, Jun. 2022. [Online]. Available: https: //ieeexplore.ieee.org/abstrac...

    2022

  28. [35]

    DTact: A Vision-Based Tactile Sensor that Measures High-Resolution 3D Geometry Directly from Darkness,

    C. Lin, Z. Lin, S. Wang, and H. Xu, “DTact: A Vision-Based Tactile Sensor that Measures High-Resolution 3D Geometry Directly from Darkness,” in2023 IEEE International Conference on Robotics and Automation (ICRA), May 2023, pp. 10 359–10 366. [Online]. Available: https://ieeexp...

    2023

  29. [36]

    Densetact 2.0: Optical tactile sensor for shape and force reconstruction,

    W. K. Do, B. Jurewicz, and M. Kennedy, “Densetact 2.0: Optical tactile sensor for shape and force reconstruction,” in2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 12 549– 12 555

    2023

  30. [37]

    Dense Tactile Force Estimation using GelSlim and inverse FEM,

    D. Ma, E. Donlon, S. Dong, and A. Rodriguez, “Dense Tactile Force Estimation using GelSlim and inverse FEM,” in2019 International Conference on Robotics and Automation (ICRA), May 2019, pp. 5418–5424, iSSN: 2577-087X. [Online]. Available: https://ieeexplore.ieee.org/abstract/d...

    2019

  31. [38]

    Sparsh: Self-supervised touch representations for vision- based tactile sensing,

    C. Higuera, A. Sharma, C. K. Bodduluri, T. Fan, P. Lancaster, M. Kalakrishnan, M. Kaess, B. Boots, M. Lambeta, T. Wu, and M. Mukadam, “Sparsh: Self-supervised touch representations for vision- based tactile sensing,” Oct. 2024, arXiv:2410.24090 [cs]. [Online]. Available: http:...

    2024 arXiv

  32. [39]

    Sim-to-real for robotic tactile sensing via physics-based simulation and learned latent projections,

    Y . Narang, B. Sundaralingam, M. Macklin, A. Mousavian, and D. Fox, “Sim-to-real for robotic tactile sensing via physics-based simulation and learned latent projections,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 6444–6451

    2021

  33. [40]

    3D Force and Contact Estimation for a Soft-Bubble Visuotactile Sensor Using FEM,

    J.-C. Peng, S. Yao, and K. Hauser, “3D Force and Contact Estimation for a Soft-Bubble Visuotactile Sensor Using FEM,” in 2024 IEEE International Conference on Robotics and Automation (ICRA), May 2024, pp. 5666–5672. [Online]. Available: https: //ieeexplore.ieee.org/abstract/do...

    2024

  34. [41]

    BiTac: A Soft Vision-Based Tactile Sensor With Bidirectional Force Perception for Robots,

    L. V . Duong, “BiTac: A Soft Vision-Based Tactile Sensor With Bidirectional Force Perception for Robots,”IEEE Sensors Journal, vol. 23, no. 9, pp. 9158–9167, May 2023. [Online]. Available: https://ieeexplore.ieee.org/document/10077553

    2023

  35. [42]

    iFEM2.0: Dense 3-D Contact Force Field Reconstruction and Assessment for Vision-Based Tactile Sensors,

    C. Zhao, J. Liu, and D. Ma, “iFEM2.0: Dense 3-D Contact Force Field Reconstruction and Assessment for Vision-Based Tactile Sensors,” IEEE Transactions on Robotics, vol. 41, pp. 289–305, 2025. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10758225

    2025

  36. [43]

    Hiera: A hierarchical vision transformer without the bells-and-whistles,

    C. Ryali, Y . Hu, D. Bolya, C. Wei, H. Fan, P. Huang, V . Aggarwal, A. Chowdhury, O. Poursaeed, J. Hoffman, J. Malik, Y . Li, and C. Feichtenhofer, “Hiera: A hierarchical vision transformer without the bells-and-whistles,” inInternational Conference on Machine Learning, ICML 2...

    2023

  37. [44]

    The effective- ness of mae pre-pretraining for billion-scale pretraining,

    M. Singh, Q. Duval, K. V . Alwala, H. Fan, V . Aggarwal, A. Adcock, A. Joulin, P. Doll´ar, C. Feichtenhofer, R. Girshicket al., “The effective- ness of mae pre-pretraining for billion-scale pretraining,” inProceedings of the IEEE/CVF international conference on computer vision...

    2023

  38. [45]

    Yale-cmu-berkeley dataset for robotic manipulation research,

    B. Calli, A. Singh, J. Bruce, A. Walsman, K. Konolige, S. Srinivasa, P. Abbeel, and A. M. Dollar, “Yale-cmu-berkeley dataset for robotic manipulation research,”The International Journal of Robotics Research, vol. 36, no. 3, pp. 261–268, 2017

    2017

  39. [46]

    On the theory of filter amplifiers,

    S. Butterworthet al., “On the theory of filter amplifiers,”Wireless Engineer, vol. 7, no. 6, pp. 536–541, 1930

    1930

  40. [47]

    The physics of rubber elasticity,

    L. Treload, “The physics of rubber elasticity,” 1975

    1975

  41. [48]

    Large elastic deformations of isotropic materials iv. further developments of the general theory,

    R. S. Rivlin, “Large elastic deformations of isotropic materials iv. further developments of the general theory,”Philosophical transactions of the royal society of London. Series A, Mathematical and physical sciences, vol. 241, no. 835, pp. 379–397, 1948

    1948

  42. [49]

    R. W. Ogden,Non-linear elastic deformations. Courier Corporation, 1997

    1997

  43. [50]

    Some forms of the strain energy function for rubber,

    O. H. Yeoh, “Some forms of the strain energy function for rubber,” Rubber Chemistry and technology, vol. 66, no. 5, pp. 754–771, 1993

    1993

  44. [51]

    Dense tactile force esti- mation using gelslim and inverse fem,

    D. Ma, E. Donlon, S. Dong, and A. Rodriguez, “Dense tactile force esti- mation using gelslim and inverse fem,” in2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 5418–5424

    2019

  45. [52]

    Measurement of the hyperelastic properties of ex vivo brain tissue slices,

    T. Kaster, I. Sack, and A. Samani, “Measurement of the hyperelastic properties of ex vivo brain tissue slices,”Journal of biomechanics, vol. 44, no. 6, pp. 1158–1163, 2011

    2011

  46. [53]

    [Online]

    Dassault Syst `emes Simulia Corp.,Abaqus/CAE User’s Manual, Version 2024, Dassault Syst `emes, Providence, RI, USA, 2024, software manual. [Online]. Available: https://www.3ds.com/products/simulia/abaqus/

    2024

  47. [55]

    Analysis of approximate nearest neighbor searching with clustered point sets,

    S. Maneewongvatana and D. M. Mount, “Analysis of approximate nearest neighbor searching with clustered point sets,”arXiv preprint cs/9901013, 1999

    1999 arXiv

  48. [56]

    Vision transformers for dense prediction,

    R. Ranftl, A. Bochkovskiy, and V . Koltun, “Vision transformers for dense prediction,” in2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021, Montreal, QC, Canada, October 10-17, 2021. IEEE, 2021, pp. 12 159–12 168. [Online]. Available: https://doi.org/10.1109...

    2021

  49. [57]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international con- ference, Munich, Germany, October 5-9, 2015, proceedings, part III 18. ...

    2015

  50. [58]

    Manipulability of robotic mechanisms,

    T. Yoshikawa, “Manipulability of robotic mechanisms,”The interna- tional journal of Robotics Research, vol. 4, no. 2, pp. 3–9, 1985

    1985

  51. [59]

    Coal: an extension of the flexible collision library,

    J. Pan, S. Chitta, D. Manocha, F. Lamiraux, J. Mirabel, J. Carpentier, L. Montautet al., “Coal: an extension of the flexible collision library,” https://github.com/coal-library/coal, 2015–2024

    2015

  52. [60]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” inProceedings of the IEEE confer- ence on computer vision and pattern recognition, 2017, pp. 4700–4708

    2017

  53. [61]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  54. [62]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,”ICLR, 2021

    2021

  55. [63]

    Vision transformers for dense prediction,

    R. Ranftl, A. Bochkovskiy, and V . Koltun, “Vision transformers for dense prediction,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 12 179–12 188

    2021

  56. [64]

    Masked autoencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Doll ´ar, and R. B. Girshick, “Masked autoencoders are scalable vision learners,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. IEEE, 2022, pp. 15 979–15 988. [Online]. Av...

    2022

  57. [65]

    Transferable tactile transformers for representation learning across diverse sensors and tasks,

    J. Zhao, Y . Ma, L. Wang, and E. H. Adelson, “Transferable tactile transformers for representation learning across diverse sensors and tasks,”arXiv preprint arXiv:2406.13640, 2024

    2024 arXiv

  58. [394]

    Available: https://ieeexplore.ieee.org/abstract/document/ 10610615

    [Online]. Available: https://ieeexplore.ieee.org/abstract/document/ 10610615

  59. [2025]

    Available: https://ieeexplore.ieee.org/abstract/document/ 10770605

    [Online]. Available: https://ieeexplore.ieee.org/abstract/document/ 10770605

Pith tools

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