Pith. sign in

REVIEW 4 major objections 3 minor 46 references

A low-cost 3D printer can be repurposed into an automated rig that calibrates vision-based tactile sensors to micron-level depth accuracy.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 23:59 UTC pith:LJ7DRCRF

load-bearing objection Useful open-source library and dataset for calibrating DIGIT/GelSight Mini with a cheap 3D printer, but the headline depth-error numbers are lower bounds computed after fitting ground-truth scale and alignment to predictions, and no baselines support the 'comparable to state-of-the-art' claim. the 4 major comments →

arxiv 2511.03078 v3 pith:LJ7DRCRF submitted 2025-11-04 cs.RO

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

Rohan Kota , Kaival Shah , J. Edward Colgate , Gregory Reardon This is my paper
classification cs.RO
keywords tactile sensingdepth reconstructionsensor calibrationvision-based tactile sensorsTouchNet3D printer probingDIGITGelSight Mini
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces 3D Cal, an open-source library that turns a low-cost 3D printer into an automated probing device for calibrating vision-based tactile sensors. The printer fabricates a socket that holds the sensor in a known location, then a spherical probe tip indents the gel in a regular 0.5 mm grid while images are captured and labeled automatically. The authors claim a lightweight convolutional network, TouchNet, trained only on these spherical-probe images, reconstructs depth maps of unseen objects with whole-surface average errors of roughly 16–52 µm and contact-region errors averaging about 156 µm (DIGIT) and 205 µm (GelSight Mini)—comparable to existing methods. They also report that reconstruction loss stabilizes when data from roughly 240 probe locations (20% of the grid) is used, providing a practical calibration recipe. If the approach holds, labs already owning a hobbyist 3D printer can skip expensive robot-arm or CNC calibration setups.

Core claim

3D Cal's central claim is that the labor-intensive calibration of vision-based tactile sensors can be fully automated with hardware most robotics labs already own: an FDM 3D printer. The printer first prints a rigid base that fixes the sensor in its coordinate frame; a 2 mm spherical probe mounted on the printhead indents the gel in a 0.5 mm × 0.5 mm grid, capturing 30 images at each of roughly 1,200 locations per sensor. These coordinate-labeled images train TouchNet, a fully convolutional network that predicts a surface gradient map which a fast Poisson integrator converts to a depth map. Trained solely on spherical indentations, the model reconstructs three unseen 3D-printed objects—hemis

What carries the argument

The central mechanism is the full 3D Cal pipeline: a G-code-driven FDM printer converted into a two-axis automatic probe; a 2 mm spherical probe tip that creates precisely known indentation geometry; and a dataset of images labeled with spatial coordinates. The learning model carrying the argument is TouchNet, a 9-layer fully convolutional network that takes an RGB image concatenated with a two-channel (x, y) coordinate embedding and outputs a surface gradient map (Gx, Gy); a fast Poisson integration reconstructs absolute depth. The coordinate embedding is what lets the model associate pixel responses with physical locations, a necessary step because illumination and response vary across the

Load-bearing premise

The load-bearing premise is that a network trained only on spherical probe indentations on a 0.5 mm grid generalizes to arbitrary unseen object geometries; the paper tests this on just three 3D-printed objects and already notes that shadowing breaks reconstruction at the pawn's neck.

What would settle it

Press a concave or undercut object of known geometry (for example a threaded screw or a ring) into DIGIT and GelSight Mini, compute the TouchNet depth map, and compare to ground truth. If contact-region errors rise far above the reported 156–205 µm range, the claim that spherical probes alone suffice for arbitrary geometry would be falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Researchers with a hobbyist 3D printer can generate thousands of coordinate-labeled tactile images in about two hours per sensor, removing the need for industrial robot arms, CNC machines, or motion-capture rigs.
  • The ablation study indicates that roughly 240 probe locations along a 0.5 mm grid are sufficient for stable reconstruction, so a new sensor can be calibrated in a single automated session.
  • Because TouchNet inference runs in under 30 ms, the pipeline can produce depth maps at 30 fps on modest laptop hardware, making closed-loop tactile manipulation practical.
  • Releasing the datasets and pre-trained DIGIT/GelSight Mini weights lets other labs fine-tune their own models instead of starting from scratch, accelerating adoption of depth-sensing touch.
  • If the generalization beyond the three test objects holds, vision-based tactile sensors can deliver physical geometry estimates without per-object retraining, supporting grasp verification, surface inspection, and shape reconstruction tasks.

Where Pith is reading between the lines

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

  • Extension: the same probe dataset could be reused to learn force, shear, or stress labels if a force-sensing probe replaces the rigid sphere, since the images already encode dense deformation information.
  • Extension: the strongest test of the spherical-probe assumption would be a concave or undercut object such as a threaded screw or ring; the pawn's shadowed neck already flags a failure region, so such tests could bound the generalization claim.
  • Extension: the 20% (≈240-point) guideline is calibrated to these two sensors; other gel sizes or optical setups may need grid density scaled by sensing area rather than copying the same fraction.
  • Extension: if the pre-trained weights transfer across sensors after fine-tuning, the pipeline effectively becomes a foundation-model approach for tactile depth estimation—a direction the paper names as future work but does not demonstrate.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 3 minor

Summary. The paper introduces 3D Cal, an open-source software library that repurposes a low-cost FDM 3D printer as an automated probing device for collecting labeled tactile-sensor data, and TouchNet, a fully convolutional network that maps RGB sensor images plus coordinate embeddings to surface gradients, which are integrated into depth maps. The authors calibrate DIGIT and GelSight Mini sensors, perform a spatial-sampling ablation to recommend a minimum number of probe locations, and report depth-reconstruction errors on three unseen 3D-printed objects. The headline results are average Type 2 errors of 156 μm for DIGIT and 205 μm for GelSight Mini, which the paper states are comparable to state-of-the-art methods.

Significance. If the reported accuracy is reliable, the paper would make a useful contribution to tactile-sensor calibration: it lowers the hardware barrier by using an inexpensive 3D printer, releases code, pre-trained weights, and a large dataset of over 70,000 probe images, and provides practical guidance on the amount of calibration data needed. The spatial ablation and the open-source infrastructure are valuable regardless of the exact error numbers. However, the central accuracy claim is currently not supported by the evaluation protocol, and no baseline method is benchmarked, so the paper's main quantitative contribution needs substantial revision.

major comments (4)
  1. [Section IV, Table I] The reported depth errors are computed after fitting the ground-truth indentation depth to minimize the MSE against the predicted depth maps: 'the indentation depths of the ground-truth CAD models were adjusted to minimize the mean squared error between the predicted and ground-truth depth maps.' This is a per-object global depth-scale fit. Any systematic under- or over-estimation of indentation depth is therefore removed before errors are reported. The abstract's 156 μm and 205 μm figures are the averages of the Type 2 errors in Table I, not the Overall Errors, and thus inherit this favorable protocol. These numbers are not unbiased estimates of absolute depth-reconstruction accuracy; they are at best lower bounds. Please report errors without the per-object depth fit, or fix/measure the indentation depth independently, and report the fitted scale factors so readers can assess the bias.
  2. [Section IV and Abstract] The claim of performance 'comparable to state-of-the-art methods' is unsupported because no state-of-the-art depth-reconstruction method is implemented, run, or compared anywhere in the paper. There is no baseline such as GelSight's classical photometric-stereo pipeline, other supervised tactile depth estimators, or a re-implementation of a published DIGIT/GelSight Mini model. Please include at least one baseline on the same test objects under the same (or clearly stated) evaluation protocol, or remove the comparative claim.
  3. [Section III–IV] The generalization claim is based on training with only spherical probe indentations and evaluating on only three 3D-printed objects (hemispheres, pill, pawn), all with relatively smooth geometry. The authors themselves note that both sensors struggle to reconstruct the pawn's neck due to shadowing, showing the spherical-probe training already fails for a non-trivial concavity. The Discussion's statement that the model generates 'high-quality depth reconstructions on unseen and complex objects' overstates the evidence. A broader test set with varied curvature, sharp features, and concavities, or a more qualified claim, is needed.
  4. [Section IV, Table I] The protocol also performs a 2D cross-correlation to align the predicted and ground-truth depth maps in the xy-plane. This is reasonable for manual indentation, but combined with the depth-scale fit it removes both translational and depth-scale errors before accuracy is reported. This means the reported Overall, Type 1, and Type 2 errors all characterize residual shape error after a three-degree-of-freedom per-object alignment, not the raw output of the system. The paper should clearly state this and, ideally, provide the errors before alignment and scaling for at least one set of results.
minor comments (3)
  1. [Section IV] The definitions of Type 1 and Type 2 errors are given in the text but would benefit from explicit equations. It would also help to state that the abstract's 156/205 μm figures are the across-object averages of the Type 2 errors, not the Overall Errors in Table I.
  2. [Section III] The spatial ablation evaluates models on held-out probe coordinates from the same spherical-probe grid, so it measures interpolation within the calibration manifold, not generalization to unseen object geometries. This should be stated explicitly when presenting the '20% of coordinates' recommendation.
  3. [Throughout] Some figures lack axis units or have clipped colorbars; for example, Figures 2 and 3 would benefit from clearer labels on the MSE values. The manuscript also repeatedly says 'high-resolution' depth maps without quantifying resolution; reporting the pixel pitch or effective spatial resolution of the output maps would be useful.

Circularity Check

1 steps flagged

Headline error numbers are computed after fitting ground-truth depth scale to the predictions, so the 156/205 μm figures are partly self-consistency rather than unbiased reconstruction errors.

specific steps
  1. fitted input called prediction [Section IV (Performance on Unseen Objects), Table I; abstract]
    "Because the indentations were performed manually, the ground-truth depth maps were spatially aligned along the xy-plane using a 2D cross-correlation. To further account for slight variations in indentation depth, the indentation depths of the ground-truth CAD models were adjusted to minimize the mean squared error between the predicted and ground-truth depth maps."

    The abstract's headline errors (156 μm DIGIT, 205 μm GelSight Mini) are the cross-object averages of the Type 2 errors in Table I. Those Type 2 errors are computed after a per-object adjustment of the ground-truth CAD depth to minimize MSE with the model's predicted depth map. This removes any systematic depth-scale/indentation-depth bias before the error is measured, so the reported figures quantify shape fidelity relative to a best-fit depth scale rather than unbiased absolute reconstruction error. The 'prediction' of average reconstruction error is therefore partly forced by the fitting step, not by the model alone.

full rationale

3D Cal's core contribution—repurposing a low-cost 3D printer for automated probe data collection, releasing the library and data, and the data-ablation study—is a real engineering contribution and is not circular: the training/validation split and the held-out-coordinate MSE curves are computed from model outputs without retrofitting labels. The circularity is confined to the unseen-object benchmark that carries the abstract's headline accuracy claim. Section IV states that after manual indentation, the ground-truth CAD depth maps were aligned in xy by cross-correlation and then 'the indentation depths of the ground-truth CAD models were adjusted to minimize the mean squared error between the predicted and ground-truth depth maps.' The Type 2 error values in Table I, whose cross-object averages (156 μm DIGIT, 205 μm GelSight Mini) are quoted in the abstract as 'average reconstruction errors,' are therefore computed against a ground truth that has been fit to the predictions for the depth-scale component. This makes the quoted absolute accuracy figures lower bounds that include a self-consistency term. The cross-section plots still show genuine shape agreement, and the spatial-ablation findings are unaffected, so the circularity is partial rather than total. The 'comparable to state-of-the-art methods' claim is not demonstrated because no baseline is run, but that is an evaluation gap, not circularity. Score 6 reflects that the paper's headline quantitative claim partially reduces by construction, while the library itself and the data-volume guidelines stand on independent evidence.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

The central quantitative claims rest on empirical calibration rather than new physical theory. The largest hidden adjustments are the per-object depth scaling and xy alignment used in the benchmark, which fit ground truth to predictions. The key domain bet is that spherical-probe training transfers to arbitrary object geometries.

free parameters (2)
  • per-object indentation depth scale = not reported; optimized per test object to minimize MSE
    Section IV: 'the indentation depths of the ground-truth CAD models were adjusted to minimize the mean squared error between the predicted and ground-truth depth maps.' This fitted scale directly affects every reported error.
  • xy alignment shift = per test object, not reported
    Section IV: ground-truth and predicted depth maps are aligned via 2D cross-correlation before computing errors, removing translation error from the metric.
axioms (5)
  • domain assumption Spherical-probe training generalizes to arbitrary unseen object geometries.
    All training data comes from spherical indentations; the central capability claim requires transfer to the shapes in Section IV.
  • domain assumption Manual indentation of test objects, followed by fitted alignment and depth scaling, yields valid ground truth.
    Section IV evaluates against CAD models after manually indenting objects and fitting the depth scale; this presupposes the manual depth can be recovered by fitting.
  • domain assumption The 3D printer's G-code positioning is accurate enough for micrometer-scale tactile calibration.
    Section II-A assumes the printer's coordinate frame implicitly defines the sensor position, with no separate metrology step.
  • domain assumption The sensor surface is planar or has small radius of curvature.
    Section II-A states this as a requirement for 3D Cal, limiting the library's scope.
  • domain assumption The predicted gradient field is integrable by a fast Poisson solver.
    Section II-B converts gradient outputs to depth via Poisson integration; this assumes the gradients are consistent enough for a unique depth solution.

pith-pipeline@v1.3.0-alltime-deepseek · 11574 in / 9319 out tokens · 81752 ms · 2026-08-03T23:59:04.664439+00:00 · methodology

0 comments
read the original abstract

Tactile sensing plays a key role in enabling dexterous and reliable robotic manipulation, but realizing this capability requires substantial calibration to convert raw sensor readings into physically meaningful quantities. Despite its near-universal necessity, the calibration process remains ad hoc and labor-intensive. Here, we introduce 3D Cal, an open-source library that transforms a low-cost 3D printer into an automated probing device capable of generating large volumes of labeled training data for calibrating vision-based tactile sensors. 3D Cal also provides an end-to-end, user-friendly pipeline for training custom convolutional networks to produce high-quality depth reconstructions. Using 3D Cal, we systematically explore the relationship between training data volume and spatial reconstruction performance on two commercially available sensors, DIGIT and GelSight Mini, and derive practical, empirically-grounded guidelines for calibrating these sensors. Finally, we demonstrate depth reconstruction performance on the DIGIT and GelSight Mini comparable to state-of-the-art methods, achieving average reconstruction errors of 156 $\mathrm{\mu m}$ and 205 $\mathrm{\mu m}$ on unseen objects, respectively. By automating tactile sensor calibration, 3D Cal can accelerate tactile sensing research, simplify sensor deployment, and facilitate the integration of tactile sensing in robotic platforms.

Figures

Figures reproduced from arXiv: 2511.03078 by Gregory Reardon, J. Edward Colgate, Kaival Shah, Rohan Kota.

Figure 1
Figure 1. Figure 1: Overview of 3D Cal library. (A) Users first 3D print a sensor base on the print bed. (B) The tactile sensor is then inserted into the base with a slide fit. (C) Next, the 3D printer nozzle is affixed with a probe tip. (D) An automated program then uses the 3D printer to probe the sensor and collect labeled calibration data. (E) The calibration data is then used to train or fine-tune a machine learning mode… view at source ↗
Figure 2
Figure 2. Figure 2: Results of training data ablation study. (A) Sensors were probed along an m × n grid with d = 0.5 mm spacing. 20% of the coordinates were reserved for validation (red dots) while the remaining coordinates were used for training (black dots). Models were trained on P = 80%, 40%, 20%, 10%, 5%, and 1% of the total coordinates. (B) Model loss when trained on different percentages, P, of the probed coordinates.… view at source ↗
Figure 3
Figure 3. Figure 3: Spatial distribution of reconstruction accuracy. (A) The mean squared error (MSE) of the output gradients for coordinates in the validation set (shaded circles) using TouchNet models trained on different percentages, P, of the probed coordinates. The MSE tended to be higher in regions with fewer training coordinates (red X’s). (B) Kernel density estimates (KDE) of the MSE values (bin width: 0.0015, dashed … view at source ↗
Figure 4
Figure 4. Figure 4: Reconstructed depth maps for DIGIT and GelSight Mini using 3D Cal. (A) Three 3D printed test objects (hemispheres, pill, pawn), all measuring 10 mm × 10 mm, were pushed into the DIGIT and GelSight Mini. The corresponding RGB images on the DIGIT (bottom left) and GelSight Mini (bottom right) are shown below each test object. (B) The TouchNet models trained on P = 80% of the probed coordinates were used pred… view at source ↗
Figure 5
Figure 5. Figure 5: Distributions of pixelwise depth map errors. (A) Violin plots of the pixelwise depth map error distributions where the ground￾truth depth was equal to zero (Type 1 Error). Type 1 error distributions were truncated at the 95th percentile for clarity (DIGIT: red, GelSight Mini: blue). (B) Violin plots of the pixelwise depth map error distributions where the ground-truth depth was non-zero (Type 2 Error). bel… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

46 extracted references · 9 linked inside Pith

  1. [1]

    Flexible tactile sensing based on piezoresistive composites: A review,

    S. Stassi, V . Cauda, G. Canavese, and C. F. Pirri, “Flexible tactile sensing based on piezoresistive composites: A review,”Sensors, vol. 14, no. 3, pp. 5296–5332, 2014

    2014

  2. [2]

    ReSkin: Versa- tile, replaceable, lasting tactile skins,

    R. Bhirangi, T. Hellebrekers, C. Majidi, and A. Gupta, “ReSkin: Versa- tile, replaceable, lasting tactile skins,” inConference on Robot Learning (CoRL), 2021

    2021

  3. [3]

    Retrographic sensing for the measurement of surface texture and shape,

    M. K. Johnson and E. H. Adelson, “Retrographic sensing for the measurement of surface texture and shape,” inIEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 1070–1077

    2009

  4. [4]

    Microgeometry capture using an elastomeric sensor,

    M. K. Johnson, F. Cole, A. Raj, and E. H. Adelson, “Microgeometry capture using an elastomeric sensor,”ACM Transactions on Graphics, vol. 30, no. 4, 2011

    2011

  5. [5]

    GelSight: High-resolution robot tactile sensors for estimating geometry and force,

    W. Yuan, S. Dong, and E. H. Adelson, “GelSight: High-resolution robot tactile sensors for estimating geometry and force,”Sensors, vol. 17, no. 12, 2017

    2017

  6. [6]

    The feeling of success: Does touch sensing help predict grasp outcomes?

    R. Calandra, A. Owens, M. Upadhyaya, W. Yuan, J. Lin, E. Adelson, and S. Levine, “The feeling of success: Does touch sensing help predict grasp outcomes?” inConference on Robot Learning (CoRL), Oct. 2017

    2017

  7. [7]

    Combining finger vision and optical tactile sensing: Reducing and handling errors while cutting vegetables,

    A. Yamaguchi and C. G. Atkeson, “Combining finger vision and optical tactile sensing: Reducing and handling errors while cutting vegetables,” in2016 IEEE-RAS 16th International Conference on Humanoid Robots, 2016, pp. 1045–1051

    2016

  8. [8]

    Learning height for top-down grasps with the digit sensor,

    T. Bernardi, Y . Fleytoux, J.-B. Mouret, and S. Ivaldi, “Learning height for top-down grasps with the digit sensor,” in2023 IEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 1737–1743

    2023

  9. [9]

    Towards learning to play piano with dexterous hands and touch,

    H. Xu, Y . Luo, S. Wang, T. Darrell, and R. Calandra, “Towards learning to play piano with dexterous hands and touch,” in2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022, pp. 10 410–10 416

    2022

  10. [10]

    Patchgraph: In- hand tactile tracking with learned surface normals,

    P. Sodhi, M. Kaess, M. Mukadam, and S. Anderson, “Patchgraph: In- hand tactile tracking with learned surface normals,” in2022 Interna- tional Conference on Robotics and Automation (ICRA), 2022, pp. 2164– 2170

    2022

  11. [11]

    Low-cost teleoperation with haptic feedback through vision-based tactile sensors for rigid and soft object manipulation,

    M. Lippi, M. C. Welle, M. K. Wozniak, A. Gasparri, and D. Kragic, “Low-cost teleoperation with haptic feedback through vision-based tactile sensors for rigid and soft object manipulation,” in2024 33rd IEEE International Conference on Robot and Human Interactive Com- munication (ROMAN), 2024, pp. 1963–1969

    2024

  12. [12]

    Leveraging tactile sensing to render both haptic feedback and virtual reality 3D object reconstruction in robotic telemanipulation,

    G. Giudici, A. A. Bonzini, C. Coppola, K. Althoefer, I. Farkhatdinov, and L. Jamone, “Leveraging tactile sensing to render both haptic feedback and virtual reality 3D object reconstruction in robotic telemanipulation,” arXiv preprint arXiv:2412.02644, 2024

    Pith/arXiv arXiv 2024

  13. [13]

    Tactile sensing systems for tumor characterization: A review,

    C.-H. Won, J.-H. Lee, and F. Saleheen, “Tactile sensing systems for tumor characterization: A review,”IEEE Sensors Journal, vol. 21, no. 11, pp. 12 578–12 588, 2021

    2021

  14. [14]

    High-resolution tactile-sensation diagnostic imaging system for thyroid cancer,

    S.-H. Cho, S.-M. Lee, N.-Y . Lee, B. C. Ko, H. Kim, D.-J. Jang, and J.-H. Lee, “High-resolution tactile-sensation diagnostic imaging system for thyroid cancer,”Sensors, vol. 23, no. 7, 2023

    2023

  15. [15]

    Assessing fruit hardness in robot hands using electric gripper actuators with tactile sensors,

    S. Li, W. Sun, Q. Liang, C. Liu, and J. Liu, “Assessing fruit hardness in robot hands using electric gripper actuators with tactile sensors,”Sensors and Actuators A: Physical, vol. 365, p. 114843, 2024

    2024

  16. [16]

    De- formation capture via soft and stretchable sensor arrays,

    O. Glauser, D. Panozzo, O. Hilliges, and O. Sorkine-Hornung, “De- formation capture via soft and stretchable sensor arrays,”ACM Trans. Graph., vol. 38, no. 2, Mar. 2019

    2019

  17. [17]

    AnySkin: Plug-and-play skin sensing for robotic touch,

    R. Bhirangi, V . Pattabiraman, E. Erciyes, Y . Cao, T. Hellebrekers, and L. Pinto, “AnySkin: Plug-and-play skin sensing for robotic touch,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 16 563–16 570

    2025

  18. [18]

    Acoustac: Tactile sensing with acoustic resonance for electronics-free soft skin,

    M. S. Li and H. S. Stuart, “Acoustac: Tactile sensing with acoustic resonance for electronics-free soft skin,”Soft Robotics, vol. 12, no. 1, pp. 109–123, 2025

    2025

  19. [19]

    DIGIT: A novel design for a low-cost compact high- resolution tactile sensor with application to in-hand manipulation,

    M. Lambeta, G. Kammerer, D. Jayaraman, R. Calandra, P.-W. Chou, S. Tian, B. Yang, B. Maloon, V . Most, D. Stroud, R. Santos, and A. Byagowi, “DIGIT: A novel design for a low-cost compact high- resolution tactile sensor with application to in-hand manipulation,”IEEE Robotics and Automation Letters, vol. 5, no. 3, pp. 3838–3845, Feb. 2020

    2020

  20. [20]

    GelSlim 4.0: Focusing on touch and reproducibility,

    A. Sipos, W. van den Bogert, and N. Fazeli, “GelSlim 4.0: Focusing on touch and reproducibility,”arXiv preprint arXiv:2409.19770, 2024

    Pith/arXiv arXiv 2024

  21. [21]

    9DTact: A compact vision-based tactile sensor for accurate 3D shape reconstruction and generalizable 6D force estimation,

    C. Lin, H. Zhang, J. Xu, L. Wu, and H. Xu, “9DTact: A compact vision-based tactile sensor for accurate 3D shape reconstruction and generalizable 6D force estimation,”arXiv preprint arXiv:2308.14277, 2023

    Pith/arXiv arXiv 2023

  22. [22]

    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. Tayloret al., “Digitizing touch with an artificial multimodal fingertip,”arXiv preprint arXiv:2411.02479, 2024

    Pith/arXiv arXiv 2024

  23. [23]

    eFlesh: Highly customizable magnetic touch sensing using cut-cell microstructures,

    V . Pattabiraman, Z. Huang, D. Panozzo, D. Zorin, L. Pinto, and R. Bhirangi, “eFlesh: Highly customizable magnetic touch sensing using cut-cell microstructures,”arXiv preprint arXiv:2506.09994, 2025

    Pith/arXiv arXiv 2025

  24. [24]

    The TacTip family: Soft optical tactile sensors with 3D-printed biomimetic morphologies,

    B. Ward-Cherrier, N. Pestell, L. Cramphorn, B. Winstone, M. Giannac- cini, J. Rossiter, and N. Lepora, “The TacTip family: Soft optical tactile sensors with 3D-printed biomimetic morphologies,”Soft Robotics, vol. 5, Jan. 2018

    2018

  25. [25]

    PyTouch: A machine learning library for touch processing,

    M. Lambeta, H. Xu, J. Xu, P.-W. Chou, S. Wang, T. Darrell, and R. Ca- landra, “PyTouch: A machine learning library for touch processing,” in2021 IEEE International Conference on Robotics and Automation (ICRA), 2021, pp. 13 208–13 214

    2021

  26. [26]

    Tacto: A fast, flexible, and open-source simulator for high-resolution vision-based tactile sensors,

    S. Wang, M. Lambeta, P.-W. Chou, and R. Calandra, “Tacto: A fast, flexible, and open-source simulator for high-resolution vision-based tactile sensors,”IEEE Robotics and Automation Letters, vol. 7, no. 2, p. 3930–3937, Apr. 2022

    2022

  27. [27]

    3D-ViTac: Learning fine-grained manipulation with visuo-tactile sensing,

    B. Huang, Y . Wang, X. Yang, Y . Luo, and Y . Li, “3D-ViTac: Learning fine-grained manipulation with visuo-tactile sensing,” inConference on Robot Learning (CoRL), 2024

    2024

  28. [28]

    Mod- eling and calibration of a tactile sensor for robust grasping,

    A. Cirillo, P. Cirillo, G. D. Maria, C. Natale, and S. Pirozzi, “Mod- eling and calibration of a tactile sensor for robust grasping,”IF AC- PapersOnLine, vol. 50, no. 1, pp. 6843–6850, 2017

    2017

  29. [29]

    Tactile image sensors employing camera: A review,

    K. Shimonomura, “Tactile image sensors employing camera: A review,” Sensors, vol. 19, no. 18, 2019

    2019

  30. [30]

    FeelAnyForce: Estimating contact force feedback from tactile sensation for vision-based tactile sensors,

    A.-H. Shahidzadeh, G. M. Caddeo, K. Alapati, L. Natale, C. Ferm ¨uler, and Y . Aloimonos, “FeelAnyForce: Estimating contact force feedback from tactile sensation for vision-based tactile sensors,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 251–257

    2025

  31. [31]

    TensorTouch: Calibration of tactile sensors for high resolution stress tensor and defor- mation for dexterous manipulation,

    W. K. Do, M. Strong, A. Swann, B. Lei, and M. K. III, “TensorTouch: Calibration of tactile sensors for high resolution stress tensor and defor- mation for dexterous manipulation,”arXiv preprint arXiv:2506.08291, 2025

    Pith/arXiv arXiv 2025

  32. [32]

    GelSight Wedge: Measuring high-resolution 3D contact geometry with a compact robot finger,

    S. Wang, Y . She, B. Romero, and E. Adelson, “GelSight Wedge: Measuring high-resolution 3D contact geometry with a compact robot finger,” inIEEE International Conference on Robotics and Automation (ICRA), May 2021, pp. 6468–6475

    2021

  33. [33]

    Improved GelSight tactile sensor for measuring geometry and slip,

    S. Dong, W. Yuan, and E. H. Adelson, “Improved GelSight tactile sensor for measuring geometry and slip,” in2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017, pp. 137– 144

    2017

  34. [34]

    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), 2022, pp. 6188–6194

    2022

  35. [35]

    Classification of vision-based tactile sensors: A review,

    H. Li, Y . Lin, C. Lu, M. Yang, E. Psomopoulou, and N. F. Lepora, “Classification of vision-based tactile sensors: A review,”IEEE Sensors Journal, 2025

    2025

  36. [36]

    Tactile robotics: An outlook,

    S. Luo, N. F. Lepora, W. Yuan, K. Althoefer, G. Cheng, and R. Dahiya, “Tactile robotics: An outlook,”IEEE Transactions on Robotics, 2025

    2025

  37. [37]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,”Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, 1998

    1998

  38. [38]

    An intriguing failing of convolutional neural networks and the CoordConv solution,

    R. Liu, J. Lehman, P. Molino, F. Petroski Such, E. Frank, A. Sergeev, and J. Yosinski, “An intriguing failing of convolutional neural networks and the CoordConv solution,”Advances in Neural Information Processing Systems, vol. 31, 2018

    2018

  39. [39]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift,

    S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” inProceedings of the 32nd International Conference on Machine Learning, vol. 37, Jul. 2015, pp. 448–456

    2015

  40. [40]

    Deep learning using rectified linear units (ReLU),

    A. F. Agarap, “Deep learning using rectified linear units (ReLU),”arXiv preprint arXiv:1803.08375, 2019

    Pith/arXiv arXiv 2019

  41. [41]

    Efficient object localization using convolutional networks,

    J. Tompson, R. Goroshin, A. Jain, Y . LeCun, and C. Bregler, “Efficient object localization using convolutional networks,” pp. 648–656, 2015

    2015

  42. [42]

    Touch2Touch: Cross-modal tactile generation for object manipulation,

    S. Rodriguez, Y . Dou, M. Oller, A. Owens, and N. Fazeli, “Touch2Touch: Cross-modal tactile generation for object manipulation,” arXiv preprint arXiv:2409.08269, 2024

    Pith/arXiv arXiv 2024

  43. [43]

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

    C. Higuera, A. Sharma, C. K. Bodduluri, T. Fan, P. Lancaster, M. Kalakr- ishnan, M. Kaess, B. Boots, M. Lambeta, T. Wuet al., “Sparsh: Self- supervised touch representations for vision-based tactile sensing,”arXiv preprint arXiv:2410.24090, 2024

    Pith/arXiv arXiv 2024

  44. [44]

    Midas- touch: Monte-carlo inference over distributions across sliding touch,

    S. Suresh, Z. Si, S. Anderson, M. Kaess, and M. Mukadam, “Midas- touch: Monte-carlo inference over distributions across sliding touch,” in Conference on Robot Learning (CoRL), 2023, pp. 319–331

    2023

  45. [45]

    Objectfolder 2.0: A multisensory object dataset for sim2real transfer,

    R. Gao, Z. Si, Y .-Y . Chang, S. Clarke, J. Bohg, L. Fei-Fei, W. Yuan, and J. Wu, “Objectfolder 2.0: A multisensory object dataset for sim2real transfer,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 10 598–10 608

    2022

  46. [46]

    Touch and go: Learning from human-collected vision and touch,

    F. Yang, C. Ma, J. Zhang, J. Zhu, W. Yuan, and A. Owens, “Touch and go: Learning from human-collected vision and touch,” inProceedings of the 36th International Conference on Neural Information Processing Systems, 2022

    2022