Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Taccel: Scaling Up Vision-based Tactile Robotics via High-performance GPU Simulation

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A GPU simulator for tactile robots runs 18x faster than real time and transfers touch skills to real robots.

desk verdict Taccel is a genuinely useful GPU tactile simulator with real sim-to-real validation, but the headline speedup is from one low-res task, the baseline comparison is not clean, and the gel-pad parameters are under-documented. read the letter →

arxiv 2504.12908 v2 pith:IPQXCX3G submitted 2025-04-17 cs.RO cs.CV

Yuyang Li , Wenxin Du , Chang Yu , Puhao Li , Zihang Zhao , Tengyu Liu , Chenfanfu Jiang , Yixin Zhu
show 1 more author
Siyuan Huang
This is my paper · ORCID
classification cs.ROcs.CV
keywords tactilesimulationvision-basedsensorsGPU-acceleratedphysicsincrementalpotentialcontactaffinebodydynamicssim-to-realtransfersoftroboticsGelSight
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

Taccel aims to make vision-based tactile simulation fast enough and accurate enough to serve as a practical research tool. Its central claim is that combining incremental potential contact (IPC) with affine body dynamics (ABD) on a GPU models the soft gel pad of a GelSight-type sensor, the robot links, and the objects they touch with physically valid, intersection-free contact while running about 18 times faster than real time across thousands of parallel environments. The paper backs this with realistic tactile images (average SSIM 0.93 against real presses), about 28-micron sim-real error in shear-deformation tracking, and successful sim-to-real transfer in object classification, grasping, and articulated object manipulation. If the claim holds, the bottleneck that has confined tactile robotics to small-scale studies is removed: researchers could generate large synthetic tactile datasets and test hand designs before building hardware.

What carries the argument

The load-bearing object is the unified ABD-IPC incremental potential, written as $$E_{\mathrm{IPC}}(y;x) = E_{\mathrm{IP}}(x) + E_{\mathrm{IP}}(y) + \$\Delta$ $t^{2}$ B(\$\varphi$(y);x) + \$\Delta$ $t^{2}$ D(\$\varphi$(y);x,\$\varphi$(y_n);x_n),$$ minimized at every timestep. Here $x$ collects the soft gel-pad vertex positions, $y$ collects the 12-degree-of-freedom affine-body coordinates (translation plus a $3\times 3$ affine map) for robot links and stiff objects, $B$ is the IPC log-barrier contact potential that grows without bound as surface primitives approach each other, and $D$ is the friction potential. An as-rigid-as-possible term with large stiffness keeps affine bodies nearly rigid, while the Neo-Hookean finite-element gel captures compliance. Solving this coupled system with a GPU-parallelized solver is what lets the simulator run thousands of environments while preserving intersection- and inversion-free contact.

What would settle it

Run the paper's shear-deformation experiment (a gripper with a GelSight-type sensor grasps a fixed bar and pulls back at 2 mm/s) at pull speeds from 0.5 to 50 mm/s. If simulated marker-flow trajectories diverge from real ones by much more than the reported about 28 $\mu$m average error at the higher speeds, the elastic gel model with one-record friction calibration fails for dynamic contacts.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that fidelity and speed need not be traded off if stiff parts are reduced to affine bodies while only the sensor gel is simulated as a full soft body, and both are co-solved in one barrier-augmented incremental potential. Each affine body carries 12 degrees of freedom (translation plus a 3x3 affine map), which keeps robot links and objects nearly rigid at low cost; the gel pad is a tetrahedral Neo-Hookean finite-element solid whose deformed coated surface is rendered into RGB images, depth maps, and marker flows. The result is reported 18-fold acceleration over real time with over 4096 parallel environments on a single GPU, and the paper demonstrates that the simulated tactile signals transfer to real robots: classification on real mechanical parts, grasping with four different hand designs, and Tac-Man execution-recovery counts that match reality to about 1 percent error.

Load-bearing premise

The claim rests on assuming that a real GelSight-type gel pad deforms exactly like the simulated Neo-Hookean elastic solid with material constants taken from prior work and friction calibrated from a single pull record, across every contact speed, load, and object shape.

Editorial extensions

If this is right

  • Synthetic tactile datasets for training perception models can be produced at scale: a reported 4096 parallel peg-insertion environments run at about 915 FPS on a single GPU, making tens of thousands of contact images cheap to generate.
  • Hand and sensor designs can be validated before fabrication: the grasping experiments across gripper, three-finger, Allegro, and full-hand-with-17-gel-pads configurations show measurable differences in contact area and classification accuracy.
  • Visuotactile sim-to-real transfer is achievable without domain adaptation in a controlled setting: a model trained only on simulated depth maps reaches about 71 percent accuracy on real mechanical parts.
  • Contact-rich articulation policies such as Tac-Man can be tuned and evaluated in simulation: simulated execution-recovery switch counts match real-world counts to within about 1 percent error for the tested microwave and drawer.

Reading between the lines

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

  • We infer the least-tested regime is dynamic contact: the gel is modeled as a time-integrated elastic solid with material constants from prior work and friction calibrated from one record, so rate-dependent effects like viscoelastic creep or speed-dependent slip are not captured; a high-speed shear test would expose this gap.
  • We infer the tactile renderer could become differentiable end-to-end: the RGB image is produced by a trained neural network from depth, so back-propagating through both physics and rendering would enable gradient-based grasp or gel-pad design optimization, something the paper does not explore.
  • We infer that the single large stiffness parameter for affine bodies hides an implicit rigidity assumption; calibrating it automatically against real link compliance would broaden the range of materials the simulator can represent.
  • We infer the simulator's coverage of tasks is likely to be stress-tested by in-hand manipulation and fast slip, exactly the scenarios where the quasi-static gel assumption and the one-record friction calibration would first show discrepancies.
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

3 major / 5 minor

Summary. The manuscript presents Taccel, a GPU-based simulation platform for vision-based tactile sensors (VBTSs) that combines Incremental Potential Contact (IPC) for soft gel pads with Affine Body Dynamics (ABD) for stiff robot links and objects. It claims precise physics simulation, realistic tactile signal generation (RGB, depth, markers), flexible robot/sensor configuration, and massive parallelization, with an 18x speedup over real-time on a low-resolution peg-insertion task. Validation includes SSIM comparison on 18 pressed objects, shear-deformation trajectory errors after calibrating friction on one record, sim-to-real object classification on 160 real depth maps, and Tac-Man execution-recovery switch counts on a microwave and a drawer. The paper argues Taccel can scale up tactile robotics research by enabling large synthetic datasets and pre-fabrication hand design evaluation.

Significance. If the results hold, Taccel would be a valuable contribution to tactile robotics because it offers a rare combination of physics-based soft-body simulation, high-resolution tactile signal synthesis, and GPU-parallel environments. The concrete external validations are a strength: SSIM of 0.93 on 18 objects, 70.94% sim-to-real classification accuracy on 160 real samples, and Tac-Man switch counts of 68.75 simulated versus 68 real are specific, reproducible claims. The underlying IPC/ABD formulation is standard and the math in Section 3 and Appendix A is sound. The main open risk is whether the physical fidelity is robust to the unreported or lightly calibrated material parameters; the paper's broad claims of precise simulation and successful transfer are currently supported only within a narrow set of objects and tasks. I therefore regard the contribution as potentially significant but needing revision to match the evidence.

major comments (3)
  1. [Sec. 5.4, Fig. 4, Abstract] The headline '18-fold acceleration over real-time across thousands of parallel environments' is measured on the low-resolution peg-insertion task with 139-node gel pads (Sec. 5.4, Fig. 4). The higher-resolution dexterous-hand task achieves only 12.67 FPS, i.e., 0.25x wallclock time. The abstract and introduction state the 18x figure without this caveat, which overstates the platform's general performance. Please qualify the speed claim to the specific task/resolution or add benchmarks that support a broader statement. In addition, Fig. 4 compares Taccel in FP64 against SAPIEN-IPC in FP32; because the authors attribute baseline convergence failures to FP32, the comparison should be repeated with matched numerical precision or the discrepancy should be discussed in the text.
  2. [Sec. 3.3, A.4, 5.3] The ARAP stiffness kappa_s is described only as 'large' and its numerical value is never reported; the Neo-Hookean gel parameters E and nu are taken from prior work (Sec. C.1), and the friction coefficient is calibrated from a single record (Sec. 5.3). No sensitivity analysis is provided for these parameters. Because SSIM (0.93), classification accuracy (70.94%), and Tac-Man switch counts (68.75 vs. 68) are the primary evidence for physical fidelity, the authors should report concrete values for kappa_s, E, nu, and mu, and quantify how sensitive these metrics are to each parameter. Without this, the claim of precise physics simulation across tasks is not established beyond the specific tested objects and loads.
  3. [Sec. 6.1, 6.3] The sim-to-real evidence is limited to 10 mechanical parts, 18 pressed objects, and two articulated objects (a microwave and a drawer). The drawer case yields zero switch counts in both real and simulated execution, so the distinguishing validation is effectively the microwave case. The paper should either add more diverse articulated-object and loading scenarios, or soften the broad claims of successful sim-to-real transfer so that they match the demonstrated operating envelope.
minor comments (5)
  1. [Sec. 3] The sentence 'For complete derivations, we refer readers to Sec. 3 and the original works' appears at the start of Section 3 itself; it should refer to Appendix A.
  2. [Sec. 5.2] The word 'perpendicullarly' is a typo and should be 'perpendicularly'.
  3. [Fig. 2 caption] The phrase 'a so़ block pressing test' contains a rendering artifact; it should read 'a soft block pressing test'.
  4. [Sec. 4.2] The RGB synthesis DNN is trained on patches from 200 real tactile images; please state whether these training images come from the same sensor model used in the Section 5 evaluations, and how the network is expected to transfer to different sensor geometries.
  5. [Fig. 4] The label 'Dext Hand' should be expanded to 'Dexterous Hand' for clarity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's sim-to-real claims are validated against held-out real data, and self-citations refer to independently established simulation methods.

full rationale

The paper's central claims are supported by external benchmarks rather than by construction. In Sec. 5.2, simulated tactile signals are compared against real measurements on 18 objects from a standard dataset, and the reported SSIM is a held-out generalization result, not a re-statement of training data. In Sec. 6.1, a ResNet-18 is trained on simulated depth maps and evaluated on 160 real-world depth maps without domain adaptation, so the 70.94% real accuracy is an independent transfer result. In Sec. 6.3, Tac-Man switch counts are compared against real-world execution with policy parameters (delta_0 = 0.4 mm, alpha = 0.6) taken from the original Tac-Man work, not fit to the reported real counts. The only calibrated quantity in Sec. 5.3 is the friction coefficient, which is fit on one record and then checked on two other records; this is a legitimate calibration/validation split and is not a fitted parameter being renamed as a prediction. The RGB tactile signal generator is a DNN trained on 200 real tactile images (Sec. 4.2), but the SSIM evaluation is performed on a different, standard shape dataset, and the depth and marker signals are produced directly from the simulated deformation field rather than learned. The paper does rely on self-citations for the underlying IPC [30], ABD [29], and unified Newton barrier method [9], but these are established, independently published methods with their own external validations, so the self-citations are not load-bearing circular support. The unspecified ARAP stiffness kappa_s and the single-record friction calibration are legitimate fidelity and robustness concerns, but they are not circularity: no central prediction reduces by construction to a fitted input or to an unverified self-citation chain. Overall, the derivation and validation chain is self-contained against real-world data and independent baselines.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claims rest on inherited modeling choices rather than parameters derived here. The gel pad material constants and affine-body stiffness are not reported, the friction coefficient is calibrated to one record, the RGB network is a fitted component, and IPC solver hyperparameters are inherited from prior work. These are not invented entities, but their values matter for precision claims and the paper does not quantify sensitivity to them.

free parameters (5)
  • Gel pad material parameters (Young's modulus E, Poisson's ratio nu) = Not reported; assumed from prior literature
    The Neo-Hookean model in Sec. 3.1/A.1 uses E and nu to define elastic energy; the paper does not state their values or measure them for its GelSight-type sensor. All deformation and tactile signal fidelity claims depend on these.
  • Affine body ARAP stiffness kappa_s = Not reported; described only as large
    Robot links and stiff objects are modeled as affine bodies whose deformation is penalized by ARAP with high stiffness; the specific value is not given, and it controls how rigid links are under contact loads.
  • Friction coefficient mu (calibrated) = Calibrated to one real record in Sec. 5.3
    In the shear/friction validation, the object friction coefficient is fit to one experimental record and then checked on two others; this is a legitimate calibration, but it is a fitted parameter of the sim-to-real pipeline.
  • RGB synthesis DNN parameters theta = Trained on 200 real tactile image patches
    RGB tactile images are generated by a pixel-to-pixel DNN mapping depth/normal to color (Sec. 4.2). The DNN's parameters are fitted to real data, so RGB realism is not purely physics-derived.
  • IPC/ABD solver hyperparameters (contact stiffness kappa, barrier threshold d_hat, velocity threshold epsilon_v) = Not reported; inherited from prior IPC works
    These control contact stiffness, barrier activation distance, and friction transition; their values affect the benchmark numbers but are not swept or stated, so their influence on the reported speeds and accuracies is unquantified.
assumptions (6)
  • domain assumption Neo-Hookean hyperelasticity adequately models the GelSight gel pad.
    Sec. 3.1 and A.1 define elastic energy using a Neo-Hookean model with Young's modulus E and Poisson's ratio nu; the paper does not include viscoelastic or rate-dependent terms, yet all deformation fidelity claims depend on this choice.
  • domain assumption Affine body dynamics with ARAP and large stiffness kappa_s accurately approximates rigid robot links and stiff objects.
    Sec. 3.3/A.4 model links and stiff objects with 12-DoF affine bodies plus an ARAP penalty with large kappa_s; the stiffness value is not reported and no convergence study is given.
  • standard math IPC barrier and friction potentials are valid for the coupled soft-affine system.
    Sec. 3.2/A.3 adopt IPC contact energies from Li et al. [30] and the unified form from Chen et al. [9]; these are established methods, but their guarantees carry over only if the solver settings stay in the intended regime.
  • domain assumption The depth-to-RGB DNN trained on 200 real image patches generalizes to all simulated depth maps, objects, and lighting conditions.
    Sec. 4.2 generates RGB signals using a pixel-to-pixel DNN following DiffTactile; 200 patches is a small training set and the paper does not quantify the DNN's own generalization error.
  • domain assumption The simulated object meshes and robot/sensor models match the real 3D-printed objects and hardware used in validation.
    Sec. C.1/C.3 describe 3D printing at 0.2 mm layer height and manual URDF creation for Tac-Man; any geometric mismatch directly reduces the meaningfulness of SSIM and switch-count comparisons.
  • domain assumption A fixed timestep Delta_t = 0.02 s keeps all benchmarked tasks stable and converged.
    Fig. 4 reports real wallclock time with Delta_t = 0.02 s and Sec. 7 notes convergence tolerance relaxation as future work; no timestep sensitivity study is included.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Taccel: Scaling Up Vision-based Tactile Robotics via High-performance GPU Simulation." pith.science (2026). https://pith.science/paper/IPQXCX3G

@misc{pith2026250412908,
  author       = {Pith},
  title        = {Pith review of: Taccel: Scaling Up Vision-based Tactile Robotics via High-performance GPU Simulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IPQXCX3G}},
  note         = {Machine review of arXiv:2504.12908}
}
read the original abstract

Tactile sensing is crucial for achieving human-level robotic capabilities in manipulation tasks. As a promising solution, Vision-Based Tactile Sensors (VBTSs) offer high spatial resolution and cost-effectiveness, but present unique challenges in robotics for their complex physical characteristics and visual signal processing requirements. The lack of efficient and accurate simulation tools for VBTSs has significantly limited the scale and scope of tactile robotics research. We present Taccel, a high-performance simulation platform that integrates IPC and ABD to model robots, tactile sensors, and objects with both accuracy and unprecedented speed, achieving an 18-fold acceleration over real-time across thousands of parallel environments. Unlike previous simulators that operate at sub-real-time speeds with limited parallelization, Taccel provides precise physics simulation and realistic tactile signals while supporting flexible robot-sensor configurations through user-friendly APIs. Through extensive validation in object recognition, robotic grasping, and articulated object manipulation, we demonstrate precise simulation and successful sim-to-real transfer. These capabilities position Taccel as a powerful tool for scaling up tactile robotics research and development, potentially transforming how robots interact with and understand their physical environment.

Figures

Figures reproduced from arXiv: 2504.12908 by the authors.

Figure 1
Figure 1. Taccel demonstration of tactile robotics simulation. An Allegro Hand with four VBTSs performing a pre￾cision grasp on a mahjong tile. The de￾formation map precisely captures the tile’s surface geometries. An ideal VBTS simulator must simultaneously achieve: • Precision: Precise modeling of robots, sensors, and ob￾jects with physically valid solutions, particularly main￾taining inversion-free and intersection-free st… view at source ↗
Figure 2
Figure 2. Comprehensive evaluation of physics simulation capabilities across VBTS simulators. (a) Bolt￾nut assembly involving contact between non-convex objects, where Taccel achieves stable simulation; (b) Soft block pressing test with soft-soft contacts, where Taccel maintains penetration-free interactions; (c) Tactile-informed articulated object manipulation, where Taccel replicates real-world interactions with 1%˜ physica… view at source ↗
Figure 3
Figure 3. Evaluations on the simulation precision. (a) Data collection setup, examples of the real and simu￾lated tactile patterns, and the sim-real SSIM distribution (violin plot) in the tactile signal evaluation. (b) Data collection setup and the shear deformation magnitude trajectories in the frictions simulation evaluation. Real wallclock time (Δt = 0.02s) Dext Hand Didnít converge Taccel (Ours) SAPIEN-IPC Low-res Peg Hig… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Parallel simulation performance analysis across environment scaling. (a) FPS achieved by Taccel (FP64) and SAPIEN-IPC (FP32) on an NVIDIA H100 80G GPU. (b) Task visualizations. 5.3 Precision on Frictions and Shear Deformation We further investigate Taccel’s precision o…
Figure 5
Figure 5. Figure 5: Object classification sim-to-real pipeline. (a): 10 mechanical parts used with diverse geometries. (b) Simulation and real-world data collection setup. (c) Classification network architecture. of FP32 precision leads to convergence issues when solving contact forces in…
Figure 6
Figure 6. Figure 6: Examples of the synthesized grasps and the simulated tactile signals. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Tac-Man manipulation simulation. (a) Demonstration of Tac-Man’s execution-recovery cycles on three articulated objects. (b-c) Three-way comparison among real-world execution, Isaac Sim implementation from Zhao et al. [62], and Taccel. The execution-recovery switch coun…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Tactile Genesis: Exploring Tactile Sensors at Scale for Learning Dexterous Tasks

    cs.RO 2026-06 unverdicted novelty 7.0 of 10

    Whole-hand tactile coverage and per-taxel force/torque dominate sensor type and resolution for learning three dexterous tasks in a new high-throughput tactile simulator.

  2. Data Pyramid for Embodied Manipulation: A Survey

    cs.RO 2026-07 conditional novelty 3.0 of 10

    Embodied training data form a five-layer pyramid—real-robot, UMI, ego/exo, simulation, general V–L—ordered by the trade-off between scale and robot alignment, and model capabilities track how those layers are mixed.

Reference graph

Works this paper leans on

63 extracted references · 58 canonical work pages · cited by 2 Pith papers

  1. [1]

    Tacsl: A library for visuotactile sensor simulation and learning.arXiv preprint arXiv:2408.06506, 2024

    Iretiayo Akinola, Jie Xu, Jan Carius, Dieter Fox, and Yashraj Narang. Tacsl: A library for visuotactile sensor simulation and learning.arXiv preprint arXiv:2408.06506, 2024. 2, 3, 5

    arXiv 2024

  2. [2]

    Tac2pose: Tactile object pose esti- mation from the first touch.International Journal of Robotics Research (IJRR), 42(13):1185– 1209, 2023

    Maria Bauza, Antonia Bronars, and Alberto Rodriguez. Tac2pose: Tactile object pose esti- mation from the first touch.International Journal of Robotics Research (IJRR), 42(13):1185– 1209, 2023. A3

    work page 2023

  3. [3]

    Trends and challenges in robot manipulation.Science, 364(6446):eaat8414, 2019

    Aude Billard and Danica Kragic. Trends and challenges in robot manipulation.Science, 364(6446):eaat8414, 2019. 1, A3

    work page 2019

  4. [4]

    Contactdb: Analyzing and predicting grasp contact via thermal imaging

    Samarth Brahmbhatt, Cusuh Ham, Charles C Kemp, and James Hays. Contactdb: Analyzing and predicting grasp contact via thermal imaging. InProceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 8

    work page 2019

  5. [5]

    Texterity: Tactile ex- trinsic dexterity

    Antonia Bronars, Sangwoon Kim, Parag Patre, and Alberto Rodriguez. Texterity: Tactile ex- trinsic dexterity. InIEEE International Conference on Robotics and Automation (ICRA), 2024. A3

    work page 2024

  6. [6]

    More than a feeling: Learning to grasp and regrasp using vision and touch.IEEE Robotics and Automation Letters (RA-L), 3(4):3300– 3307, 2018

    Roberto Calandra, Andrew Owens, Dinesh Jayaraman, Justin Lin, Wenzhen Yuan, Jitendra Malik, Edward H Adelson, and Sergey Levine. More than a feeling: Learning to grasp and regrasp using vision and touch.IEEE Robotics and Automation Letters (RA-L), 3(4):3300– 3307, 2018. A3

    work page 2018

  7. [7]

    The ycb object and model set: Towards common benchmarks for manipulation re- search

    Berk Calli, Arjun Singh, Aaron Walsman, Siddhartha Srinivasa, Pieter Abbeel, and Aaron M Dollar. The ycb object and model set: Towards common benchmarks for manipulation re- search. InIEEE International Conference on Robotics and Automation (ICRA), 2015. 8

    work page 2015

  8. [8]

    General-purpose sim2real protocol for learning contact-rich manipulation with marker-based visuotactile sen- sors.IEEE Transactions on Robotics (T-RO), 40:1509–1526, 2024

    Weihang Chen, Jing Xu, Fanbo Xiang, Xiaodi Yuan, Hao Su, and Rui Chen. General-purpose sim2real protocol for learning contact-rich manipulation with marker-based visuotactile sen- sors.IEEE Transactions on Robotics (T-RO), 40:1509–1526, 2024. 1, 2, 3, 6, 7, A3, A4 10

    work page 2024

Show all 63 references
  1. [9]

    A unified newton barrier method for multibody dynamics.ACM Transactions on Graphics (TOG), 41(4):1–14,

    Yunuo Chen, Minchen Li, Lei Lan, Hao Su, Yin Yang, and Chenfanfu Jiang. A unified newton barrier method for multibody dynamics.ACM Transactions on Graphics (TOG), 41(4):1–14,

  2. [10]

    Tacchi: A pluggable and low computational cost elastomer deformation simulator for optical tactile sensors.IEEE Robotics and Automation Letters (RA-L), 8(3):1239–1246, 2023

    Zixi Chen, Shixin Zhang, Shan Luo, Fuchun Sun, and Bin Fang. Tacchi: A pluggable and low computational cost elastomer deformation simulator for optical tactile sensors.IEEE Robotics and Automation Letters (RA-L), 8(3):1239–1246, 2023. 3

    2023

  3. [11]

    Improved gelsight tactile sensor for measuring geometry and slip

    Siyuan Dong, Wenzhen Yuan, and Edward H Adelson. Improved gelsight tactile sensor for measuring geometry and slip. InIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017. A3

    2017

  4. [12]

    Tacipc: Intersection-and inversion-free fem-based elastomer simulation for optical tactile sensors.IEEE Robotics and Automation Letters (RA-L), 2024

    Wenxin Du, Wenqiang Xu, Jieji Ren, Zhenjun Yu, and Cewu Lu. Tacipc: Intersection-and inversion-free fem-based elastomer simulation for optical tactile sensors.IEEE Robotics and Automation Letters (RA-L), 2024. 1, 2, 3

    2024

  5. [13]

    A tale of two explanations: Enhancing human trust by explaining robot behavior.Science Robotics, 4(37):eaay4663, 2019

    Mark Edmonds, Feng Gao, Hangxin Liu, Xu Xie, Siyuan Qi, Brandon Rothrock, Yixin Zhu, Ying Nian Wu, Hongjing Lu, and Song-Chun Zhu. A tale of two explanations: Enhancing human trust by explaining robot behavior.Science Robotics, 4(37):eaay4663, 2019. 1

    2019

  6. [14]

    Intersection-free rigid body dynamics.ACM Transactions on Graphics (TOG), 40(4), 2021

    Zachary Ferguson, Minchen Li, Teseo Schneider, Francisca Gil-Ureta, Timothy Langlois, Chenfanfu Jiang, Denis Zorin, Danny M Kaufman, and Daniele Panozzo. Intersection-free rigid body dynamics.ACM Transactions on Graphics (TOG), 40(4), 2021. 6

    2021

  7. [15]

    Planning optimal grasps

    Carlo Ferrari, John F Canny, et al. Planning optimal grasps. InIEEE International Conference on Robotics and Automation (ICRA), 1992. A3

    1992

  8. [16]

    Generation of gelsight tactile images for sim2real learning.IEEE Robotics and Automation Letters (RA-L), 6(2):4177–4184, 2021

    Daniel Fernandes Gomes, Paolo Paoletti, and Shan Luo. Generation of gelsight tactile images for sim2real learning.IEEE Robotics and Automation Letters (RA-L), 6(2):4177–4184, 2021. 6, A3

    2021

  9. [17]

    Gelsight simulation for sim2real learn- ing

    Daniel Fernandes Gomes, Achu Wilson, and Shan Luo. Gelsight simulation for sim2real learn- ing. InICRA ViTac Workshop, 2019. 3

    2019

  10. [18]

    Estimating properties of solid particles inside container using touch sensing

    Xiaofeng Guo, Hung-Jui Huang, and Wenzhen Yuan. Estimating properties of solid particles inside container using touch sensing. InIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023. A3

    2023

  11. [19]

    Deep residual learning for im- age recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for im- age recognition. InProceedings of Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016. 8

    2016

  12. [20]

    Tactile dexterity: Manipulation primitives with tactile feedback

    Francois R Hogan, Jose Ballester, Siyuan Dong, and Alberto Rodriguez. Tactile dexterity: Manipulation primitives with tactile feedback. InIEEE International Conference on Robotics and Automation (ICRA), 2020. 2, A3

    2020

  13. [21]

    A moving least squares material point method with displacement discontinuity and two-way rigid body coupling.ACM Transactions on Graphics (TOG), 37(4):1–14, 2018

    Yuanming Hu, Yu Fang, Ziheng Ge, Ziyin Qu, Yixin Zhu, Andre Pradhana, and Chenfanfu Jiang. A moving least squares material point method with displacement discontinuity and two-way rigid body coupling.ACM Transactions on Graphics (TOG), 37(4):1–14, 2018. 3

    2018

  14. [22]

    Chainqueen: A real-time differentiable physi- cal simulator for soft robotics

    Yuanming Hu, Jiancheng Liu, Andrew Spielberg, Joshua B Tenenbaum, William T Freeman, Jiajun Wu, Daniela Rus, and Wojciech Matusik. Chainqueen: A real-time differentiable physi- cal simulator for soft robotics. InIEEE International Conference on Robotics and Automation (ICRA), 2019. 3

    2019

  15. [23]

    Understanding dynamic tactile sensing for liquid property estimation.Robotics: Science and Systems (RSS), 2022

    Hung-Jui Huang, Xiaofeng Guo, and Wenzhen Yuan. Understanding dynamic tactile sensing for liquid property estimation.Robotics: Science and Systems (RSS), 2022. A3

    2022

  16. [24]

    Hierarchical fall avoidance strategy for small-scale humanoid robots

    Bassam Jalgha, Daniel Asmar, Elie Shammas, and Imad Elhajj. Hierarchical fall avoidance strategy for small-scale humanoid robots. In2012 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2012. 1

    2012

  17. [25]

    Control frame- work for dexterous manipulation using dynamic visual servoing and tactile sensors’ feedback

    Carlos A Jara, Jorge Pomares, Francisco A Candelas, and Fernando Torres. Control frame- work for dexterous manipulation using dynamic visual servoing and tactile sensors’ feedback. Sensors, 14(1):1787–1804, 2014. 1, 9 11

    2014

  18. [26]

    Capturing forceful interaction with deformable objects using a deep learning-powered stretchable tactile array.Nature Com- munications, 15(1):9513, 2024

    Chunpeng Jiang, Wenqiang Xu, Yutong Li, Zhenjun Yu, Longchun Wang, Xiaotong Hu, Zhengyi Xie, Qingkun Liu, Bin Yang, Xiaolin Wang, et al. Capturing forceful interaction with deformable objects using a deep learning-powered stretchable tactile array.Nature Com- munications, 15(1...

    2024

  19. [27]

    Coding and use of tactile signals from the finger- tips in object manipulation tasks.Nature Reviews Neuroscience, 10(5):345–359, 2009

    Roland S Johansson and J Randall Flanagan. Coding and use of tactile signals from the finger- tips in object manipulation tasks.Nature Reviews Neuroscience, 10(5):345–359, 2009. A3

    2009

  20. [28]

    Texterity–tactile ex- trinsic dexterity: Simultaneous tactile estimation and control for extrinsic dexterity

    Sangwoon Kim, Antonia Bronars, Parag Patre, and Alberto Rodriguez. Texterity–tactile ex- trinsic dexterity: Simultaneous tactile estimation and control for extrinsic dexterity. InIEEE International Conference on Robotics and Automation (ICRA), 2024. 2

    2024

  21. [29]

    Affine body dy- namics: fast, stable and intersection-free simulation of stiff materials.ACM Transactions on Graphics (TOG), 41(4):1–14, 2022

    Lei Lan, Danny M Kaufman, Minchen Li, Chenfanfu Jiang, and Yin Yang. Affine body dy- namics: fast, stable and intersection-free simulation of stiff materials.ACM Transactions on Graphics (TOG), 41(4):1–14, 2022. 2, 3, 4, A2

    2022

  22. [30]

    Incremental potential contact: intersection- and inversion-free, large-deformation dynamics.ACM Transactions on Graphics (TOG), 39(4):49, 2020

    Minchen Li, Zachary Ferguson, Teseo Schneider, Timothy R Langlois, Denis Zorin, Daniele Panozzo, Chenfanfu Jiang, and Danny M Kaufman. Incremental potential contact: intersection- and inversion-free, large-deformation dynamics.ACM Transactions on Graphics (TOG), 39(4):49, 2020...

    2020

  23. [31]

    Localization and manipulation of small parts using gelsight tac- tile sensing

    Rui Li, Robert Platt, Wenzhen Yuan, Andreas Ten Pas, Nathan Roscup, Mandayam A Srini- vasan, and Edward Adelson. Localization and manipulation of small parts using gelsight tac- tile sensing. InIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS),

  24. [32]

    Wanlin Li, Zihang Zhao, Leiyao Cui, Weiyi Zhang, Hangxin Liu, Li-An Li, and Yixin Zhu. Minitac: An ultra-compact 8 mm vision-based tactile sensor for enhanced palpation in robot-assisted minimally invasive surgery.IEEE Robotics and Automation Letters (RA-L), 9(12):11170–11177,...

    2024

  25. [33]

    Grasp multiple objects with one hand.IEEE Robotics and Automation Letters (RA-L), 9(5):4027–4034, 2024

    Yuyang Li, Bo Liu, Yiran Geng, Puhao Li, Yaodong Yang, Yixin Zhu, Tengyu Liu, and Siyuan Huang. Grasp multiple objects with one hand.IEEE Robotics and Automation Letters (RA-L), 9(5):4027–4034, 2024. A3

    2024

  26. [34]

    9dtact: A compact vision-based tactile sensor for accurate 3d shape reconstruction and generalizable 6d force estimation.IEEE Robotics and Automation Letters (RA-L), 2023

    Changyi Lin, Han Zhang, Jikai Xu, Lei Wu, and Huazhe Xu. 9dtact: A compact vision-based tactile sensor for accurate 3d shape reconstruction and generalizable 6d force estimation.IEEE Robotics and Automation Letters (RA-L), 2023. 1, A3

    2023

  27. [35]

    Tactile image based contact shape recognition using neural network

    Hongbin Liu, Juan Greco, Xiaojing Song, Joao Bimbo, Lakmal Seneviratne, and Kaspar Al- thoefer. Tactile image based contact shape recognition using neural network. In2012 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI),

  28. [36]

    Tengyu Liu, Zeyu Liu, Ziyuan Jiao, Yixin Zhu, and Song-Chun Zhu. Synthesizing diverse and physically stable grasps with arbitrary hand structures using differentiable force closure estimator.IEEE Robotics and Automation Letters (RA-L), 7(1):470–477, 2021. 8, A3, A4

    2021

  29. [37]

    Pose-and-shear-based tactile servoing.International Journal of Robotics Research (IJRR), 43(7):1024–1055, 2024

    John Lloyd and Nathan F Lepora. Pose-and-shear-based tactile servoing.International Journal of Robotics Research (IJRR), 43(7):1024–1055, 2024. 2, A3

    2024

  30. [38]

    Warp: A high-performance python framework for gpu simulation and graph- ics.https://github.com/nvidia/warp, March 2022

    Miles Macklin. Warp: A high-performance python framework for gpu simulation and graph- ics.https://github.com/nvidia/warp, March 2022. NVIDIA GPU Technology Conference (GTC). 2, 6

    2022

  31. [39]

    Dex-net 2.0: Deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics

    Jeffrey Mahler, Jacky Liang, Sherdil Niyaz, Michael Laskey, Richard Doan, Xinyu Liu, Juan Aparicio Ojea, and Ken Goldberg. Dex-net 2.0: Deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics. InRobotics: Science and Systems (RSS),

  32. [40]

    Isaac gym: High performance gpu-based physics simulation for robot learning.arXiv preprint arXiv:2108.10470, 2021

    Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, Nikita Rudin, Arthur Allshire, Ankur Handa, et al. Isaac gym: High performance gpu-based physics simulation for robot learning.arXiv preprint arXiv:2108.10470, 2021. 6 12

    2021 arXiv

  33. [41]

    Recent progress in advanced tactile sensing technologies for soft grippers.Advanced Functional Materials, 33(41):2306249, 2023

    Juntian Qu, Baijin Mao, Zhenkun Li, Yining Xu, Kunyu Zhou, Xiangyu Cao, Qigao Fan, Minyi Xu, Bin Liang, Houde Liu, et al. Recent progress in advanced tactile sensing technologies for soft grippers.Advanced Functional Materials, 33(41):2306249, 2023. A3

    2023

  34. [42]

    Cable manipulation with a tactile-reactive gripper.International Journal of Robotics Research (IJRR), 40(12-14):1385–1401, 2021

    Yu She, Shaoxiong Wang, Siyuan Dong, Neha Sunil, Alberto Rodriguez, and Edward Adelson. Cable manipulation with a tactile-reactive gripper.International Journal of Robotics Research (IJRR), 40(12-14):1385–1401, 2021. A3

    2021

  35. [43]

    Taxim: An example-based simulation model for gelsight tactile sensors.IEEE Robotics and Automation Letters (RA-L), 7(2):2361–2368, 2022

    Zilin Si and Wenzhen Yuan. Taxim: An example-based simulation model for gelsight tactile sensors.IEEE Robotics and Automation Letters (RA-L), 7(2):2361–2368, 2022. 2, 3

    2022

  36. [44]

    Difftactile: A physics-based differentiable tactile simulator for contact-rich robotic manipula- tion

    Zilin Si, Gu Zhang, Qingwei Ben, Branden Romero, Zhou Xian, Chao Liu, and Chuang Gan. Difftactile: A physics-based differentiable tactile simulator for contact-rich robotic manipula- tion. InProceedings of International Conference on Learning Representations (ICLR), 2024. 1, 2, 3, 5

    2024

  37. [45]

    Robotics and the handbook

    Bruno Siciliano and Oussama Khatib. Robotics and the handbook. InSpringer Handbook of Robotics, pages 1–6. Springer, 2016. A3

    2016

  38. [46]

    A particle method for history-dependent materials.Computer methods in applied mechanics and engineering, 118(1-2):179–196, 1994

    Deborah Sulsky, Zhen Chen, and Howard L Schreyer. A particle method for history-dependent materials.Computer methods in applied mechanics and engineering, 118(1-2):179–196, 1994. 3

    1994

  39. [47]

    Neuralfeels with neural fields: Visuotactile perception for in-hand manipulation.Science Robotics, 9(96):eadl0628,

    Sudharshan Suresh, Haozhi Qi, Tingfan Wu, Taosha Fan, Luis Pineda, Mike Lambeta, Jitendra Malik, Mrinal Kalakrishnan, Roberto Calandra, Michael Kaess, et al. Neuralfeels with neural fields: Visuotactile perception for in-hand manipulation.Science Robotics, 9(96):eadl0628,

  40. [48]

    Tacto: A fast, flexible, and open-source simulator for high-resolution vision-based tactile sensors.IEEE Robotics and Automation Letters (RA-L), 7(2):3930–3937, 2022

    Shaoxiong Wang, Mike Lambeta, Po-Wei Chou, and Roberto Calandra. Tacto: A fast, flexible, and open-source simulator for high-resolution vision-based tactile sensors.IEEE Robotics and Automation Letters (RA-L), 7(2):3930–3937, 2022. 1, 2, 3, 5

    2022

  41. [49]

    Doorbot: Closed-loop task plan- ning and manipulation for door opening in the wild with haptic feedback

    Zhi Wang, Yuchen Mo, Shengmiao Jin, and Wenzhen Yuan. Doorbot: Closed-loop task plan- ning and manipulation for door opening in the wild with haptic feedback. InIEEE International Conference on Robotics and Automation (ICRA), 2025. A3

    2025

  42. [50]

    The tactip family: Soft optical tactile sensors with 3d-printed biomimetic morphologies.Soft Robotics, 5(2):216–227, 2018

    Benjamin Ward-Cherrier, Nicholas Pestell, Luke Cramphorn, Benjamin Winstone, Maria Elena Giannaccini, Jonathan Rossiter, and Nathan F Lepora. The tactip family: Soft optical tactile sensors with 3d-printed biomimetic morphologies.Soft Robotics, 5(2):216–227, 2018. A3

    2018

  43. [51]

    Factors influencing the force control during precision grip.Experimental Brain Research, 53:277–284, 1984

    Goran Westling and Roland S Johansson. Factors influencing the force control during precision grip.Experimental Brain Research, 53:277–284, 1984. 1, A3

    1984

  44. [52]

    Sapien: A simulated part-based interactive envi- ronment

    Fanbo Xiang, Yuzhe Qin, Kaichun Mo, Yikuan Xia, Hao Zhu, Fangchen Liu, Minghua Liu, Hanxiao Jiang, Yifu Yuan, He Wang, et al. Sapien: A simulated part-based interactive envi- ronment. InProceedings of Conference on Computer Vision and Pattern Recognition (CVPR),

  45. [53]

    Efficient tactile simulation with differentiability for robotic manip- ulation

    Jie Xu, Sangwoon Kim, Tao Chen, Alberto Rodriguez Garcia, Pulkit Agrawal, Wojciech Ma- tusik, and Shinjiro Sueda. Efficient tactile simulation with differentiability for robotic manip- ulation. InConference on Robot Learning (CoRL), 2023. 2, 3, 5

    2023

  46. [54]

    Visual-tactile sensing for in-hand object reconstruction

    Wenqiang Xu, Zhenjun Yu, Han Xue, Ruolin Ye, Siqiong Yao, and Cewu Lu. Visual-tactile sensing for in-hand object reconstruction. InProceedings of Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 1, A3

    2023

  47. [55]

    In-hand object classification and pose estimation with sim-to-real tactile transfer for robotic manipulation.IEEE Robotics and Automation Letters (RA-L), 9(1):659–666, 2023

    Sanghoon Yang, Won Dong Kim, Hyunkyu Park, Seojung Min, Hyonyoung Han, and Jung Kim. In-hand object classification and pose estimation with sim-to-real tactile transfer for robotic manipulation.IEEE Robotics and Automation Letters (RA-L), 9(1):659–666, 2023. 8, A3

    2023

  48. [56]

    Exploiting kinematic redundancy for robotic grasping of multiple objects.IEEE Transactions on Robotics (T-RO), 39(3):1982–2002, 2023

    Kunpeng Yao and Aude Billard. Exploiting kinematic redundancy for robotic grasping of multiple objects.IEEE Transactions on Robotics (T-RO), 39(3):1982–2002, 2023. A3 13

    1982

  49. [57]

    Dynamic reconstruction of hand-object interaction with distributed force-aware contact representation.arXiv preprint arXiv:2411.09572, 2024

    Zhenjun Yu, Wenqiang Xu, Pengfei Xie, Yutong Li, and Cewu Lu. Dynamic reconstruction of hand-object interaction with distributed force-aware contact representation.arXiv preprint arXiv:2411.09572, 2024. A3

    2024 arXiv

  50. [58]

    Gelsight: High-resolution robot tactile sensors for estimating geometry and force.Sensors, 17(12):2762, 2017

    Wenzhen Yuan, Siyuan Dong, and Edward H Adelson. Gelsight: High-resolution robot tactile sensors for estimating geometry and force.Sensors, 17(12):2762, 2017. 1, A3

    2017

  51. [59]

    Measurement of shear and slip with a gelsight tactile sensor

    Wenzhen Yuan, Rui Li, Mandayam A Srinivasan, and Edward H Adelson. Measurement of shear and slip with a gelsight tactile sensor. InIEEE International Conference on Robotics and Automation (ICRA), pages 304–311. IEEE, 2015. A3

    2015

  52. [60]

    Tactile sensing in dex- terous robot hands-review.Robotics & Autonomous Systems, 74:195–220, 2015

    Kappassov Zhanat, Juan-Antonio Corrales, and Perdereau Veronique. Tactile sensing in dex- terous robot hands-review.Robotics & Autonomous Systems, 74:195–220, 2015. 1

    2015

  53. [61]

    Embedding high-resolution touch across robotic hands enables adaptive human-like grasping.arXiv preprint arXiv:2412.14482, 2024

    Zihang Zhao, Wanlin Li, Yuyang Li, Tengyu Liu, Boren Li, Meng Wang, Kai Du, Hangxin Liu, Yixin Zhu, Qining Wang, et al. Embedding high-resolution touch across robotic hands enables adaptive human-like grasping.arXiv preprint arXiv:2412.14482, 2024. 1, 9

    2024 arXiv

  54. [62]

    Tac-Man: Tactile-informed prior-free manipulation of articulated objects.IEEE Transactions on Robotics (T-RO), 41:538–557, 2024

    Zihang Zhao, Yuyang Li, Wanlin Li, Zhenghao Qi, Lecheng Ruan, Yixin Zhu, and Kaspar Althoefer. Tac-Man: Tactile-informed prior-free manipulation of articulated objects.IEEE Transactions on Robotics (T-RO), 41:538–557, 2024. 1, 2, 3, 5, 9, A3, A4

    2024

  55. [63]

    Dark, beyond deep: A paradigm shift to cognitive ai with humanlike common sense.Engineering, 6(3):310–345, 2020

    Yixin Zhu, Tao Gao, Lifeng Fan, Siyuan Huang, Mark Edmonds, Hangxin Liu, Feng Gao, Chi Zhang, Siyuan Qi, Ying Nian Wu, et al. Dark, beyond deep: A paradigm shift to cognitive ai with humanlike common sense.Engineering, 6(3):310–345, 2020. A3 14 A Unified ABD-IPC inTaccel InTac...

    2020

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.