Pith. sign in

REVIEW 2 major objections 6 minor 1 cited by

FasTac: A Curved Multispectral Vision-Based Tactile Sensor for High-Speed High-Precision 3D Shape and Force Perception

T0 review · 2 major / 6 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read A compact curved fingertip sensor recovers sub-0.05 mm contact depth, three-axis force, and millisecond feedback from one multispectral camera.

desk verdict Solid curved VBTS systems paper: clean depth and FPGA wins, force maps still only globally supervised on spheres. read the letter →

arxiv 2607.28416 v1 pith:3L2DDI2W submitted 2026-07-30 cs.RO

Xiaofan Lu , Kaiji Huang , Jiahui Chen , Yuankai Lin , Hua Yang , Zhouping Yin This is my paper · ORCID
classification cs.RO
keywords vision-basedtactilesensing3Dreconstructionthree-axisforcephotometricstereoFPGAaccelerationcurvedfingertipmultispectralimagingHyper
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

Dexterous robot fingertips need fine contact shape, separate normal and shear loads, and fast response on a curved surface—and most vision-based designs cannot do all three in a small package. FasTac pairs a single RGB-NIR camera with four-color illumination so photometric stereo stays well-posed on high curvature, then integrates normals with a CAD boundary prior to get depth. A position-aware dynamic convolution model (HyperForce) turns depth and marker motion into three-axis force that respects the elastomer’s spatially varying stiffness. Putting the image-to-normal-force path on an FPGA cuts latency to about a millisecond. The paper shows that this stack is accurate enough for delicate geometry, friction-aware grasp tightening, and tracking vibrations up to 100 Hz.

What carries the argument

HyperForce: FEM-inspired position-aware dynamic convolution that maps a fused 3D displacement field (depth-derived normal plus marker-derived tangential) through coordinate-conditioned kernels so local force responds to the curved elastomer’s nonuniform stiffness; paired with single-sensor RGB-NIR photometric stereo and Dirichlet boundary-prior fast Poisson depth.

What would settle it

Run HyperForce on non-sphere contacts (edges, multi-point, soft objects) with a calibrated local force reference or dense FEM ground truth and check whether distributed Fx/Fy/Fz maps and resultant NMAE stay near the reported ~2–3% sphere-test levels; large local or resultant error would break the central force claim.

Watch

Extended reading notes

Core claim

On a compact curved fingertip, single-sensor RGB-NIR photometric stereo plus boundary-prior Poisson depth reconstruction reaches 0.0415 mm depth MAE (versus roughly 0.27 mm without the prior), HyperForce estimates three-axis contact force at NMAE 2.74% normal and 2.39% shear, and an FPGA image-to-Fz pipeline runs in 1.09 ms—enough for fine geometry, friction-aware grasp feedback, and high-frequency dynamic contact sensing.

Load-bearing premise

That a force model trained only on global wrench labels from spherical-indenter contacts, with intermediate distributed force maps left unchecked, still matches real multi-object and grasp contacts on the same curved skin.

Editorial extensions

If this is right

  • Curved fingertips can report contact geometry at tens of microns without multi-camera or beam-splitter bulk.
  • Closed-loop grasp controllers can tighten from estimated friction coefficient µ = Fs/Fn in real time using on-sensor three-axis force.
  • Millisecond FPGA image-to-Fz feedback can track contact vibrations near 100 Hz that CPU/GPU pipelines alias.
  • A single RGB-NIR CFA plus diffuse skeleton lighting is enough to keep photometric stereo over-determined on high-curvature pads.

Reading between the lines

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

  • If position-aware kernels are the main win, the same hypernetwork idea may transfer to other curved soft sensors once a cheap position encoding is available, not only vision-based gels.
  • Deploying only Fz on the FPGA leaves shear and full 3D geometry host-side; full three-axis edge pipelines would be the natural next stress test for multi-finger hands.
  • Sphere-only supervision may understate error on sharp edges; a mixed-geometry calibration set would be a direct way to harden the force claim without redesigning the hardware.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. FasTac is a compact curved vision-based tactile fingertip that combines single-sensor RGB-NIR multispectral photometric stereo, boundary-prior Poisson depth reconstruction, a position-aware dynamic-convolution force estimator (HyperForce), and an FPGA image-to-Fz pipeline. On spherical-indenter data, NIR plus the boundary prior reduces depth MAE from ~0.27 mm to 0.0415 mm; HyperForce reports NMAE of 2.74% (normal) and 2.39% (shear) against an ATI Nano17; FPGA deployment cuts end-to-end latency from 3.26 ms (GPU) to 1.09 ms and supports vibration tracking up to 100 Hz. Qualitative multi-object reconstruction, friction-coefficient grasp feedback, and vibration experiments are used to illustrate geometry, stable force feedback, and dynamic sensing.

Significance. If the reported results hold under broader contact conditions, FasTac is a strong systems contribution: it jointly targets fine curved-surface geometry, three-axis force, and deterministic low-latency edge processing in a fingertip form factor that prior curved visuotactile designs rarely combine. Strengths include clear ablations (RGB vs RGB-NIR; depth prior on/off; dynamic vs fixed kernels; displacement-input variants), external ATI and CAD-aligned geometric ground truth, and concrete CPU/GPU/FPGA latency–energy–bandwidth comparisons with frequency-domain validation. The single-sensor RGB-NIR demultiplexing and boundary-prior DST Poisson solver are practical engineering choices that address known curved-photometric-stereo failure modes without multi-camera registration.

major comments (2)
  1. [§IV-B, Eqs. (6)–(11); §V-A; Table IV; Fig. 9] §IV-B (Eqs. 6–11) and §V-A/V-C: HyperForce is supervised only by global ATI wrench L1 after spatial integration of pixel-wise maps, on spherical-indenter contacts (0.5–1.5 N normal; 0.3/0.6 Fz shear on spatially gridded poses). Distributed f(p) fields are never compared to local force ground truth, and Table IV ablations remain inside that protocol. The grasp demo (Fig. 9) and multi-object use cases therefore assume that position-aware kernels fitted under sphere indentation transfer to non-spherical, multi-finger contacts. Please either (i) add a quantitative check of local/distributed force consistency (e.g., multi-indenter, known contact patches, or FEM reference maps) or (ii) clearly scope the force claim to resultant wrench under sphere-like contacts and mark distributed maps as intermediate visualizations.
  2. [§IV-C; Table V; Abstract; Contributions] §IV-C and Table V/S1: Only the normal-force branch is FPGA-deployed (image-to-Fz), while the abstract and contribution list advertise three-axis force and high-speed perception together. Tangential force still depends on marker detection, RBF interpolation, and the full HyperForce path off-FPGA. State explicitly what runs on-edge vs host, report end-to-end rates for the full three-axis path if claimed for closed-loop use, and avoid implying that 1.09 ms / 100 Hz applies to Fx/Fy as well as Fz.
minor comments (6)
  1. [Fig. 2] Fig. 2 contains residual Chinese labels (e.g., “侧视图”, “结构分解图”, gel/skeleton callouts). Replace with English for journal production.
  2. [Table I] Table I notes that evaluation settings differ across sensors; still, DenseTact/Insight/GelStereo numbers are presented side-by-side with FasTac. Add a short caveat in the table caption that MAE/NMAE/speed are not strictly commensurate.
  3. [§IV-A-2, Eq. (1)] Eq. (1): define α, β, and I_max numerically (or ranges used), and state whether they were tuned per unit or fixed across experiments.
  4. [§V-B, Fig. 7] §V-B strawberry/fingerprint/LED results are qualitative only. If possible, add a simple geometric proxy (e.g., known LED pitch or ridge spacing error) so multi-object claims are not purely visual.
  5. [Supplementary §I] Supplementary §I: training/validation/test grids are spatially disjoint—good—but report contact-location coverage relative to the full curved sensing area and any edge-region degradation.
  6. [Abstract; §IV-C] Notation: Fz vs F_z, and “image-to-Fz” vs three-axis F, should be consistent in abstract, §IV-C, and Table V.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: depth, force, and latency claims are empirical measurements against external ATI, CAD/CNC, and platform benchmarks.

full rationale

FasTac is a systems/hardware paper whose load-bearing numbers are not derived by rearranging fitted constants or self-defined quantities. Surface normals and depth are supervised and scored against CAD-aligned geometric ground truth rendered from CNC pose, indenter geometry, and the sensor model (Sec. V-A; Tables II–III); the boundary-prior Poisson step injects a known CAD Dirichlet condition into the RHS (Eqs. 3, 13) and is evaluated by residual MAE to that external geometry, not by restating the prior as a prediction. HyperForce is an FEM-motivated dynamic-convolution regressor (Eqs. 6–11) trained with global ATI Nano17 wrench L1 after spatial integration; reported MAE/NMAE/R² are held-out comparisons to the same external force sensor (Fig. 8, Table IV), not algebraic identities of the fit. FPGA image-to-Fz latency/energy (Table V, Fig. 10) are wall-clock and power measurements versus CPU/GPU. Prior GelSplitter-line citations supply related NIR imaging context but do not underwrite uniqueness theorems or force the reported errors. Generalization limits of sphere-trained distributed maps are a validation concern, not circularity. No step reduces a claimed prediction to its inputs by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 2 invented entities

Engineering sensor paper: load-bearing content is hardware design choices, classical photometric-stereo/Poisson machinery, and learned maps fitted to sphere-contact data. No new physical entities; free parameters are calibration/network/quantization choices that the reported errors depend on.

free parameters (6)
  • IR crosstalk subtraction strength α and NIR clamp β
    Eq. (1) uses hand-set correction gains that directly affect demultiplexed RGB used for normals.
  • Normal-estimation MLP weights = Trained up to 300 epochs, Adam lr 1e-3 (suppl.)
    Pixel-wise F(I⊕p) is trained on rendered sphere normals (Eq. 2); all depth numbers inherit this fit.
  • HyperForce hypernetwork / dynamic kernels H_α(c(p)) = AdamW lr 1e-4, ≤900 epochs (suppl.)
    Position-conditioned kernels (Eqs. 8–11) are fit to ATI global forces; central force NMAE is this fit’s test error.
  • RBF IMQ shape ε and marker-derived tangential field
    Eq. (4) kernel parameter and sparse-to-dense marker interpolation affect u_tan input to HyperForce.
  • FPGA quantization scales s_k, s_d and crop/downsample to 128×128 = 128×128 deployment input (suppl. III)
    Deployed Fz (Eqs. 14–16) and Table V accuracy depend on quantization and reduced resolution chosen for buffering/frame rate.
  • Gel mix ratio and reflective coating formulation = Solaris A:B:Slacker = 1:1:3
    1:1:3 elastomer and silver-powder paint set optical/mechanical response; all geometry and force maps are conditional on this build.
assumptions (5)
  • domain assumption Lambertian (or network-approximable) image formation under near-field multi-spectral illumination suffices for stable normals on the coated curved gel.
    Section IV-A invokes photometric stereo I=ρ L n and replaces constant L with position-aware MLP; silver paint and NIR are engineering mitigations, not proofs of the reflectance model.
  • domain assumption CAD-derived Dirichlet boundary depth D_prior correctly constrains the Poisson solve on the real molded fingertip.
    Eq. (3) and Table III show the prior removes ~0.27 mm drift; claim of 0.0415 mm MAE assumes mold/CAD alignment matches the physical boundary.
  • ad hoc to paper Local force response is a position-dependent linear combination of a K×K 3D displacement patch (FEM-inspired dynamic convolution).
    Section IV-B motivates f(p)≈Σ G_{p,q} u(p+q) and implements it as hypernetwork kernels; this is a modeling choice, not derived stiffness identification.
  • domain assumption Global wrench supervision is adequate to train pixel-wise force maps used for contact reasoning.
    Eq. (11) trains only on integrated F̂ vs ATI; distributed maps in figures are not independently labeled.
  • standard math Discrete sine-transform Poisson solver and streaming MLP/conv on FPGA faithfully implement the floating-point perception chain within reported error.
    Section IV-C / Fig. 5; Table V shows small Fz MAE change vs GPU, supporting the numerical claim for the deployed path.
invented entities (2)
  • HyperForce position-aware dynamic-convolution force estimator
    purpose: Map fused normal/tangential displacement on a curved elastomer to three-axis force with spatially varying kernels.
    Named method introduced in this paper; it is a learned architecture, not a new physical field. Independent handle is ATI test error and ablations, not an external physical discovery.
  • FasTac RGB-NIR curved visuotactile fingertip + FPGA image-to-Fz pipeline
    purpose: Single compact device combining multispectral shape, three-axis force, and low-latency normal-force edge compute.
    Hardware/system contribution; evidence is the authors’ build and experiments only unless others reproduce.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FasTac: A Curved Multispectral Vision-Based Tactile Sensor for High-Speed High-Precision 3D Shape and Force Perception." pith.science (2026). https://pith.science/paper/3L2DDI2W

@misc{pith2026260728416,
  author       = {Pith},
  title        = {Pith review of: FasTac: A Curved Multispectral Vision-Based Tactile Sensor for High-Speed High-Precision 3D Shape and Force Perception},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3L2DDI2W}},
  note         = {Machine review of arXiv:2607.28416}
}
read the original abstract

Curved tactile fingertips for dexterous manipulation must resolve fine contact geometry, distinguish normal and tangential loads, and capture transient signals. Existing curved vision-based tactile sensors struggle to combine accurate 3D reconstruction, three-axis force estimation, and high-speed processing in a compact form. This article presents FasTac, a curved vision-based tactile sensor integrating multispectral photometric stereo, dynamic-convolution force estimation, and hardware acceleration on a field-programmable gate array (FPGA). Single-image-sensor simultaneous multispectral imaging provides spatially aligned observations for robust surface normal estimation, followed by boundary-prior fast Poisson depth reconstruction. HyperForce uses position-aware dynamic convolution to model the spatially nonuniform mechanical response of curved elastomers and estimate three-axis forces. The complete image-to-normal-force pipeline is deployed on an FPGA. Experiments show that near-infrared (NIR) illumination and the boundary prior decrease depth mean absolute error (MAE) from 0.2730 mm to 0.0415 mm; HyperForce achieves normalized mean absolute error (NMAE) values of 2.74% and 2.39% for normal and shear forces, respectively; and FPGA deployment shortens processing latency from 3.26 ms on the GPU to 1.09 ms. Multi-object reconstruction, feedback grasping, and vibration measurement validate fine geometric perception, stable force feedback, and dynamic contact sensing.

Figures

Figures reproduced from arXiv: 2607.28416 by the authors.

Figure 1
Figure 1. High-precision 3D perception of delicate objects using the FasTac sensor. (a) A three-fingered dexterous hand equipped with FasTac sensors grasping a strawberry. (b-c) RGB and NIR images captured by FasTac. (d) Reconstructed surface normal map of the strawberry contact region. (e) Corresponding depth map. (f) Three-axis contact force distribution estimation results visualized on the curved surface. characteristics m… view at source ↗
Figure 1
Figure 1. FasTac uses RGB-NIR multispectral illumination with [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Mechanical design and integration of FasTac. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (8 more)
Figure 3
Figure 3. Figure 3: RGB-NIR spectral demultiplexing and preprocessing pipeline. The [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]
Figure 4
Figure 4. Figure 4: 3D reconstruction, Tangential displacement extraction, and three-axis force estimation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FPGA architecture of the image-to-Fz pipeline. (a) Quantized surface normal estimation. (b) Boundary-prior depth reconstruction. (c) Streaming Fz estimation. 2) Depth Reconstruction Module: The depth-reconstruction module converts estimated surface normals to gradients…
Figure 6
Figure 6. Figure 6: Data acquisition and geometric ground-truth generation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: 3D reconstruction results on three representative objects. The RGB/NIR inputs, predicted normal maps, and reconstructed point clouds under RGB-only and RGB-NIR illumination are compared. GT Fx (N) Pred Fx (N) GT Fy (N) GT Fz (N) Pred Fy (N) Pred Fz (N) R2=0.9920 R2=0.9…
Figure 8
Figure 8. Figure 8: Regression between predicted three-axis contact forces and ATI ground truth. TABLE V END-TO-END IMAGE-TO-NORMAL-FORCE PERFORMANCE OF CPU, GPU, AND FPGA IMPLEMENTATIONS. Platform Latency (ms) Force MAE (N) Energy (mJ/frame) CPU 6.82 0.0669 238.05 GPU 3.26 0.0667 33.60 F…
Figure 9
Figure 9. Figure 9: Contact force F, friction coefficient µ, and motor angle variation ∆θ without and with friction-coefficient feedback. Panel (a) shows the case without feedback, while panel (b) shows the case with feedback. The green dashed lines indicate representative time points dur…
Figure 10
Figure 10. Figure 10: Frequency spectra (top row) and time-domain normal-force responses (bottom row) of CPU, GPU, and FPGA under 50 Hz, 70 Hz, and 100 Hz inputs. VI. CONCLUSION FasTac is a compact curved tactile sensor integrating RGB￾NIR imaging, boundary-prior 3D reconstruction, three-a…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Near-sensor Computing for Rapid Visuotactile Perception

    cs.RO 2026-08 conditional novelty 6.0 of 10

    A streaming spectral Poisson solver on an FPGA reconstructs visuotactile depth maps with deterministic 0.211 ms latency, enabling a 28 ms protective reflex loop.

Reference graph

Works this paper leans on

38 extracted references · cited by 1 Pith paper

  1. [1]

    Tactile sensing in dexterous robot hands: Review,

    Z. Kappassov, J.-A. Corrales, and V . Perdereau, “Tactile sensing in dexterous robot hands: Review,”Robotics and Autonomous Systems, vol. 74, pp. 195–220, 2015

    2015

  2. [2]

    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. Kammereret al., “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, 2020

    2020

  3. [3]

    Soft, round, high resolution tactile fingertip sensors for dexterous robotic manipulation,

    B. Romero, F. Veiga, and E. Adelson, “Soft, round, high resolution tactile fingertip sensors for dexterous robotic manipulation,” in2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2020, pp. 4796–4802

    2020

  4. [4]

    BaroTac: Barometric three-axis tactile sensor with slip detection capability,

    G. Kim and D. Hwang, “BaroTac: Barometric three-axis tactile sensor with slip detection capability,”Sensors, vol. 23, no. 1, p. 428, 2023

    2023

  5. [5]

    A tactile sensor with slippage prediction by unequal-height dome array,

    Y . Yang, M. Zhao, Y . Jia, L. Zhao, D. Piao, Y . Zheng, and L. Song, “A tactile sensor with slippage prediction by unequal-height dome array,” IEEE Sensors Journal, vol. 23, no. 16, pp. 17 958–17 967, 2023

    2023

  6. [6]

    MinSight: A fingertip-sized vision-based tactile sensor for robotic manipulation,

    I. Andrussow, H. Sun, K. J. Kuchenbecker, and G. Martius, “MinSight: A fingertip-sized vision-based tactile sensor for robotic manipulation,” Advanced Intelligent Systems, vol. 5, no. 8, p. 2300042, 2023

    2023

  7. [7]

    A large-area robotic skin for intelligent tactile interaction of collaborative robots,

    W. Zheng, D. Guo, W. Yang, and H. Liu, “A large-area robotic skin for intelligent tactile interaction of collaborative robots,”IEEE/ASME Transactions on Mechatronics, pp. 1–11, 2025, early Access, Jan. 10, 2025, doi: 10.1109/TMECH.2024.3520953

    arXiv 2025

  8. [8]

    Stretchable capacitive tactile sensor array for accurate distributed pressure recognition,

    J. Yu, S. Yao, X. Jiang, and Z. Yao, “Stretchable capacitive tactile sensor array for accurate distributed pressure recognition,”IEEE Sensors Journal, vol. 24, no. 22, pp. 37 836–37 845, 2024

    2024

Show all 38 references
  1. [9]

    A soft multi-axis high force range magnetic tactile sensor for force feedback in robotic surgical systems,

    M. Rehan, M. M. Saleem, M. I. Tiwana, R. I. Shakoor, and R. Cheung, “A soft multi-axis high force range magnetic tactile sensor for force feedback in robotic surgical systems,”Sensors, vol. 22, no. 9, p. 3500, 2022

    2022

  2. [10]

    Large-area magnetic skin for multi-point and multi-scale tactile sensing with super-resolution,

    H. Hu, C. Zhang, X. Lai, H. Dai, C. Pan, H. Sun, D. Tang, Z. Hu, J. Fu, T. Liet al., “Large-area magnetic skin for multi-point and multi-scale tactile sensing with super-resolution,”npj Flexible Electronics, vol. 8, no. 1, p. 42, 2024

    2024

  3. [11]

    Recent progress in high-resolution tactile sensor array: From sensor fabrication to advanced applications,

    R. Wang, S. Hu, W. Zhu, Y . Huang, W. Wang, Y . Li, Y . Yang, J. Yu, and Y . Deng, “Recent progress in high-resolution tactile sensor array: From sensor fabrication to advanced applications,”Progress in Natural Science: Materials International, vol. 33, no. 1, pp. 55–66, 2023

    2023

  4. [12]

    Flexible tactile sensor array mounted on the curved surface: Analytical modeling and experimental validation,

    Y . Wang, G. Liang, D. Mei, and Z. Chen, “Flexible tactile sensor array mounted on the curved surface: Analytical modeling and experimental validation,”Journal of Microelectromechanical Systems, vol. 26, no. 5, pp. 1002–1011, 2017

    2017

  5. [13]

    Curvature-aware calibration of tactile sensors for accurate force estimation on non-planar surfaces,

    L. Zhong, H. J. H. Kim, D. P. Losey, and C. M. Nunez, “Curvature-aware calibration of tactile sensors for accurate force estimation on non-planar surfaces,”arXiv preprint arXiv:2510.25965, 2025

    2025

  6. [14]

    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, p. 2762, 2017

    2017

  7. [15]

    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,”IEEE Robotics and Automation Letters, vol. 9, no. 2, pp. 923–930, 2024

    2024

  8. [16]

    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). IEEE, 2019, pp. 5418–5424

    2019

  9. [17]

    GelTip: A finger-shaped optical tactile sensor for robotic manipulation,

    D. F. Gomes, Z. Lin, and S. Luo, “GelTip: A finger-shaped optical tactile sensor for robotic manipulation,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2020, pp. 9903–9909

    2020

  10. [18]

    GelSight360: An omnidirectional camera-based tactile sensor for dexterous robotic manipulation,

    M. H. Tippur and E. H. Adelson, “GelSight360: An omnidirectional camera-based tactile sensor for dexterous robotic manipulation,” in2023 IEEE International Conference on Soft Robotics (RoboSoft). IEEE, 2023, pp. 1–8

    2023

  11. [19]

    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

  12. [20]

    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, 2022

    2022

  13. [21]

    GelStereo BioTip: Self-calibrating bionic fingertip visuotactile sensor for robotic manipulation,

    S. Cui, S. Wang, C. Zhang, R. Wang, B. Zhang, S. Zhang, and Y . Wang, “GelStereo BioTip: Self-calibrating bionic fingertip visuotactile sensor for robotic manipulation,”IEEE/ASME Transactions on Mechatronics, vol. 29, no. 4, pp. 2451–2462, 2024

    2024

  14. [22]

    3D force and contact estimation for a SoftBubble visuotactile sensor using FEM,

    J.-C. Peng, S. Yao, and K. Hauser, “3D force and contact estimation for a SoftBubble visuotactile sensor using FEM,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 5666–5672. 11

    2024

  15. [23]

    3D vision-tactile reconstruction from infrared and visible images for robotic fine-grained tactile percep- tion,

    Y . Lin, X. Lu, J. Chen, and H. Yang, “3D vision-tactile reconstruction from infrared and visible images for robotic fine-grained tactile percep- tion,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 16 023–16 029

    2025

  16. [24]

    Photometric stereo: A reflectance map technique for determining surface orientation from image intensity,

    R. J. Woodham, “Photometric stereo: A reflectance map technique for determining surface orientation from image intensity,” inImage understanding systems and industrial applications I, vol. 155. SPIE, 1979, pp. 136–143

    1979

  17. [25]

    StereoTacTip: Vision-based tactile sensing with biomimetic skin-marker arrangements,

    C. Lu, K. Tang, X. Hui, H. Li, S. Nam, and N. F. Lepora, “StereoTacTip: Vision-based tactile sensing with biomimetic skin-marker arrangements,” IEEE/ASME Transactions on Mechatronics, vol. 30, no. 6, pp. 7558– 7570, 2025

    2025

  18. [26]

    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, 2022

    2022

  19. [27]

    GelFlow: Self-supervised learning of optical flow for vision-based tactile sensor displacement measurement,

    Z. Zhang, H. Yang, and Z. Yin, “GelFlow: Self-supervised learning of optical flow for vision-based tactile sensor displacement measurement,” inInternational Conference on Intelligent Robotics and Applications. Springer, 2023, pp. 26–37

    2023

  20. [28]

    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,” in2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009, pp. 1070–1077

    2009

  21. [29]

    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). IEEE, 2023, pp. 8979–8984

    2023

  22. [30]

    GelSplitter: Tactile reconstruction from near infrared and visible images,

    Y . Lin, Y . Zhou, K. Huang, Q. Zhong, T. Cheng, H. Yang, and Z. Yin, “GelSplitter: Tactile reconstruction from near infrared and visible images,” inInternational Conference on Intelligent Robotics and Applications. Springer, 2023, pp. 14–25

    2023

  23. [31]

    RainbowSight: A family of gener- alizable, curved, camera-based tactile sensors for shape reconstruction,

    M. H. Tippur and E. H. Adelson, “RainbowSight: A family of gener- alizable, curved, camera-based tactile sensors for shape reconstruction,” in2024 IEEE International Conference on Robotics and Automation (ICRA), 2024, pp. 1114–1120

    2024

  24. [32]

    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

    2025

  25. [33]

    AllSight: A low-cost and high-resolution round tactile sensor with zero-shot learning capability,

    O. Azulay, N. Curtis, R. Sokolovsky, G. Levitski, D. Slomovik, G. Lilling, and A. Sintov, “AllSight: A low-cost and high-resolution round tactile sensor with zero-shot learning capability,”IEEE Robotics and Automation Letters, vol. 9, no. 1, pp. 483–490, 2024

    2024

  26. [34]

    Recent progress of optical tactile sensors: A review,

    N. Yao and S.-P. Wang, “Recent progress of optical tactile sensors: A review,”Optics and Laser Technology, vol. 176, p. 111040, 2024

    2024

  27. [35]

    HiVTac: A high-speed vision-based tactile sensor for precise and real-time force reconstruction with fewer markers,

    S. Quan, X. Liang, H. Zhu, M. Hirano, and Y . Yamakawa, “HiVTac: A high-speed vision-based tactile sensor for precise and real-time force reconstruction with fewer markers,”Sensors, vol. 22, no. 11, p. 4196, 2022

    2022

  28. [36]

    EveTac: An event-based optical tactile sensor for robotic manipulation,

    N. Funk, E. Helmut, G. Chalvatzaki, R. Calandra, and J. Peters, “EveTac: An event-based optical tactile sensor for robotic manipulation,”IEEE Transactions on Robotics, vol. 40, pp. 3812–3832, 2024

    2024

  29. [37]

    FPGA-based tactile sensor suite electronics for real-time embedded processing,

    O. Oballe-Peinado, J. A. Hidalgo-Lopez, J. Castellanos-Ramos, J. A. Sanchez-Duran, R. Navas-Gonzalez, J. Herran, and F. Vidal-Verd ´u, “FPGA-based tactile sensor suite electronics for real-time embedded processing,”IEEE Transactions on Industrial Electronics, vol. 64, no. 12, ...

    2017

  30. [38]

    Fast reactive grasping with in-finger vision and in-hand FPGA-accelerated CNNs,

    F. Hundhausen, R. Grimm, L. Stieber, and T. Asfour, “Fast reactive grasping with in-finger vision and in-hand FPGA-accelerated CNNs,” in2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2021, pp. 6825–6832. 1 Supplementary Material for “Fas...

    2021

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.