Pith. sign in

REVIEW 3 major objections 5 minor 45 references

Pair each mechanical pin on a soft robot skin with an optical fiber, and touch information can be carried to a remote camera where simple image moments—no deep learning—reveal where, how big, and what shape the contact is, down to sub-milli

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

2026-08-01 22:26 UTC pith:23XBJRZF

load-bearing objection A real new sensor idea—aligned one-to-one pin-to-fiber skin—with localization results that hold up; the width model is a calibration fit and needs held-out validation, but the paper is worth refereeing. the 3 major comments →

arxiv 2607.15746 v1 pith:23XBJRZF submitted 2026-07-17 cs.RO

Towards Artificial Nerves: Biomimetic Optical-Fiber Tactile Sensing for Robots

Laura E. Butcher , Chris J. Ford , Nathan F. Lepora , Efi Psomopoulou This is my paper
classification cs.RO
keywords biomimetic tactile sensingoptical fiber arrayimage momentsHu momentscontact localizationshape classificationsoft roboticshyperacuity
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.

This paper claims that a one-to-one pairing of mechanical skin pins with optical fibers creates an artificial-nerve architecture for robotic touch: each fiber acts like a nerve carrying local tactile information away from the skin to a separate camera. Because the physical structure produces clean, localized intensity changes, simple image-moment analysis suffices to locate contacts with about 0.4 mm error, estimate contact width, and classify contact shapes with at least 0.90 F1-score. The authors argue this offers an interpretable, scalable alternative to deep-learning black boxes for distributed tactile sensing.

Core claim

The central discovery is that biomimetic physical pre-processing—one pin per optical fiber, arranged in a coherent hexagonal array—structures tactile signals so that contacts appear as localized negative intensity dips. From these, raw image moments give the centroid (calibrated to sub-millimeter RMSE), central-moment eigenvalues give contact width via a Gaussian-profile model, and Hu moments M1–M3 fed to a Gaussian Mixture Model classify an edge and four flat shapes with 96% average accuracy. This demonstrates that a sensor can transmit tactile information away from the skin, preserve high spatial resolution, and support interpretable inference without end-to-end learning.

What carries the argument

The key machinery is the coherent optical-fiber array: 217 plastic fibers (1 mm diameter, 1.5 mm spacing in a hexagonal lattice) run from an acrylic window behind the soft skin to a remote camera. Each fiber is aligned with one skin pin, so pin levering changes the fiber's light output. The raw camera frames are segmented, per-fiber intensities are averaged, and cubic interpolation creates high-resolution surface plots. Image moments computed on these plots—first moments for centroid, central moments for width, Hu invariants for shape—carry the inference, with a k-d tree plus Gaussian-filter calibration surface correcting spatial errors.

Load-bearing premise

The coherent optical-fiber array keeps its one-to-one spatial correspondence and signal quality when fibers are longer, bent, or routed through moving joints; the paper only tested short (95 mm), straight, fixed fibers under controlled laboratory conditions, and the Discussion explicitly calls for further engineering studies to quantify this robustness.

What would settle it

Take the fiber array, bend it to a radius of a few centimeters or extend it to a meter, then press a 5 mm indenter at known positions and measure the calibrated centroid RMSE; if it grows beyond the 1.5 mm taxel spacing, the distributed-sensing claim collapses. Separately, indent with a large flat square to create a distinctly non-Gaussian contact profile and check whether the width model's linear fit (R2 0.95–0.97) systematically breaks down, which would falsify the Gaussian assumption in Eq. 6.

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

If this is right

  • Tactile sensing hardware can be physically separated from the imaging module, allowing larger or distributed skin areas to share a single camera.
  • Sub-millimeter contact localization is achievable with a transparent analytical pipeline, matching or approaching the accuracy of deep-learning optical-fiber sensors.
  • Contact width can be estimated with errors below the 1.5 mm taxel spacing, especially for small contacts under 10 mm diameter.
  • Edges and flat shapes (circle, square, triangle, ellipse) are classifiable with a minimum F1-score of 0.90 using just three Hu moments and an unsupervised Gaussian Mixture Model.
  • The sensor exhibits hyperacuity: contact centroids are resolved more than three times finer than the spacing between adjacent pin-fiber pairs.

Where Pith is reading between the lines

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

  • If the coherent one-to-one mapping degrades when fibers are lengthened or bent (e.g., routed through a robot joint), the distributed-sensing promise would require per-configuration recalibration or coherent bundle maintenance; this is directly testable.
  • The width model's Gaussian assumption (Eq. 6) is a simplification; real contact profiles may be non-Gaussian, and the calibrated offsets (3.43 mm, 2.47 mm) and scales (2.35, 2.51) might not transfer to other skin geometries without retuning.
  • Because tactile data are treated as images, the method could likely be extended to arbitrary object shapes by porting richer vision descriptors, or to force and shear estimation, which this paper does not address.
  • The modular separation of skin and camera suggests a single camera could serve multiple skin patches, effectively distributing touch coverage; whether light budget and fiber packaging allow this is an open engineering question.

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

3 major / 5 minor

Summary. This manuscript presents OptiTac, a soft optical tactile sensor that pairs each skin pin with a plastic optical fiber in a coherent fiber array, transmitting tactile information to a remote camera. The authors compare sparse, dense, and aligned pin patterns, adopt the aligned one-to-one pattern, and invert the resulting intensity 'images' to infer contact centroid, width, and shape using image moments, central moments, and Hu moments with a Gaussian mixture model. They report calibration-improved contact localization RMSE ≈0.4 mm on an unseen dataset (hyperacuity), contact-width estimates with R² ≈0.95–0.97 after applying per-axis offsets and scales, and shape classification with a minimum F1 of 0.90. The paper frames this as a biomimetic artificial-nerve architecture for distributed tactile sensing.

Significance. If the quantitative claims hold, the paper makes a useful architectural contribution: physical structured pre-processing at the sensor level enables simple, interpretable analytical inference without deep learning, and the coherent optical-fiber transmission with sub-millimeter localization is a valuable proof-of-concept. The strongest evidence is the localization calibration, which is tested on a separately collected 'unseen' dataset and is clearly reported. The authors also state limitations honestly, particularly regarding long-fiber routing and packaging. However, two of the three headline capabilities—contact width and shape classification—are currently supported by evaluations that appear to be in-sample or are not described as held-out, so the quantitative claims need tightening before the architecture can be considered validated.

major comments (3)
  1. [§2.4, §5.3.2] Contact-width estimation is reported as a validated model (R²=0.95/0.97, RMSE=1.2/1.0 mm), but the pipeline fits per-axis offsets (3.43, 2.47 mm) and scale factors (2.35, 2.51) to the same diameter series on which accuracy is then reported, after assuming a Gaussian intensity profile (Eq. 6). The Gaussian assumption is not independently checked; if the actual intensity profile is top-hat, saturated, or threshold-dependent, the fitted constants can absorb an arbitrary monotonic mapping and the reported RMSE measures the calibration fit rather than predictive accuracy. Because 'measuring contact size' is a headline capability, this circularity is load-bearing. Please provide a leave-one-diameter-out or otherwise held-out validation, and report residuals across widths.
  2. [§2.5, §5.3.3] The shape-classification evaluation does not state whether the GMM was fit and evaluated on the same indentation set. As written, the GMM components are set to 5 and confusion matrices are computed from the clustered data, which would make the reported minimum F1 of 0.90 an in-sample value. Please clarify the evaluation protocol; if no held-out or cross-validated split was used, add one, since shape classification is one of the three main claims.
  3. [§5.3.3, Eq. (7)] Hu moments are scale-invariant only when computed from normalized central moments η_ij = µ_ij / µ_00^{1+(i+j)/2}. Equation (7) as written uses raw central moments µ_ij, and the normalization in Eq. (4) is introduced only for the width eigenvalues. If the classification reported here used unnormalized moments, the shape descriptors are not invariant to indenter size or indentation depth, which could confound the clustering. Please state which quantities were used and, if unnormalized, re-run with standard normalized moments.
minor comments (5)
  1. [Section numbering] There are two '5.3 Contact Inference' headings and three '5.5 Experimental Setup' headings; renumber the sections.
  2. [Fig. 5c] Report the number of indenter sizes, the number of trials per size, and add error bars or confidence intervals; currently the figure shows point distributions and linear fits without per-size statistics.
  3. [§5.4] The Gaussian filter standard deviation σ=10 is a free parameter; state how it was chosen and whether the calibration results are sensitive to its value.
  4. [Localization datasets] Clarify how many indentations constitute the 'seen' and 'unseen' localization datasets and confirm that the unseen dataset includes points between OFs in both axes, not only along one axis.
  5. [Abstract] The abstract's 'scalable route' is stronger than the current evidence; consider softening to 'proof-of-concept' or 'step toward', as the Discussion appropriately notes that long-fiber routing, bending, and packaging robustness are untested.

Circularity Check

1 steps flagged

Contact-width 'model' is an in-sample affine calibration: reported R2/RMSE validate the fitted offsets/scales, not Eq. 6's Gaussian prediction.

specific steps
  1. fitted input called prediction [Section 2.4 (Results); Eq. 6 in Methods 5.3.2]
    "An offset of 3.43 mm was removed from the measurements of the main axis and an offset of 2.47 mm was removed from the measurements of the minor axis. The measured lengths of the major and minor axes were scaled by 2.35 and 2.51, respectively, to produce the same linear response. Both the major and minor axes follow a strong linear relationship with R2 values of 0.95 and 0.97 respectively, indicating the model’s suitability in measuring contact width over a large range of contact sizes (Figure 5c). The RMSE values for the width of the major and minor axes were 1.2 mm and 1.0 mm respectively, wh"

    The contact-width capability is presented as a validated model, but the reported R2/RMSE are computed after subtracting per-axis offsets (3.43/2.47 mm) and applying per-axis scales (2.35/2.51) chosen on the same set of indenter diameters. No held-out or leave-one-diameter-out validation is described. With four free constants, an arbitrary monotone relation between the raw moment width and the true diameter can be absorbed, so the accuracy numbers measure the affine calibration fit to its own training data rather than an independent prediction of contact size. Eq. 6's Gaussian-profile assumption is stated but never validated, so the model is not tested independently of its calibration.

full rationale

The derivation chain is mostly self-contained empirically: centroid localization is calibrated on OF-aligned indentations and evaluated on a distinct unseen inter-OF dataset, so its sub-millimeter RMSE is a genuine held-out result, and the paper explicitly labels seen/unseen data. Shape classification is framed as a proof-of-principle; however, the Hu-moment combination was selected using the same data on which the F1 scores are reported, introducing in-sample selection bias, though this is not a definitional circularity. The serious circular step is contact-width estimation: Eq. 6 assumes a Gaussian tactile profile, then per-axis offsets (3.43/2.47 mm) and scales (2.35/2.51) are applied to the same diameter dataset before reporting R2/RMSE. No held-out validation is described, so the reported width accuracy reflects the affine calibration's fit to its own training data rather than an independent model prediction. Since measuring contact size is one of the three headline capabilities, this makes the central claim partially circular. Self-citations to TacTip are background/hardware context, not load-bearing mathematical premises; the Discussion's concession about short fibers and controlled conditions is an honest limitation, not circularity. Overall score 6.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The paper's analytical machinery is standard image moments; what it adds is hardware and empirical measurements. The width estimation carries four fitted constants and the localization calibration is an empirical lookup table; the scalability premise is acknowledged by the authors as untested. No new physical entities are postulated.

free parameters (5)
  • Major-axis width offset = 3.43 mm
    Removed from measured major-axis width to bring the linear response through the origin; fitted to the same indenter data used to report R2=0.95.
  • Minor-axis width offset = 2.47 mm
    Removed from measured minor-axis width; fitted to the same indenter data used to report R2=0.97.
  • Major-axis width scale factor = 2.35
    Multiplicative scale applied to major-axis measurements to match true diameter; fitted to data.
  • Minor-axis width scale factor = 2.51
    Multiplicative scale applied to minor-axis measurements; fitted to data.
  • Calibration Gaussian filter standard deviation = 10
    Chosen by hand to smooth nearest-neighbor error surfaces and reduce overfitting in the localization calibration (Section 5.4).
axioms (6)
  • domain assumption The light-intensity distribution created by an indentation is approximately Gaussian, so the FWHM formula (Eq. 6) can be used to estimate contact width.
    Invoked in Section 5.3.2 for width estimation; not experimentally validated, and the fitted offset/scale constants may absorb non-Gaussian bias.
  • domain assumption The coherent optical-fiber array preserves one-to-one spatial correspondence from the sensing face to the camera under the tested conditions, and can in principle do so at larger scale.
    Used throughout Sections 2.1 and 5.2; the Discussion concedes robustness under longer routing and bending is untested.
  • domain assumption Average OF pixel intensity, after background subtraction, changes locally and monotonically with pin deflection so that contacts appear as local intensity extrema.
    Basis for image-moment inference in Section 5.3; supported only qualitatively by Figure 3.
  • standard math Hu moment invariants are valid and stable discriminators for binarized tactile images.
    Standard result (Hu 1962, Eq. 7) applied without modification.
  • domain assumption A Gaussian Mixture Model with five components is an appropriate generative model for the five indenter shapes.
    Section 2.5 sets the number of components to 5, equal to the number of shapes, with no model-selection criterion.
  • domain assumption The calibration error surface sampled at 217 OF positions generalizes smoothly to arbitrary contact positions.
    Section 5.4/5.5; unseen data supports this at interpolated positions between OFs, but only within the tested grid and indentation protocol.

pith-pipeline@v1.3.0-alltime-deepseek · 14479 in / 15200 out tokens · 120911 ms · 2026-08-01T22:26:04.143712+00:00 · methodology

0 comments
read the original abstract

Robotic systems increasingly demand tactile sensing that approaches the adaptability and resolution of human skin to enable dexterous manipulation and safe interaction. OptiTac is a biomimetic tactile sensor that emulates the mechanoreceptor-to-nerve architecture of human touch by pairing each mechanical pin on a soft skin with an optical fiber acting as an artificial nerve. This design demonstrates an architectural principle for routing tactile information away from the sensing surface while preserving high spatial resolution, establishing a practical route toward distributed tactile sensing in future robotic systems. By treating tactile signals as images, simple analytical methods, rather than opaque deep-learning models, are used to infer contact location, size, and shape, providing interpretable and scalable tactile intelligence. This work demonstrates how evolutionary principles from biology can guide the development of artificial nerve systems for robots, offering a pathway toward human-like tactile perception in next-generation robotic platforms. More broadly, OptiTac establishes an artificial nerve-inspired sensing framework for interpretable robotic touch and a scalable route toward future distributed tactile systems.

Figures

Figures reproduced from arXiv: 2607.15746 by Chris J. Ford, Efi Psomopoulou, Laura E. Butcher, Nathan F. Lepora.

Figure 1
Figure 1. Figure 1: Concept and biomimetic design of the OptiTac tactile sensor. There are two subprocesses; tactile trans￾duction and high spatial resolution sampling. Tactile transduction in the OptiTac is inspired by the way humans transduce tactile information. Hairless skin contains features such as Merkel cells and Meissner Corpuscles that are key to the trans￾duction of mechanical stimulation of the skin and send activ… view at source ↗
Figure 2
Figure 2. Figure 2: Design and operating principle of the OptiTac. a. OptiTac sensor CAD schematic. b. the 217 OFs are arranged in a coherent hexagonal array. To demonstrate its coherence a repeating patterned object was imaged through the array when the tactile sensing module was not attached. Clear stripes are visible at the output of the OF array. c. the pre-processing of the raw tactile images to intensity surface plots f… view at source ↗
Figure 3
Figure 3. Figure 3: Biomimetic one-to-one pin-OF pattern. TacTip skin marker patterns a. sparse, b. dense and c. aligned. Left to right; CAD schematic, a cropped undeformed tactile image, a cropped tactile image under 3 mm of deformation by 5 mm diameter ball indenter and its resultant surface plot. 16 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Development of the contact centroid localization calibration method. To calibrate the measured cen￾troid values, pre-requisite information on how the errors vary over the surface of the sensor is required. Indentations are made directly above each OF within the array. The errors of the 217 indentations vary in both x and y axis so two cali￾bration plots are generated. For each plot, a nearest neighbor sear… view at source ↗
Figure 5
Figure 5. Figure 5: Tactile contact characterization of the OptiTac. a. To demonstrate the effect of the calibration process (see “Methods”), it was tested on seen data which was used in the development of the calibration protocol. An unseen data set which was not used in the development of the protocol was tested to demonstrate the generalizability of the calibration. The two scatter plots in the far right illustrate the eff… view at source ↗
Figure 6
Figure 6. Figure 6: Contact shape classification using the OptiTac. a. Dimensions in mm of the 3D printed indenters. An edge and four flat shapes (circle, square, ellipse, triangle). The output intensity surface plots are shown for each shape. b. Hu moment combinations and the resultant average of the diagonal elements in the corresponding confusion matrices for shape classification using a GMM. c. Confusion matrices for diff… 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

45 extracted references · 1 linked inside Pith

  1. [1]

    The gentle touch receptors of mammalian skin,

    A. Zimmerman, L. Bai, and D. D. Ginty, “The gentle touch receptors of mammalian skin,” Science, vol. 346, no. 6212, pp. 950–954, 2014

    2014

  2. [2]

    Tactile sensing—from humans to humanoids,

    R. S. Dahiya, G. Metta, M. Valle, and G. Sandini, “Tactile sensing—from humans to humanoids,”IEEE Transactions on Robotics, vol. 26, no. 1, pp. 1–20, 2009

    2009

  3. [3]

    Tactile sensibility in the human hand: relative and absolute den- sities of four types of mechanoreceptive units in glabrous skin.,

    R. S. Johansson and A. B. Vallbo, “Tactile sensibility in the human hand: relative and absolute den- sities of four types of mechanoreceptive units in glabrous skin.,” The Journal of Physiology, vol. 286, no. 1, pp. 283–300, 1979

    1979

  4. [4]

    Coding and use of tactile signals from the fingertips in object manipulation tasks,

    R. S. Johansson and J. R. Flanagan, “Coding and use of tactile signals from the fingertips in object manipulation tasks,” Nature Reviews Neuroscience, vol. 10, no. 5, pp. 345–359, 2009

    2009

  5. [5]

    Human tactile perception as a standard for artificial tactile sensing—a review,

    J. Dargahi and S. Najarian, “Human tactile perception as a standard for artificial tactile sensing—a review,” The International Journal of Medical Robotics and Computer Assisted Surgery, vol. 1, no. 1, pp. 23–35, 2004

    2004

  6. [6]

    Novel tactile sensor technology and smart tactile sensing systems: A review,

    L. Zou, C. Ge, Z. J. Wang, E. Cretu, and X. Li, “Novel tactile sensor technology and smart tactile sensing systems: A review,” Sensors, vol. 17, no. 11, p. 2653, 2017

    2017

  7. [7]

    Recent progress in technologies for tactile sensors,

    C. Chi, X. Sun, N. Xue, T. Li, and C. Liu, “Recent progress in technologies for tactile sensors,” Sen- sors, vol. 18, no. 4, p. 948, 2018

    2018

  8. [8]

    A review on sensory perception for dexterous robotic manipulation,

    Z. Xia, Z. Deng, B. Fang, Y. Yang, and F. Sun, “A review on sensory perception for dexterous robotic manipulation,” International Journal of Advanced Robotic Systems, vol. 19, no. 2, 2022

    2022

  9. [9]

    A survey of multifingered robotic manipulation: Biological results, structural evolvements, and learning methods,

    Y. Li, P. Wang, R. Li, M. Tao, Z. Liu, and H. Qiao, “A survey of multifingered robotic manipulation: Biological results, structural evolvements, and learning methods,” Frontiers in Neurorobotics, vol. 16, p. 843267, 2022

    2022

  10. [10]

    Review on human-like robot manipulation using dexterous hands.,

    S. K. Sampath, N. Wang, H. Wu, and C. Yang, “Review on human-like robot manipulation using dexterous hands.,” Cogn. Comput. Syst., vol. 5, no. 1, pp. 14–29, 2023

    2023

  11. [11]

    Tactile sensing technology in bionic skin: A review,

    C. Wang, C. Liu, F. Shang, S. Niu, L. Ke, N. Zhang, B. Ma, R. Li, X. Sun, and S. Zhang, “Tactile sensing technology in bionic skin: A review,” Biosensors and Bioelectronics, vol. 220, p. 114882, 2023. 11 5.5 Experimental Setup

    2023

  12. [12]

    Tactile innervation densities across the whole body,

    G. Corniani and H. P. Saal, “Tactile innervation densities across the whole body,” Journal of Neuro- physiology, vol. 124, no. 4, pp. 1229–1240, 2020

    2020

  13. [13]

    The GRASP taxonomy of human grasp types,

    T. Feix, J. Romero, H.-B. Schmiedmayer, A. M. Dollar, and D. Kragic, “The GRASP taxonomy of human grasp types,” IEEE Transactions on Human-Machine Systems, vol. 46, no. 1, pp. 66–77, 2015

    2015

  14. [14]

    The sensory neurons of touch,

    V. Abraira and D. Ginty, “The sensory neurons of touch,” Neuron, vol. 79, no. 4, pp. 618–639, 2013

    2013

  15. [15]

    The encoding of touch by somatotopically aligned dor- sal column subdivisions,

    J. Turecek, B. P. Lehnert, and D. D. Ginty, “The encoding of touch by somatotopically aligned dor- sal column subdivisions,” Nature, vol. 612, no. 7939, pp. 310–315, 2022

    2022

  16. [16]

    Mechanoreceptor signal convergence and transformation in the dorsal horn flexi- bly shape a diversity of outputs to the brain,

    A. M. Chirila, G. Rankin, S.-Y. Tseng, A. J. Emanuel, C. L. Chavez-Martinez, D. Zhang, C. D. Har- vey, and D. D. Ginty, “Mechanoreceptor signal convergence and transformation in the dorsal horn flexi- bly shape a diversity of outputs to the brain,” Cell, vol. 185, no. 24, pp. 4541–4559, 2022

    2022

  17. [17]

    The neural mechanisms of manual dexterity,

    A. R. Sobinov and S. J. Bensmaia, “The neural mechanisms of manual dexterity,” Nature Reviews Neuroscience, vol. 22, no. 12, pp. 741–757, 2021

    2021

  18. [18]

    Development of a tactile sensor based on biolog- ically inspired edge encoding,

    C. Chorley, C. Melhuish, T. Pipe, and J. Rossiter, “Development of a tactile sensor based on biolog- ically inspired edge encoding,” in 2009 International Conference on Advanced Robotics, pp. 1–6, IEEE, 2009

    2009

  19. [19]

    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, 2018

    2018

  20. [20]

    Hardware technology of vision-based tactile sensor: A review,

    S. Zhang, Z. Chen, Y. Gao, W. Wan, J. Shan, H. Xue, F. Sun, Y. Yang, and B. Fang, “Hardware technology of vision-based tactile sensor: A review,” IEEE Sensors Journal, vol. 22, no. 22, pp. 21410–21427, 2022

    2022

  21. [21]

    Soft biomimetic optical tactile sensing with the TacTip: A review,

    N. F. Lepora, “Soft biomimetic optical tactile sensing with the TacTip: A review,” IEEE Sensors Journal, vol. 21, no. 19, pp. 21131–21143, 2021

    2021

  22. [22]

    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

  23. [23]

    Anyrotate: Gravity-invariant in-hand object rotation with sim-to-real touch,

    M. Yang, C. Lu, A. Church, Y. Lin, C. Ford, H. Li, E. Psomopoulou, D. A. Barton, and N. F. Lep- ora, “Anyrotate: Gravity-invariant in-hand object rotation with sim-to-real touch,” arXiv preprint arXiv:2405.07391, 2024

    Pith/arXiv arXiv 2024

  24. [24]

    Dataset of tactile signatures of the human right hand in twenty-one activities of daily living using a high spatial resolution pressure sensor,

    J. Cepriá-Bernal and A. Pérez-González, “Dataset of tactile signatures of the human right hand in twenty-one activities of daily living using a high spatial resolution pressure sensor,” Sensors, vol. 21, no. 8, p. 2594, 2021

    2021

  25. [25]

    A high-resolution and whole-body dataset of hand-object contact areas based on 3d scanning method,

    Z. Chen, H. Chen, Y. Ouyang, C. Cao, W. Gao, Q. Hu, H. Jin, and S. Zhang, “A high-resolution and whole-body dataset of hand-object contact areas based on 3d scanning method,” Scientific Data, vol. 12, no. 1, pp. 1–17, 2025

    2025

  26. [26]

    Fiber optic sensors in tactile sensing: A review,

    C. Lyu, P. Li, J. Zhang, and Y. Du, “Fiber optic sensors in tactile sensing: A review,” IEEE Trans- actions on Instrumentation and Measurement, 2025

    2025

  27. [27]

    Numerical analysis of light propagation in image fibers or coherent fiber bundles,

    K. L. Reichenbach and C. Xu, “Numerical analysis of light propagation in image fibers or coherent fiber bundles,” Optics Express, vol. 15, pp. 2151–2165, Mar. 2007. Publisher: Optica Publishing Group

    2007

  28. [28]

    A review of tactile information: Perception and action through touch,

    Q. Li, O. Kroemer, Z. Su, F. F. Veiga, M. Kaboli, and H. J. Ritter, “A review of tactile information: Perception and action through touch,” IEEE Transactions on Robotics, vol. 36, no. 6, pp. 1619–1634, 2020

    2020

  29. [29]

    Using fiber optic bundles to miniaturize vision-based tactile sensors,

    J. Di, Z. Dugonjic, W. Fu, T. Wu, R. Mercado, K. Sawyer, V. R. Most, G. Kammerer, S. Speidel, R. E. Fan, et al., “Using fiber optic bundles to miniaturize vision-based tactile sensors,” IEEE Transactions on Robotics, 2024

    2024

  30. [30]

    An optical tactile sensor for manipulators,

    J. L. Schneiter and T. B. Sheridan, “An optical tactile sensor for manipulators,” Robotics and Computer- Integrated Manufacturing, vol. 1, pp. 65–71, Jan. 1984

    1984

  31. [31]

    Planar and finger-shaped optical tactile sensors for robotic applications,

    S. Begej, “Planar and finger-shaped optical tactile sensors for robotic applications,” IEEE Journal on Robotics and Automation, vol. 4, pp. 472–484, Oct. 1988

    1988

  32. [32]

    Shear, Torsion and Pressure Tactile Sensor via Plastic Optofiber Guided Imaging,

    D. Baimukashev, Z. Kappassov, and H. A. Varol, “Shear, Torsion and Pressure Tactile Sensor via Plastic Optofiber Guided Imaging,” IEEE Robotics and Automation Letters, vol. 5, pp. 2618–2625, Apr. 2020. 12 5.5 Experimental Setup

    2020

  33. [33]

    Feasibility study- novel optical soft tactile array sensing for minimally invasive surgery,

    J. Back, P. Dasgupta, L. Seneviratne, K. Althoefer, and H. Liu, “Feasibility study- novel optical soft tactile array sensing for minimally invasive surgery,” in 2015 IEEE/RSJ International Conference on In- telligent Robots and Systems (IROS), (Hamburg, Germany), pp. 1528–1533, IEEE, Sept. 2015

    2015

  34. [34]

    Optical soft tactile sensor algo- rithm based on multiscale resnet,

    Z. Lu, T. Yang, Z. Cao, D. Luo, Q. Zhang, Y. Liang, and Y. Dong, “Optical soft tactile sensor algo- rithm based on multiscale resnet,” IEEE Sensors Journal, vol. 23, no. 10, pp. 10731–10738, 2023

    2023

  35. [35]

    Tactile sensation for robots.,

    K. J. Overton and T. Williams, “Tactile sensation for robots.,” in IJCAI, pp. 791–795, 1981

    1981

  36. [36]

    A detailed review of feature extraction in image processing systems,

    G. Kumar and P. K. Bhatia, “A detailed review of feature extraction in image processing systems,” in 2014 Fourth international conference on advanced computing & communication technologies, pp. 5–12, IEEE, 2014

    2014

  37. [37]

    Opening a door with a humanoid robot using multi-sensory tactile feedback,

    A. J. Schmid, N. Gorges, D. Goger, and H. Worn, “Opening a door with a humanoid robot using multi-sensory tactile feedback,” in 2008 IEEE international conference on robotics and automation, pp. 285–291, IEEE, 2008

    2008

  38. [38]

    Haptic object recognition using passive joints and haptic key features,

    N. Gorges, S. E. Navarro, D. Göger, and H. Wörn, “Haptic object recognition using passive joints and haptic key features,” in 2010 IEEE International Conference on Robotics and Automation, pp. 2349–2355, IEEE, 2010

    2010

  39. [39]

    A control framework for tactile servoing.,

    Q. Li, C. Schürmann, R. Haschke, and H. J. Ritter, “A control framework for tactile servoing.,” in Robotics: Science and systems, 2013

    2013

  40. [40]

    Going in blind: Object motion classification using distributed tactile sensing for safe reaching in clutter,

    R. Thomasson, E. Roberge, M. R. Cutkosky, and J.-P. Roberge, “Going in blind: Object motion classification using distributed tactile sensing for safe reaching in clutter,” in 2022 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS), pp. 1440–1446, IEEE, 2022

    2022

  41. [41]

    Visual pattern recognition by moment invariants,

    M.-K. Hu, “Visual pattern recognition by moment invariants,” IRE Transactions on Information The- ory, vol. 8, no. 2, pp. 179–187, 1962

    1962

  42. [42]

    Object recognition by a smart tactile sensor,

    R. A. Russell, “Object recognition by a smart tactile sensor,” in Proceedings of the Australian Con- ference on Robotics and Automation, pp. 93–8, 2000

    2000

  43. [43]

    Design of a flexible tactile sensor for classifica- tion of rigid and deformable objects,

    A. Drimus, G. Kootstra, A. Bilberg, and D. Kragic, “Design of a flexible tactile sensor for classifica- tion of rigid and deformable objects,” Robotics and Autonomous Systems, vol. 62, no. 1, pp. 3–15, 2014

    2014

  44. [44]

    Assessing grasp stability based on learning and haptic data,

    Y. Bekiroglu, J. Laaksonen, J. A. Jorgensen, V. Kyrki, and D. Kragic, “Assessing grasp stability based on learning and haptic data,” IEEE Transactions on Robotics, vol. 27, no. 3, pp. 616–629, 2011

    2011

  45. [45]

    The mechanosensory neurons of touch and their mechanisms of activa- tion,

    A. Handler and D. D. Ginty, “The mechanosensory neurons of touch and their mechanisms of activa- tion,” Nature Reviews Neuroscience, vol. 22, no. 9, pp. 521–537, 2021. 13 5.5 Experimental Setup Figure 1:Concept and biomimetic design of the OptiTac tactile sensor.There are two subprocesses; tactile trans- duction and high spatial resolution sampling. Tacti...

    2021