Pith. sign in

REVIEW 3 major objections 6 minor 25 references

A wearable fingertip sensor with 24 capacitive taxels can estimate a dense 35×26 contact-depth map, and that dense representation materially improves robot grasping and human-to-robot replay beyond what raw taxel readings provide.

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 11:07 UTC pith:TYM5SSTX

load-bearing objection Solid wearable tactile sensor with clear task-level wins; the sub-mm depth accuracy claims are sim-to-real consistency numbers, not physical deformation accuracy—exactly what the authors' own Limitations concede. the 3 major comments →

arxiv 2607.29231 v1 pith:TYM5SSTX submitted 2026-07-31 cs.RO

TacPrint: A Wearable Fingertip Tactile Sensor for Human-to-Robot Contact Reproduction

Yongxi Liu , Chaofan Zhang , Xingyu Zhang , Xiangyin Bao , Boyue Zhang , Shaowei Cui , Shuo Wang This is my paper
classification cs.RO
keywords wearable tactile sensorcapacitive fingertip sensingcontact depth mapreal-to-sim-to-realhuman-to-robot transferclosed-loop graspingLSTM depth estimationtactile-guided compensation
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.

TacPrint is a wearable fingertip tactile sensor that uses 24 capacitive taxels and a real-to-sim-to-real pipeline to estimate a dense 35×26 map of contact depth on the fingertip. The paper's central claim is that this low-cost, sparse sensing strategy can recover spatially detailed contact information sufficient to correct robot replay and guide closed-loop grasping. The authors show that when the estimated depth map is used as feedback, grasping success rises from 67.5% (raw-taxel centroid) to 87.5% overall, and from 45% to 85% under edge-contact conditions; similarly, tactile-guided normal compensation turns a 0% success replay into 91.67% (grasping) and 90% (wiping). A sympathetic reader would care because wearable tactile data could let humans demonstrate contact-rich skills naturally, and the dense-but-cheap representation narrows the gap between sparse sensing and high-resolution tactile geometry.

Core claim

The paper claims that a 24-channel capacitive fingertip sensor, worn on a human finger, can be trained to output a 35×26 contact-depth map whose accuracy is near-millimeter: on simulation-generated labels the model reaches a contact-region RMSE of 0.223±0.161 mm and an IoU of 0.829±0.169; on physical indentations the predicted depth at the contact center deviates by only 0.085±0.057 mm, and the predicted contact position by 0.250±0.208 mm. The authors further claim that this dense-depth estimate, not the raw 6×4 taxel readings, is what makes closed-loop grasp adjustment reliable, and that tactile compensation along the fingertip normal recovers contact that vision-based replay misses.

What carries the argument

The core mechanism is the one-to-one alignment between 24 hemispherical silicone protrusions on the inner skin and the 24 capacitive taxels; each protrusion localizes the capacitive response to a small region. Around that hardware sits a real-to-sim-to-real loop: controlled physical indentations provide synchronized capacitive sequences; a physics-based soft-body simulation of the silicone elastomer (Neo-Hookean, E=0.30 MPa, ν=0.47, friction 0.20) generates the corresponding 35×26 depth labels; and an LSTM temporal encoder feeding a spatial decoder maps a 9-frame window of 24-channel signals to the depth map. A foreground-weighted L1+L2 loss emphasizes pixels in contact during training. The

Load-bearing premise

The simulation-generated depth labels—computed with Neo-Hookean parameters E=0.30 MPa, ν=0.47, and friction 0.20 fixed from the literature—faithfully represent the deformation of the fabricated silicone skin; the paper validates them only at configuration level (center depth and centroid), and its Limitations explicitly state that residual discrepancies may remain in contact boundaries and local shapes.

What would settle it

Mount a calibrated structured-light or stereo camera under the TacPrint silicone skin (or use a transparent indenter with fiducial markers) and record the full 3D deformation of the elastomer during the same controlled indentations used for training; compute a pixel-wise error map between the measured deformation and the simulated labels. If the pixel-level RMSE substantially exceeds the reported 0.223 mm contact-region RMSE, or if the error is systematic (e.g., biased contact boundaries), the simulator's fidelity—and therefore the ground truth for the network—is not what the paper assumes.

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

If this is right

  • A $50 wearable fingertip sensor can enrich human demonstrations with local contact geometry, reducing the need for tight visual alignment during teleoperation or replay.
  • Dense-depth feedback outperforms raw-taxel centroid feedback for closed-loop grasping, especially near the sensing boundary (85% vs 45% success on edge contacts).
  • Tactile-guided normal compensation converts vision-only replay failures into high-success grasps and wipes (91.67% and 90%), suggesting contact information is a practical substitute for visual contact recovery.
  • The real-to-sim-to-real pipeline with fixed, literature-derived material parameters is sufficient for task-level success across multiple indenter geometries and contact conditions.

Where Pith is reading between the lines

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

  • The method implies a general recipe for turning sparse, inexpensive taxel arrays into high-resolution contact maps: use physics simulation to generate dense labels, then train a temporal network. This could transfer to other sensor form factors (e.g., gloves, palm pads) without hardware redesign.
  • Because the physical validation only checks center depth and centroid, the claimed pixel-level accuracy remains a claim about simulator fidelity; a direct full-field deformation measurement would either confirm or refute the sim-to-real transfer more rigorously than the reported configuration-level errors.
  • The task gains in edge-contact conditions suggest that dense reconstruction corrects a systematic boundary bias in centroid estimation from sparse arrays; extending this approach to other sparse tactile layouts might yield similar benefits near sensor edges.
  • A natural next test is to use TacPrint in a large-scale imitation-learning collection where the dense depth maps become part of the training signal (not just replay compensation), probing whether the added modality improves policy learning rather than only reactive control.

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 / 6 minor

Summary. The paper presents TacPrint, a wearable fingertip tactile sensor with 24 capacitive taxels and an elastomer skin whose protrusions align one-to-one with the taxels. A real-to-sim-to-real pipeline uses LSTM encoding and a spatial decoder to estimate a 35×26 contact-depth map from short temporal windows of 24-channel capacitive signals. The authors evaluate the sensor in three ways: (1) prediction accuracy against simulation-generated labels and against controlled physical indentations at the contact center and centroid; (2) human-to-robot replay with tactile-guided normal compensation, raising fruit-grasping success from 0% to 91.67% and whiteboard-wiping from 0% to 90%; and (3) closed-loop grasping with dense-depth feedback, achieving 87.5% overall and 85% edge-contact success versus 37.5%/20% for contact-only and 67.5%/45% for raw-taxel feedback. The paper is transparent about several limitations, including the fact that physical validation checks depth at the center and contact location but not full-field pixel-level deformation.

Significance. If the dense depth reconstruction is physically accurate, the contribution is significant: a $50 wearable sensor with a compact form factor that provides spatially resolved contact geometry during natural human demonstration, with demonstrated utility in both replay correction and closed-loop grasping. The paper's strengths include a concrete hardware design, a reproducible training pipeline with fixed hyperparameters, controlled physical indentation experiments, and task-level experiments that show large improvements over no-feedback baselines. The authors also explicitly state the boundaries of their physical validation. However, the central quantitative claim of dense depth-map accuracy currently rests on simulation-generated labels whose full-field physical fidelity is not directly measured; this limits the strength of the 'contact reproduction' claim and makes the significance conditional on future validation.

major comments (3)
  1. [§III.B.2, Table I, Abstract] The headline dense-map accuracy metrics — contact-region RMSE 0.223±0.161 mm, centroid error 1.213±2.379 px, IoU 0.829±0.169 — are computed against depth labels produced by the TacFlex simulator (ref. [23]), which shares authors with this paper and whose Neo-Hookean parameters (E=0.30 MPa, ν=0.47, friction=0.20) are fixed literature values rather than identified from the fabricated skin. The physical validation in §IV.A/§IV.C only compares center depth and contact centroid, not pixel-level deformation, and the Limitations explicitly concede that residual discrepancies may remain in contact boundaries and local shapes. This is load-bearing because the title and abstract claim contact reproduction and the quantitative sub-mm accuracy is presented as a central result. As written, the paper overstates what is measured. I recommend either adding a direct physical full-field deformation valida
  2. [§IV.B, Eq. (7), Exp. 2] The tactile-guided compensation experiment compares direct replay against replay with the additional term α_i d_i(t) n_i(t). The gains α_i are empirically selected as 'the smallest value that enabled repeatable contact recovery in preliminary trials,' and direct replay achieves 0% success. Since d_i(t) is the mean of the 10 largest values in the estimated depth map, it is likely an approximately constant scalar during sustained contact; the improvement might therefore be largely a fixed inward displacement rather than evidence that the reconstructed depth magnitude provides informative feedback. An ablation with a constant normal displacement of the same mean magnitude would be needed to isolate the contribution of the tactile depth signal. Without this, the causal claim that tactile information 'increased success rates from 0% to 91.67%' is not fully established.
  3. [§IV.C, Exp. 3] The dense-depth control versus raw-taxel control comparison, while practically useful, is not a controlled test of depth-map accuracy: dense-depth uses a 35×26 map while raw-taxel uses a 6×4 centroid. The improved success rate (87.5% vs 67.5%) is consistent with the dense representation providing better localization, but it could also be due to algorithmic differences in thresholding or centroid weighting. A fairer comparison would use the same localization algorithm on a densified/interpolated taxel map, or report the theoretical resolution limits of the 6×4 array. This does not invalidate the task-level result, but it should be framed as a system-level comparison rather than evidence that the reconstructed depth map is physically accurate.
minor comments (6)
  1. [Figure 2] The figure caption contains extraneous Chinese text ('图2 结构图') and the dimensions in subfigure (b) are not all labeled with units. Please clean up the captions and ensure all units are included.
  2. [Figure 4] The caption includes the stray phrase '新增流程图' (Chinese for 'new flowchart'). Remove it and standardize all figure captions.
  3. [Figure 9] The caption contains 'ExpB 图8擦白板v3', which appears to be an internal editing note. Remove it.
  4. [Eq. (2)] D_max is used in the normalization but the symbol is not defined in the main text near Eq. (2); it is later stated as 3 mm. Please define it when first introduced.
  5. [§IV.A, Exp. 1] The paper uses a 0.4-mm segmentation threshold for physical position evaluation but τ=0.1 mm for IoU/contact masks in Eq. (8). The sensitivity of the reported position error and IoU to these threshold choices is not discussed. A sentence justifying the thresholds would improve reproducibility.
  6. [Related Work] The related-work section is concise but would benefit from explicit comparison of the spatial resolution, form factor, and cost of TacPrint against ThimbleSense, FingerTac, and TacCap, rather than a purely descriptive listing.

Circularity Check

0 steps flagged

No significant circularity: sim-based labels are evaluated against independent physical references, and the paper's own limitations separate configuration-level checks from full-field deformation.

full rationale

The paper's derivation chain is not circular. The 35×26 depth labels are generated by reproducing each nominal contact configuration in TacFlex, but the physical evaluations do not use those labels as ground truth: the 0.085±0.057 mm center-depth error and 0.250±0.208 mm position error compare network predictions against guide-calibrated physical references. The performance against simulation-generated labels (Table I) is a standard held-out supervised-learning consistency check, not a prediction that is equivalent to the training inputs by construction. The paper explicitly states that the simulation calibration is 'configuration-level correspondence between the physical and simulated contacts rather than independent full-field deformation calibration', and its Limitations acknowledge that 'the current physical evaluations validate contact depth and location rather than complete pixel-level deformation of the real elastomer.' That is a validation gap, not a circular derivation. The TacFlex citation [23] is prior work by overlapping authors, but it is used as a physics backend with stated literature-based material parameters, not as a self-justifying uniqueness claim, and its output is independently checked against controlled physical indentations at the configuration level. Downstream grasping and wiping results are measured physical task successes, not quantities derived from the fitted model. The paper's central claims therefore do not reduce to their inputs by definition.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The central depth-map claim rests on simulation-generated labels produced by the authors' own TacFlex simulator, with material parameters chosen from literature rather than characterized for the fabricated skin. The network and downstream controllers add numerous fitted parameters and hand-chosen thresholds/gains. No new physical entities or theoretical constructs are introduced.

free parameters (7)
  • Silicone Young's modulus E = 0.30 MPa
    Chosen from literature/benchmarks for Neo-Hookean model in TacFlex; not measured from the fabricated skin. All simulated depth labels depend on it.
  • Poisson's ratio ν = 0.47
    Fixed with E as the constitutive model for the silicone; uncertainty directly affects simulated deformation.
  • Indenter–elastomer friction coefficient = 0.20
    Fixed value in simulation; affects contact boundary and peripheral deformation.
  • Segmentation threshold for contact region = 0.4 mm
    Selected post hoc from {0.3, 0.4, 0.5} mm based on lowest median/95th percentile errors; used to compute centroid position and contact-region metrics.
  • Tactile compensation gains = α_grasp=[0.005,0.005,0.005] m/mm; α_wipe=[0.004,0.010,0.010] m/mm
    Empirically selected as the smallest values enabling repeatable contact recovery in preliminary trials; directly determine the replay corrections.
  • Foreground loss parameters = λ=0.5, δ=0.3, γ=2
    Hand-chosen weights for the squared/absolute loss and foreground emphasis; affect the learned depth maps.
  • Depth normalization scale D_max = 3 mm
    Fixed global scale for clipping/normalizing depth labels; sets the absolute scale of network outputs.
axioms (5)
  • domain assumption The TacFlex physics simulation accurately models the deformation of the TacPrint silicone skin for the tested indenter configurations.
    Invoked to treat simulation-generated depth maps as training labels and as reference for depth/centroid metrics (Sec. III-B.2).
  • domain assumption The Neo-Hookean material model with E=0.30 MPa, ν=0.47, and friction 0.20 adequately represents the fabricated silicone.
    These parameters are fixed from literature/benchmarks, not measured for this sensor; they determine all simulated deformation fields (Sec. III-B.2).
  • domain assumption The prescribed CNC guide displacements and CAD geometry define the physical ground-truth contact depth and location within acceptable error (mold-support mismatch ≤0.2 mm; bonding variation unquantified).
    Used as ground truth for the 40 physical trials and for generating sim labels (Sec. III-A/B).
  • domain assumption A network trained on simulation-generated labels transfers to real capacitive inputs.
    The real-to-sim-to-real pipeline assumes the distribution gap is small enough that physical sensor readings map to accurate depth predictions (Sec. III-B.2/IV-C).
  • domain assumption The reported center-depth and centroid metrics are sufficient proxies for the accuracy of the full depth map.
    The paper's Limitations state that pixel-level deformation is not validated; this assumption is required to support the 35×26 depth-map accuracy claim (Sec. V).

pith-pipeline@v1.3.0-daily-deepseek · 11324 in / 16975 out tokens · 173092 ms · 2026-08-03T11:07:40.708053+00:00 · methodology

0 comments
read the original abstract

Human-centric data collection is emerging as a significant paradigm for robot skill acquisition, but seamlessly integrating low-cost, scalable tactile sensing systems that capture fine-grained fingertip interactions without compromising natural operation remains a key challenge. This reduces the reliability of human-to-robot transfer in contact-rich tasks. In this work, we present TacPrint, a wearable fingertip tactile sensor, where protrusions on the inner surface of the silicone skin are aligned one-to-one with 24 capacitive taxels to enable localized capacitive responses. A real-to-sim-to-real pipeline estimates a 35 $\times$ 26 contact-depth map from 24-channel capacitive signals. Against simulation-generated labels, the model achieved a contact-region RMSE of 0.223 $\pm$ 0.161 mm, a weighted-centroid error of 1.213 $\pm$ 2.379 pixels, and an IoU of 0.829 $\pm$ 0.169. With measured capacitive inputs, the network-predicted depth evaluated at the guide-calibrated contact center showed a mean absolute error of 0.085 $\pm$ 0.057 mm across all 40 controlled trials, while the mean contact-position error was 0.250 $\pm$ 0.208 mm across the 37 trials whose reference contact regions were not truncated by the sensing boundary. In human-to-robot replay, tactile-guided compensation increased grasping and wiping success rates from 0% to 91.67% and 90%, respectively. In closed-loop grasping, dense-depth feedback achieved success rates of 87.5% over all tested positions and 85% under edge-contact conditions, compared with 67.5% and 45% for raw-taxel feedback.

Figures

Figures reproduced from arXiv: 2607.29231 by Boyue Zhang, Chaofan Zhang, Shaowei Cui, Shuo Wang, Xiangyin Bao, Xingyu Zhang, Yongxi Liu.

Figure 1
Figure 1. Figure 1: TacPrint is a compact, low-cost wearable fingertip sensor that estimates [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Structure design of TacPrint. (a) TacPrint components: a silicone [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the tactile-to-depth learning pipeline. Controlled physi [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Pipeline for tactile-guided compensation. Direct replay uses vision [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Setup of the contact-position-aware grasping experiment. (a) Robotic [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Representative depth-map prediction results on test samples. For each [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: TacPrint outputs during contact with a water bottle on two different [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Trajectory comparison between direct replay and replay with tactile [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Whiteboard-wiping experiments. The first row shows the human [PITH_FULL_IMAGE:figures/full_fig_p006_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Representative contact-position-aware grasping trials at Pos. 1. Contact only detects valid contact but performs no lateral adjustment and fails to [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Grasping performance of the three control methods. (a) Success rates [PITH_FULL_IMAGE:figures/full_fig_p007_11.png] 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

25 extracted references · 2 canonical work pages

  1. [1]

    Learning fine-grained bimanual manipulation with low-cost hardware,

    T. Z. Zhao, V . Kumar, S. Levine, and C. Finn, “Learning fine-grained bimanual manipulation with low-cost hardware,” 2023. [Online]. Available: https://arxiv.org/abs/2304.13705

    Pith/arXiv arXiv 2023

  2. [2]

    Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation,

    Z. Fu, T. Z. Zhao, and C. Finn, “Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation,” 2024. [Online]. Available: https://arxiv.org/abs/2401.02117

    Pith/arXiv arXiv 2024

  3. [3]

    Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators,

    P. Wu, Y . Shentu, Z. Yi, X. Lin, and P. Abbeel, “Gello: A general, low-cost, and intuitive teleoperation framework for robot manipulators,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 12 156–12 163

    2024

  4. [4]

    How to train your robots? the impact of demonstration modality on imitation learning,

    H. Li, Y . Cui, and D. Sadigh, “How to train your robots? the impact of demonstration modality on imitation learning,” 2025. [Online]. Available: https://arxiv.org/abs/2503.07017

    Pith/arXiv arXiv 2025

  5. [5]

    Learn to grasp objects with dexterous robot manipulator from human demonstration,

    Y . Hu, K. Li, and N. Wei, “Learn to grasp objects with dexterous robot manipulator from human demonstration,” in2022 International Conference on Advanced Robotics and Mechatronics (ICARM), 2022, pp. 1062–1067

    2022

  6. [6]

    Universal manipulation interface: In-the-wild robot teaching without in-the-wild robots,

    C. Chi, Z. Xu, C. Pan, E. Cousineau, B. Burchfiel, S. Feng, R. Tedrake, and S. Song, “Universal manipulation interface: In-the-wild robot teaching without in-the-wild robots,” 2024. [Online]. Available: https://arxiv.org/abs/2402.10329

    Pith/arXiv arXiv 2024

  7. [7]

    Freetacman: Robot-free visuo-tactile data collection system for contact-rich manipulation,

    L. Wu, C. Yu, J. Ren, L. Chen, Y . Jiang, R. Huang, G. Gu, and H. Li, “Freetacman: Robot-free visuo-tactile data collection system for contact-rich manipulation,” 2026. [Online]. Available: https://arxiv.org/abs/2506.01941

    arXiv 2026

  8. [8]

    Dexcap: Scalable and portable mocap data collection system for dexterous manipulation,

    C. Wang, H. Shi, W. Wang, R. Zhang, L. Fei-Fei, and C. K. Liu, “Dexcap: Scalable and portable mocap data collection system for dexterous manipulation,” 2024. [Online]. Available: https://arxiv.org/ abs/2403.07788

    Pith/arXiv arXiv 2024

  9. [9]

    Arcap: Collecting high-quality human demonstrations for robot learning with augmented reality feedback,

    S. Chen, C. Wang, K. Nguyen, L. Fei-Fei, and C. K. Liu, “Arcap: Collecting high-quality human demonstrations for robot learning with augmented reality feedback,” 2024. [Online]. Available: https://arxiv.org/abs/2410.08464

    Pith/arXiv arXiv 2024

  10. [10]

    Dart: Dexterous augmented reality teleoperation platform for large-scale robot data collection in simulation,

    Y . Park, J. S. Bhatia, L. Ankile, and P. Agrawal, “Dart: Dexterous augmented reality teleoperation platform for large-scale robot data collection in simulation,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 13 883–13 889

    2025

  11. [11]

    Dexumi: Using human hand as the universal manipulation interface for dexterous manipulation,

    M. Xu, H. Zhang, Y . Hou, Z. Xu, L. Fan, M. Veloso, and S. Song, “Dexumi: Using human hand as the universal manipulation interface for dexterous manipulation,” 2025. [Online]. Available: https://arxiv.org/abs/2505.21864

    arXiv 2025

  12. [12]

    Dexvitac: Collecting human visuo-tactile-kinematic demonstrations for contact-rich dexterous manipulation,

    X. Chen, Y . Pan, M. Li, and X. Ding, “Dexvitac: Collecting human visuo-tactile-kinematic demonstrations for contact-rich dexterous manipulation,” 2026. [Online]. Available: https://arxiv.org/abs/2603. 17851

    2026

  13. [13]

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

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

    2020

  14. [14]

    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, pp. 135–145, 2022. [Online]. Available: https://doi.org/10.1038/s42256-021-00439-3

    work page doi:10.1038/s42256-021-00439-3 2022

  15. [15]

    Dtact: A vision-based tactile sensor that measures high-resolution 3d geometry directly from darkness,

    C.-Y . Lin, Z.-X. Lin, S. Wang, and H. Xu, “Dtact: A vision-based tactile sensor that measures high-resolution 3d geometry directly from darkness,”2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 10 359–10 366, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:252568209

    2023

  16. [16]

    Dottip: Enhancing dexterous robotic manipulation with a tactile fingertip featuring curved perceptual morphology,

    H. Zheng, X. Shi, A. Bao, Y . Jin, and P. Zhao, “Dottip: Enhancing dexterous robotic manipulation with a tactile fingertip featuring curved perceptual morphology,”IEEE Robotics and Automation Letters, vol. 10, no. 1, pp. 772–779, 2025

    2025

  17. [17]

    Active shape reconstruction using a novel visuotactile palm sensor,

    J. Hu, S. Cui, S. Wang, R. Wang, and Y . Wang, “Active shape reconstruction using a novel visuotactile palm sensor,” Biomimetic Intelligence and Robotics, vol. 4, no. 3, p. 100167, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S2667379724000251

    2024

  18. [18]

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

    P. Lin, Y . Huang, W. Li, J. Ma, C. Xiao, and Z. Jiao, “Pp-tac: Paper picking using tactile feedback in dexterous robotic hands,” 2025. [Online]. Available: https://arxiv.org/abs/2504.16649

    Pith/arXiv arXiv 2025

  19. [19]

    Thimblesense: A fingertip-wearable tactile sensor for grasp analysis,

    E. Battaglia, M. Bianchi, A. Altobelli, G. Grioli, M. G. Catalano, A. Se- rio, M. Santello, and A. Bicchi, “Thimblesense: A fingertip-wearable tactile sensor for grasp analysis,”IEEE Transactions on Haptics, vol. 9, no. 1, pp. 121–133, 2016

    2016

  20. [20]

    Machine learning-enabled tactile sensor design for dynamic touch decoding,

    Y . Lu, D. Kong, G. Yang, R. Wang, G. Pang, H. Luo, H. Yang, and K. Xu, “Machine learning-enabled tactile sensor design for dynamic touch decoding,”Advanced Science, vol. 10, no. 32, p. 2303949,

  21. [21]

    Fingertac - an interchangeable and wearable tactile sensor for the fingertips of human and robot hands,

    P. Sathe, A. Schmitz, T. P. Tomo, S. Somlor, S. Funabashi, and S. Shigeki, “Fingertac - an interchangeable and wearable tactile sensor for the fingertips of human and robot hands,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023, pp. 10 813–10 820

    2023

  22. [22]

    Taccap: A wearable fbg-based tactile sensor for seamless human-to-robot skill transfer,

    C. Xing, H. Li, Y .-L. Wei, T.-A. Ren, T. Tu, Y . Lin, E. Schumann, W.-S. Zheng, and M. R. Cutkosky, “Taccap: A wearable fbg-based tactile sensor for seamless human-to-robot skill transfer,” 2025. [Online]. Available: https://arxiv.org/abs/2503.01789

    Pith/arXiv arXiv 2025

  23. [23]

    Tacflex: Multimode tactile imprints simulation for visuotactile sensors with coating patterns,

    C. Zhang, S. Cui, J. Hu, T. Jiang, T. Zhang, R. Wang, and S. Wang, “Tacflex: Multimode tactile imprints simulation for visuotactile sensors with coating patterns,”IEEE Transactions on Robotics, vol. 41, pp. 3965–3985, 2025

    2025

  24. [24]

    Bridging the human to robot dexterity gap through object-oriented rewards,

    I. Guzey, Y . Dai, G. Savva, R. Bhirangi, and L. Pinto, “Bridging the human to robot dexterity gap through object-oriented rewards,” in2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp. 3344–3351

    2025

  25. [2023]

    Available: https://advanced.onlinelibrary.wiley.com/doi/ abs/10.1002/advs.202303949

    [Online]. Available: https://advanced.onlinelibrary.wiley.com/doi/ abs/10.1002/advs.202303949

    work page doi:10.1002/advs.202303949