Pith. sign in

REVIEW 4 major objections 6 minor 44 references

SuperMag: Vision-based Tactile Data Guided High-resolution Tactile Shape Reconstruction for Magnetic Tactile Sensors

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Sparse magnetic tactile readings can be converted into dense 200x200 contact-shape maps at 95 Hz by a conditional variational autoencoder trained with vision-based tactile supervision.

desk verdict A novel cross-modal tactile super-resolution idea that is undermined by an internal contradiction: the 'identical' contact modules are not actually identical, so the VBTS ground truth may not represent MBTS deformation. read the letter →

arxiv 2507.20002 v1 pith:WMKWC3A2 submitted 2025-07-26 cs.RO

Peiyao Hou , Danning Sun , Meng Wang , Yuzhe Huang , Zeyu Zhang , Hangxin Liu , Wanlin Li , Ziyuan Jiao This is my paper
classification cs.RO
keywords tactilesensingsuper-resolutionmagneticsensorvision-basedconditionalvariationalautoencodershapereconstructioncross-modallearningroboticmanipulation
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

SuperMag proposes that the low spatial resolution of magnetic tactile sensors is not a hardware ceiling: with a vision-based tactile sensor providing high-resolution depth images of the same contact, a conditional variational autoencoder can learn to turn the sparse magnetic readings into dense 200x200 contact-shape maps. The authors build a magnetic sensor with a 4x4 magnetometer array and a matching vision sensor whose contact module has the same material and dimensions, then mount them on opposite sides of a gripper so that both record the same press. On unseen objects the full model improves PSNR from 20.86 dB (single-axis variant) to 24.24 dB and SSIM from 0.69 to 0.78, and the reconstructed images feed an in-hand pose estimator that corrects object orientation in 6 of 6 trials versus 1 of 6 with a 4x4 baseline. The inference runs at about 95 Hz, preserving the magnetic sensor's speed while gaining the vision sensor's detail.

What carries the argument

The carrier is a conditional variational autoencoder (CVAE), a generative network that models the distribution of possible high-resolution depth maps given a magnetic reading. The encoder takes the vision-sensor depth image and embeds the magnetic condition through multiplicative conditioning at two stages; the decoder samples a latent variable z and produces the reconstructed image, with the condition injected the same way. Training maximizes the evidence lower bound (ELBO), balancing reconstruction fidelity against a KL term that keeps the encoder near the prior. The physical enabler is the symmetric calibration setup: the magnetic sensor, a 4x4 array of three-axis magnetometers under a magnetized silicone contact pad, and the vision sensor, a camera-based device measuring contact depth from the darkness of a semi-transparent elastomer, are mounted on opposite sides of a gripper. Their contact modules have identical material and thickness, so a 200x200 depth map and a 4x4x3 magnetic vector reading are co-located and synchronized while the gripper presses with a 10 N force threshold.

What would settle it

Press a flat-faced object onto the magnetic sensor while independently measuring the contact shape with a calibrated high-resolution reference, and compare the reconstructed depth to that reference; the authors already state that vision-based depth is unreliable for large planar surfaces, so SuperMag should fail there. A second check is to introduce a small shear or rotation during the symmetric press, breaking the co-location assumption, and measure whether SSIM on held-out objects drops.

Watch

Extended reading notes

Core claim

The central claim is that a sparse 4x4x3 magnetic array contains enough information to generate a 200x200 depth image of the contact surface, provided the mapping is learned as a conditional generative model rather than an interpolation or a direct regression. The paper frames the reconstruction as learning the conditional distribution p(I|x) of vision-sensor depth images I given the magnetic reading x, which allows multiple plausible shapes for a single measurement and handles the inherent ambiguity of cross-modal mapping. The evidence is a comparison on unseen objects: bilinear and bicubic interpolation score below 10 dB PSNR and 0.10 SSIM, the z-axis-only model reaches 20.86 dB and 0.69, the single-object-trained model reaches 22.36 dB and 0.65, and the full three-axis, multi-object SuperMag reaches 24.24 dB and 0.78, with the best FID of 210.10. The paper further claims that the reconstructed shapes carry enough geometry for a downstream pose-estimation task, where super-resolution enables perfect angle correction in a small reorientation trial.

Load-bearing premise

The load-bearing premise is that the symmetric calibration setup makes the magnetic and vision sensors experience exactly the same contact, so the vision depth map is a valid label for the magnetic reading; any misalignment, timing offset, or systematic depth error from the vision sensor would train the model to reproduce that error instead of true contact shape.

Editorial extensions

If this is right

  • MBTS hardware, unchanged, can report dense 200x200 depth maps at 95 Hz, making sparse magnetic arrays practical for high-precision contact tasks.
  • The 100% versus 16.7% success in the in-hand orientation trial suggests super-resolved tactile images carry enough geometric signal for closed-loop manipulation.
  • Because training needs only two simple objects (an Allen key and a letter 'R'), the data pipeline is cheap and reproducible with the open-source sensor designs.
  • Single-axis magnetic data is a real bottleneck: the z-axis-only variant lags the full three-axis model, so any magnetic sensor providing only normal force would need to be augmented to reach this quality.
  • The approach is tied to the matching contact module; sensors with different dimensions or materials would need re-calibration or domain adaptation.

Reading between the lines

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

  • A learned alignment step could replace the symmetric calibration, allowing the method to transfer to arbitrary magnetic/vision sensor pairs with different contact modules, which the authors list as future work.
  • The CVAE's latent sampling means one magnetic reading yields several plausible shapes; a robot could treat the spread as contact uncertainty and plan conservative grasps, something the paper does not explore.
  • The same cross-modal supervision idea should extend to other sparse taxel modalities (capacitive, piezoresistive) as long as a co-located vision sensor can be mounted, provided the contact mechanics are matched.
  • The reported numbers are on two training objects and a handful of unseen objects; scaling to a larger object library is the natural stress test for the generalization claim.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes SuperMag, a conditional variational autoencoder (CVAE) that reconstructs high-resolution tactile shape images (200×200 pixels) from sparse magnetic-based tactile sensor (MBTS) readings (4×4×3 taxel arrays). The supervision signal is provided by paired data from a co-designed vision-based tactile sensor (VBTS), collected through a symmetric calibration setup that presses an object between the two sensors. The authors report that SuperMag outperforms bilinear/bicubic interpolation and its own ablations on FID, PSNR, and SSIM on unseen objects, and demonstrate an in-hand pose estimation / object reorientation task. The claimed inference time is 2.5 ms, enabling roughly 95 Hz operation. The paper includes open-source sensor designs and acknowledges several limitations, including inferior fine-detail reconstruction relative to VBTS and the VBTS's weakness on planar surfaces.

Significance. If the results hold, the cross-modal supervision idea is a worthwhile contribution: using high-resolution VBTS depth as a training signal to upgrade MBTS from sparse taxels to dense shape reconstruction could help bridge the resolution-versus-frequency trade-off in tactile sensing. The open-source hardware designs, the explicit formulation of the problem as conditional generation, and the reported inference speed are concrete strengths. However, the central claim of 'identical contact modules' is contradicted by the paper's own specifications, which undermines the validity of the VBTS depth labels as ground truth for MBTS deformation. The evaluation also lacks state-of-the-art learning-based baselines and has very limited pose-estimation trials. These issues prevent the current evidence from fully supporting the conclusions, although they appear addressable with additional experiments and careful re-framing.

major comments (4)
  1. [Section III-A, III-B, III-C] The paper asserts in Section III-A that both sensors 'must have contact modules with identical dimensions and materials,' but the specifications in Sections III-B and III-C do not satisfy this. The MBTS module is 20×20×2.5 mm of magnetized silicone (Ecoflex 00-30 with MQP-15-7 powder), while the VBTS module is 21×21×2.5 mm with a PDMS base, a soft silicone layer, and a black coating. These differ in size, composition, and stiffness. Under the same 10 N gripper force, the two surfaces will deform differently, so the VBTS depth map records the deformation of the VBTS contact surface, not the MBTS contact surface. Since the CVAE is trained with VBTS depth as the target for MBTS input, this bias propagates into the reconstruction. The paper provides no calibration experiment showing that the two surfaces produce the same depth profile for a known indenter. This is a load-bearing issue: please either validate that the deformation fields are equivalent (e.g., with a calibration indenter comparing depth profiles), use the same physical contact module for both modalities, or explicitly model the material/geometric mismatch and show it does not affect the reported metrics.
  2. [Section V-B, Table I] The baselines in Table I are limited to bilinear/bicubic interpolation of z-axis MBTS data and two ablations of the proposed method. No comparison is made to existing learning-based tactile super-resolution methods, e.g., the MLP/kNN/CNN approaches cited in Section II-B or the GAN-based method of Wu et al. [16]. The claim that SuperMag 'outperforms all baselines' is therefore only relative to naive interpolation and to themselves. To support the central contribution, please add at least one deterministic regression baseline (e.g., a U-Net or CNN trained on the same paired data) and, if feasible, a prior learning-based tactile SR method adapted to this data. This would clarify whether the CVAE formulation itself is beneficial rather than merely the use of multi-axis MBTS data.
  3. [Section V-C] The in-hand pose estimation experiment uses only six trials in total (two random orientations per object). The 100% vs 16.7% success rate corresponds to 6/6 vs 1/6, which is not statistically reliable; a single failure changes the conclusion. Please report more trials per object with confidence intervals or a statistical test, and specify the number of trials explicitly. The angular error distribution (not just binary success) would also be more informative than a coarse success rate.
  4. [Section V-A, V-B] The data split and test-set composition are not described. The text says 4050 total pairs (2025 per training object) and 3645 training samples, but no validation/test split is given, and the set of unseen test objects is not enumerated in a table or list. The FID metric, in particular, is sensitive to the number of samples used; reporting FID on an unspecified and likely small test set makes the number difficult to interpret. Please report the number of test objects, the number of trials per object, the exact split, and the number of images used for each metric (FID, PSNR, SSIM).
minor comments (6)
  1. [Section VI] The limitation paragraph says the method is constrained to MBTS sensors with contact modules that 'match the dimensions and silicone material of the VBTS,' but the VBTS contact module also includes a PDMS base and a black coating. This sentence should be corrected to reflect the actual materials described in Section III-C.
  2. [Section V-A] The dataset description reports 2025 pairs per object and 3645 training samples total, leaving 405 pairs unaccounted for. Clarify how many samples are used for validation and test, and whether the test objects are entirely disjoint from the training objects.
  3. [Section IV-B] The architecture is described as 'lightweight' but has 47.2M parameters, which is substantial. Please remove the 'lightweight' characterization or justify it relative to typical generative models; also specify the number of floating-point operations if the claim is about computational efficiency.
  4. [Section III-B] The ratio '3:1:1' for mixing magnetic powder with silicone is ambiguous (powder:part A:part B?). Please specify the exact formulation and the mass/volume units.
  5. [Abstract and Section I] The phrase 'identical contact modules' appears in the abstract and introduction, but the papers' own details contradict it. Please revise the wording to 'matched as closely as possible' or 'co-designed to approximate matching,' or adjust the hardware design to make the modules truly identical.
  6. [General] The FID metric is computed with features from an ImageNet-trained Inception network. Tactile depth images are far from natural images, so the perceptual meaning of FID here is unclear. Consider adding a task-specific metric (e.g., contact-area IoU or edge error) or at least discuss the validity of FID for this domain.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SuperMag is a supervised generative mapping fitted to paired MBTS/VBTS data, not a derivation equivalent to its inputs.

full rationale

The central claim is that a conditional VAE learns p_theta(I|x) from paired MBTS readings x and VBTS depth images I (Eqs. 1-2), with the network trained by maximizing the ELBO. The reconstructed image is therefore a fitted prediction, not an analytic consequence of the magnetic input. The metrics are computed against the same VBTS modality used as supervision, but this is standard supervised evaluation, not a reduction of the output to the input. The paper relies on the external ReSkin and DTact sensor designs for hardware and adopts a standard CVAE architecture; no load-bearing premise is justified only by self-citation. There is an internal inconsistency: Section III-A requires identical contact modules, while Sections III-B and III-C specify 20x20x2.5 mm magnetized silicone versus 21x21x2.5 mm PDMS/silicone/coating modules, and Section VI admits that VBTS cannot detect large planar surfaces. These are empirical validity threats to the ground-truth co-location assumption, not circularity. No equation or fitted parameter is equivalent to the predicted target by construction.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim depends on the learned neural mapping and on domain assumptions about sensor alignment and ground-truth fidelity. The model's 47.2M parameters are the main free parameters; no new physical entities are introduced.

free parameters (3)
  • CVAE network weights = 47.2M parameters learned via AdamW
    The central reconstruction mapping is entirely learned from the paired dataset; the weights are fitted to the 3645 training samples.
  • Latent dimension = 512
    Chosen by hand to balance reconstruction accuracy and generative diversity; not derived from theory.
  • Learning rate and epochs = 1e-4, 350 epochs
    Chosen by hand; standard hyperparameters for training.
assumptions (4)
  • domain assumption VBTS depth images are accurate estimates of the true contact geometry
    The paper uses VBTS depth as ground truth without independent validation of its accuracy (Section IV).
  • domain assumption The symmetric calibration setup ensures identical contact position and force between MBTS and VBTS during data collection
    Section III-D states the sensors are mounted on opposite sides of a gripper; no quantitative verification of co-location is provided.
  • domain assumption Training on two objects (Allen key and letter R) is sufficient to learn a mapping that generalizes to unseen objects
    The paper tests on unseen objects but does not quantify the diversity or similarity of those objects.
  • standard math Conditional VAE and reparameterization trick are valid for modeling the conditional distribution p(I|x)
    Standard machine learning techniques; no proof of applicability to this domain.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SuperMag: Vision-based Tactile Data Guided High-resolution Tactile Shape Reconstruction for Magnetic Tactile Sensors." pith.science (2026). https://pith.science/paper/WMKWC3A2

@misc{pith2026250720002,
  author       = {Pith},
  title        = {Pith review of: SuperMag: Vision-based Tactile Data Guided High-resolution Tactile Shape Reconstruction for Magnetic Tactile Sensors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WMKWC3A2}},
  note         = {Machine review of arXiv:2507.20002}
}
read the original abstract

Magnetic-based tactile sensors (MBTS) combine the advantages of compact design and high-frequency operation but suffer from limited spatial resolution due to their sparse taxel arrays. This paper proposes SuperMag, a tactile shape reconstruction method that addresses this limitation by leveraging high-resolution vision-based tactile sensor (VBTS) data to supervise MBTS super-resolution. Co-designed, open-source VBTS and MBTS with identical contact modules enable synchronized data collection of high-resolution shapes and magnetic signals via a symmetric calibration setup. We frame tactile shape reconstruction as a conditional generative problem, employing a conditional variational auto-encoder to infer high-resolution shapes from low-resolution MBTS inputs. The MBTS achieves a sampling frequency of 125 Hz, whereas the shape reconstruction sustains an inference time within 2.5 ms. This cross-modality synergy advances tactile perception of the MBTS, potentially unlocking its new capabilities in high-precision robotic tasks.

Figures

Figures reproduced from arXiv: 2507.20002 by the authors.

Figure 1
Figure 1. SuperMag: High-resolution Tactile Shape Reconstruction for Magnetic-based Tactile Sensors (MBTS) with Vision-based Tactile Sensors (VBTS) data. (a) Training: High-Resolution (HR) VBTS depth images serve as supervisory signals to guide Low￾Resolution (LR) MBTS in reconstructing high-resolution tactile shapes of the object. (b) Inference: Sparse MBTS data are used to reconstruct the tactile shape of an unseen test obj… view at source ↗
Figure 2
Figure 2. Exploded and schematics views demonstrating the key components of the two tactile sensors examined in this work. (a) The 9 mm-thick MBTS [20] features a magnetized contact module, mounted over a circuit board with 4×4 magnetometers spaced 5 mm apart. (b) The 27 mm-thick VBTS [21] uses a camera module embedded beneath an acrylic window and illumination ring, together with a contact module composed of a coating layer … view at source ↗
Figure 4
Figure 4. Network architecture of SuperMag. A. Problem Formulation We consider the MBTS readings as a condition x, which are typically vectors or spatially organized signals from magnetometers. Our goal is to learn a mapping x 7→I, where I represents the corresponding reconstructed depth image from VBTS. Formally, we aim to learn the conditional distribution pθ (I | x). In a CVAE, this distribution is represented using a late… view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Shape reconstruction results for ground truth, baselines, and SuperMag on seen training object and unseen testing objects. The ground truth is from the vision-based tactile sensor (VBTS). Baselines are Bilinear and Bicubic interpolation of z-axis magnetic-based tactile…
Figure 6
Figure 6. Figure 6: In-hand Pose Estimation for Object Reorientation. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Example results of SuperMag shape reconstruction for [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 43 canonical work pages

  1. [16]

    Tactile pattern super resolution with taxel-based sensors,

    B. Wu, Q. Liu, and Q. Zhang, “Tactile pattern super resolution with taxel-based sensors,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 3644–3650, IEEE, 2022

    work page 2022

  2. [1]

    Force and tactile sensors,

    M. R. Cutkosky, R. D. Howe, and W. R. Provancher, “Force and tactile sensors,” Springer Handbook of Robotics , pp. 455–476, 2008

    work page 2008

  3. [2]

    Tactile sensing for dexterous in-hand manipulation in robotics—a review,

    H. Yousef, M. Boukallel, and K. Althoefer, “Tactile sensing for dexterous in-hand manipulation in robotics—a review,” Sensors and Actuators A: physical , vol. 167, no. 2, pp. 171–187, 2011

    work page 2011

  4. [3]

    Learning decoupled multi-touch force estimation, localization and stretch for soft ca- pacitive e-skin,

    A. B. Dawood, C. Coppola, and K. Althoefer, “Learning decoupled multi-touch force estimation, localization and stretch for soft ca- pacitive e-skin,” in IEEE International Conference on Robotics and Automation (ICRA) , pp. 614–619, IEEE, 2023

    work page 2023

  5. [4]

    Recent advances in resistive sensor technology for tactile perception: A review,

    Y . Zhu, Y . Liu, Y . Sun, Y . Zhang, and G. Ding, “Recent advances in resistive sensor technology for tactile perception: A review,” IEEE Sensors Journal, vol. 22, no. 16, pp. 15635–15649, 2022

    work page 2022

  6. [5]

    Flexible tactile sens- ing based on piezoresistive composites: A review,

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

    work page 2014

  7. [6]

    Skin- inspired piezoelectric tactile sensor array with crosstalk-free row+ column electrodes for spatiotemporally distinguishing diverse stimuli,

    W. Lin, B. Wang, G. Peng, Y . Shan, H. Hu, and Z. Yang, “Skin- inspired piezoelectric tactile sensor array with crosstalk-free row+ column electrodes for spatiotemporally distinguishing diverse stimuli,” Advanced Science , vol. 8, no. 3, p. 2002817, 2021

    work page 2021

  8. [7]

    Fingerprint-shaped triboelectric tactile sensor,

    X. Qu, J. Xue, Y . Liu, W. Rao, Z. Liu, and Z. Li, “Fingerprint-shaped triboelectric tactile sensor,” Nano Energy, vol. 98, p. 107324, 2022

    work page 2022

Show all 44 references
  1. [8]

    Location and orientation super- resolution sensing with a cost-efficient and repairable barometric tactile sensor,

    J. Hou, X. Zhou, and A. J. Spiers, “Location and orientation super- resolution sensing with a cost-efficient and repairable barometric tactile sensor,” IEEE Transactions on Robotics (T-RO) , 2024

    2024

  2. [9]

    Recent progress of optical tactile sensors: A review,

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

    2024

  3. [10]

    Recent progress of biomimetic tactile sensing technology based on magnetic sensors,

    J. Man, G. Chen, and J. Chen, “Recent progress of biomimetic tactile sensing technology based on magnetic sensors,” Biosensors, vol. 12, no. 11, p. 1054, 2022

    2022

  4. [11]

    Design methodology for magnetic field-based soft tri-axis tactile sensors,

    H. Wang, G. De Boer, J. Kow, A. Alazmani, M. Ghajari, R. Hewson, and P. Culmer, “Design methodology for magnetic field-based soft tri-axis tactile sensors,” Sensors, vol. 16, no. 9, p. 1356, 2016

    2016

  5. [12]

    Guiding the design of superresolution tactile skins with taxel value isolines theory,

    H. Sun and G. Martius, “Guiding the design of superresolution tactile skins with taxel value isolines theory,”Science Robotics, vol. 7, no. 63, p. eabm0608, 2022

    2022

  6. [13]

    Multipole magnetic tactile sensor with super-resolution for 3d pose estimation,

    J. Huang, Y . Lou, X. Xiong, X. Yang, and Y . Li, “Multipole magnetic tactile sensor with super-resolution for 3d pose estimation,” IEEE Transactions on Instrumentation and Measurement (TIM) , 2024

    2024

  7. [14]

    Soft magnetic skin for super-resolution tactile sensing with force self- decoupling,

    Y . Yan, Z. Hu, Z. Yang, W. Yuan, C. Song, J. Pan, and Y . Shen, “Soft magnetic skin for super-resolution tactile sensing with force self- decoupling,” Science Robotics , vol. 6, no. 51, p. eabc8801, 2021

    2021

  8. [15]

    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. Li, et 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

  9. [17]

    On the design and development of vision-based tactile sensors,

    U. H. Shah, R. Muthusamy, D. Gan, Y . Zweiri, and L. Seneviratne, “On the design and development of vision-based tactile sensors,” Journal of Intelligent & Robotic Systems , vol. 102, pp. 1–27, 2021

    2021

  10. [18]

    Touching a nerf: Leveraging neural radiance fields for tactile sensory data gen- eration,

    S. Zhong, A. Albini, O. P. Jones, P. Maiolino, and I. Posner, “Touching a nerf: Leveraging neural radiance fields for tactile sensory data gen- eration,” in Annual Conference on Robot Learning (CoRL) , pp. 1618– 1628, PMLR, 2023

    2023

  11. [19]

    When vision meets touch: A contemporary review for visuotactile sensors from the signal processing perspective,

    S. Li, Z. Wang, C. Wu, X. Li, S. Luo, B. Fang, F. Sun, X.-P. Zhang, and W. Ding, “When vision meets touch: A contemporary review for visuotactile sensors from the signal processing perspective,” IEEE Journal of Selected Topics in Signal Processing , 2024

    2024

  12. [20]

    Reskin: versatile, replaceable, lasting tactile skins,

    R. Bhirangi, T. Hellebrekers, C. Majidi, and A. Gupta, “Reskin: versatile, replaceable, lasting tactile skins,” in Annual Conference on Robot Learning (CoRL) , 2021

    2021

  13. [21]

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

    C. Lin, Z. Lin, S. Wang, and H. Xu, “Dtact: A vision-based tactile sensor that measures high-resolution 3d geometry directly from dark- ness,” in IEEE International Conference on Robotics and Automation (ICRA), pp. 10359–10366, IEEE, 2023

    2023

  14. [22]

    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

  15. [23]

    A bimodal soft electronic skin for tactile and touchless interaction in real time,

    J. Ge, X. Wang, M. Drack, O. V olkov, M. Liang, G. S. Cañón Bermúdez, R. Illing, C. Wang, S. Zhou, J. Fassbender, et al. , “A bimodal soft electronic skin for tactile and touchless interaction in real time,” Nature communications, vol. 10, no. 1, p. 4405, 2019

    2019

  16. [24]

    Split-type magnetic soft tactile sensor with 3d force decoupling,

    H. Dai, C. Zhang, C. Pan, H. Hu, K. Ji, H. Sun, C. Lyu, D. Tang, T. Li, J. Fu, et al. , “Split-type magnetic soft tactile sensor with 3d force decoupling,” Advanced Materials , vol. 36, no. 11, p. 2310145, 2024

    2024

  17. [25]

    A tactile sensor based on magnetic sensing: Design and mechanism,

    J. Li, H. Qin, Z. Song, L. Hou, and H. Li, “A tactile sensor based on magnetic sensing: Design and mechanism,” IEEE Transactions on Instrumentation and Measurement (TIM) , 2024

    2024

  18. [26]

    A novel approach to real-time bilinear interpolation,

    K. T. Gribbon and D. G. Bailey, “A novel approach to real-time bilinear interpolation,” in Proceedings. DELTA 2004. Second IEEE international workshop on electronic design, test and applications , pp. 126–131, IEEE, 2004

    2004

  19. [27]

    Wireless flexible magnetic tactile sensor with super-resolution in large-areas,

    H. Hu, C. Zhang, C. Pan, H. Dai, H. Sun, Y . Pan, X. Lai, C. Lyu, D. Tang, J. Fu, et al. , “Wireless flexible magnetic tactile sensor with super-resolution in large-areas,” ACS nano, vol. 16, no. 11, pp. 19271– 19280, 2022

    2022

  20. [28]

    Superresolution with an optical tactile sensor,

    N. F. Lepora and B. Ward-Cherrier, “Superresolution with an optical tactile sensor,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 2686–2691, IEEE, 2015

    2015

  21. [29]

    Implementing tactile behaviors us- ing fingervision,

    A. Yamaguchi and C. G. Atkeson, “Implementing tactile behaviors us- ing fingervision,” in IEEE/RAS International Conference on Humanoid Robotics (Humanoids) , pp. 241–248, IEEE, 2017

    2017

  22. [30]

    Self- supervised contact geometry learning by gelstereo visuotactile sens- ing,

    S. Cui, R. Wang, J. Hu, C. Zhang, L. Chen, and S. Wang, “Self- supervised contact geometry learning by gelstereo visuotactile sens- ing,” IEEE Transactions on Instrumentation and Measurement (TIM) , vol. 71, pp. 1–9, 2021

    2021

  23. [31]

    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

  24. [32]

    Finger-shaped gelforce: sensor for measuring surface traction fields for robotic hand,

    K. Sato, K. Kamiyama, N. Kawakami, and S. Tachi, “Finger-shaped gelforce: sensor for measuring surface traction fields for robotic hand,” IEEE Transactions on Haptics (ToH) , vol. 3, no. 1, pp. 37–47, 2009

    2009

  25. [33]

    Large-scale deployment of vision-based tactile sensors on multi- fingered grippers,

    M. Wang, W. Li, H. Liang, B. Li, K. Althoefer, Y . Su, and H. Liu, “Large-scale deployment of vision-based tactile sensors on multi- fingered grippers,” in IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS) , pp. 13946–13952, IEEE, 2024

    2024

  26. [34]

    Minitac: An ultra-compact 8 mm vision-based tactile sensor for enhanced palpation in robot-assisted minimally invasive surgery,

    W. Li, Z. Zhao, L. Cui, W. Zhang, H. Liu, L.-A. Li, and Y . Zhu, “Minitac: An ultra-compact 8 mm vision-based tactile sensor for enhanced palpation in robot-assisted minimally invasive surgery,” Robotics and Automation Letters (RA-L) , 2024

    2024

  27. [35]

    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

  28. [36]

    Gelslim: A high-resolution, compact, robust, and calibrated tactile- sensing finger,

    E. Donlon, S. Dong, M. Liu, J. Li, E. Adelson, and A. Rodriguez, “Gelslim: A high-resolution, compact, robust, and calibrated tactile- sensing finger,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pp. 1927–1934, IEEE, 2018

    1927

  29. [37]

    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, et al. , “Digit: A novel design for a low-cost compact high-resolution tactile sensor with application to in-hand manipulation,” Robotics and Automation Letters (RA...

    2020

  30. [38]

    A fast poisson solver for complex geometries,

    A. McKenney, L. Greengard, and A. Mayo, “A fast poisson solver for complex geometries,” Journal of Computational Physics , vol. 118, no. 2, pp. 348–355, 1995

    1995

  31. [39]

    Measurement of shear and slip with a gelsight tactile sensor,

    W. Yuan, R. Li, M. A. Srinivasan, and E. H. Adelson, “Measurement of shear and slip with a gelsight tactile sensor,” in IEEE International Conference on Robotics and Automation (ICRA) , pp. 304–311, IEEE, 2015

    2015

  32. [40]

    L3f-touch: A wireless gelsight with decoupled tactile and three-axis force sensing,

    W. Li, M. Wang, J. Li, Y . Su, D. K. Jha, X. Qian, K. Althoefer, and H. Liu, “L3f-touch: A wireless gelsight with decoupled tactile and three-axis force sensing,” Robotics and Automation Letters (RA-L) , 2023

    2023

  33. [41]

    Tac- man: Tactile-informed prior-free manipulation of articulated objects,

    Z. Zhao, Y . Li, W. Li, Z. Qi, L. Ruan, Y . Zhu, and K. Althoefer, “Tac- man: Tactile-informed prior-free manipulation of articulated objects,” IEEE Transactions on Robotics (T-RO) , 2024

    2024

  34. [42]

    Deep learning for image super- resolution: A survey,

    Z. Wang, J. Chen, and S. C. Hoi, “Deep learning for image super- resolution: A survey,” Transactions on Pattern Analysis and Machine Intelligence (TPAMI), vol. 43, no. 10, pp. 3365–3387, 2020

    2020

  35. [43]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”Trans- actions on Image Processing (TIP) , vol. 13, no. 4, pp. 600–612, 2004

    2004

  36. [44]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” Advances in Neural Information Processing Systems (NIPS), vol. 30, 2017

    2017

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.