Pith. sign in

REVIEW 2 major objections 5 minor 55 references

Grasp Like Humans: Learning Generalizable Multi-Fingered Grasping from Human Proprioceptive Sensorimotor Integration

T0 review · 2 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a glove that records human touch and joint motion, feeding a spatio-temporal graph network, transfers human grasping skills across six robotic hands with a 91.67% success rate.

desk verdict Solid empirical systems paper on glove-based tactile-kinesthetic imitation learning; the cross-robot claim rests on an under-specified manual mapping that needs referee attention. read the letter →

arxiv 2509.08354 v1 pith:SD5VBAXW submitted 2025-09-10 cs.RO cs.AI

Ce Guo , Xieyuanli Chen , Zhiwen Zeng , Zirui Guo , Yihong Li , Haoran Xiao , Dewen Hu , Huimin Lu This is my paper
classification cs.ROcs.AI
keywords multi-fingeredgraspinglearningfromdemonstrationtactilesensingkinestheticfeedbackgraphneuralnetworkforcecontroldeformableobjectsrobothandgeneralization
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

This paper tries to show that human grasping skill can be captured once by a glove that reads joint motion and full-palm contact forces, then executed by different multi-fingered robot hands without collecting new demonstrations or retraining the model. The authors build a pipeline that arranges joint angles, angular velocities, and contact forces as a graph, feeds it to a spatio-temporal graph network (TK-STGN), and converts the network's predictions to robot commands through a force-position hybrid mapping. Across 20 objects that include deformable, slippery, and irregular shapes, they report the highest grasp success rate (91.67%) among the robot methods they compare, with contact forces close to those of human operation. They also report that success stays high under random sensor dropout, random approach angles, random initial postures, and random object placement, and that the same frozen model transfers to six robotic hand configurations. If true, this would mean the skill lives in a human-format tactile-kinesthetic representation rather than in a particular robot body, so adding a new robot hand would require only calibrating a small mapping rather than re-demonstrating the task.

What carries the argument

TK-STGN (Tactile-Kinesthetic Spatio-Temporal Graph Network): a graph neural network that applies weighted multidimensional subgraph convolutions to the hand graph, then bidirectional LSTM layers and multi-head attention over time, to predict future joint angles, contact forces, and a steady-state indicator for each node. The hand graph is the other load-bearing object: 20 joint nodes plus 25 tactile-pad nodes, with fingers linked sequentially and palm nodes fully connected, reflecting anatomical proprioceptive coupling. The graph structure preserves where forces occur on the hand, and the polar-coordinate joint encoding (r, θ, θ̇) folds bone length into the representation so different demons

What would settle it

Take a seventh robotic hand with kinematics not among the six tested, keep TK-STGN frozen, hand-tune the mapping matrices, and run the same 20-object protocol; if success falls below the reported range or requires per-object retuning, the no-retraining generalization claim fails. A faster check: replace the manually tuned mapping with a fixed baseline mapping on the same hand and measure whether the reported success rate collapses.

Watch

Extended reading notes

Core claim

The central claim is that proprioceptive sensorimotor integration in humans—coordinating what the hand feels with how it moves—can be treated as a learnable mapping from tactile and kinesthetic feedback to joint-level states, and that this mapping can be learned from natural human demonstrations rather than from operating a specific robot. The evidence is a glove with 25 tactile pads and 20 tracked joints, worn by humans to demonstrate grasps and then worn by robot hands to give the same input format. The paper's graph representation binds each joint to its nearby tactile pad, uses polar coordinates (bone length, joint angle, angular velocity) to make motion representation independent of dem

Load-bearing premise

The force-position hybrid mapping matrices are set experimentally and manually tuned for each robotic hand, so the claim that a new robot hand needs only simple manual calibration depends on a sparse, hand-chosen linear mapping being good enough to convert human-level joint state predictions into actuator commands for that hand.

Editorial extensions

If this is right

  • A single set of human grasp demonstrations becomes a reusable skill library; adding a new robot hand only requires recalibrating the force-position mapping, not re-collecting data.
  • Robotic hands can maintain stable grasps on deformable and slippery objects without vision or object-specific models, relying on touch and joint feedback alone once positioned.
  • Force management improves: lower terminal forces and smaller deformation of clay specimens than the tested alternatives, which matters for fragile or easily damaged objects.
  • The graph representation makes the system tolerant to missing tactile channels: randomly disabling three pads barely changes success rate.
  • The approach applies to both underactuated and fully actuated hands, as long as the action space of the hand does not exceed that of a human hand.

Reading between the lines

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

  • Editorial extension: the same glove-plus-graph pipeline could plausibly be extended beyond grasping to in-hand manipulation and arm-hand coordination, but the paper's experiments stop at pre-positioned grasping, so the extension is untested.
  • Editorial extension: the paper's own limitation section says the mapping matrices are hand-tuned per hardware; the 'no retraining' promise therefore rests on a small, manually adjusted linear map, and a harder test would be an optimization-based or learned mapping.
  • Editorial extension: because the ablation shows tactile-only performance is weaker than combined tactile-kinesthetic performance, adding higher-density tactile sensing (vibration, shear, temperature) could improve performance on more complex tasks, though the paper currently uses sparse normal-force pads.
  • Editorial extension: a direct falsifier-friendly benchmark would fix the training set at one or two objects and report success on the full 20-object set; the paper's dataset-expansion table suggests performance on unseen objects climbs from 28% to 90% as training data grows, so results are sensitive to dataset scale.
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

2 major / 5 minor

Summary. The paper proposes a glove-mediated framework for transferring human grasping skills to multi-fingered robotic hands. A custom data glove with 25 tactile pads and 20-joint IMU-based motion capture records human demonstrations; these multimodal signals are encoded as graph node states using polar-coordinate kinematic features plus tactile forces. A Tactile-Kinesthetic Spatio-Temporal Graph Network (TK-STGN), combining graph convolutions, Bi-LSTM, and multi-head attention, predicts next-step joint angles, contact forces, and a steady-state flag. Predicted human-level states are converted to robot commands through a force-position hybrid mapping. Experiments on 20 objects across six robot-hand configurations report a 91.67% grasp success rate, force metrics closest to human operation among non-human methods, robustness to randomized perturbations, and ablations supporting the architecture choices.

Significance. If the results hold, the contribution is substantial: demonstration collection from natural human proprioception rather than teleoperation, a sparse tactile+kinesthetic graph representation, and an unusually broad real-robot validation campaign (multiple hands, blocked fingers, deformable/slippery/irregular objects, randomized initial conditions). The paper reports many concrete quantitative results and makes the dataset available. The main novelty—model-level cross-robot transfer without retraining—is plausible, but it rests on an underspecified force-position mapping, and one of the headline force metrics is computed in a non-standard way. These issues must be resolved before the claims can be taken at face value.

major comments (2)
  1. [Sec. III-D and Sec. VI-D(iii), Eq. (21)] The central claim of cross-robot generalization without re-collecting demonstrations or retraining depends on the force-position hybrid mapping. The matrices Γθ, Γh ∈ R^{C×N} are described as 'determined experimentally' and 'fine-tuned' per hand, but the paper gives no calibration procedure, no coefficient values, no tuning cost, and no statement about whether test objects were used during tuning. For underactuated hands (e.g., 6-DoF Inspire vs 15-DoF RY-H1), these matrices encode a control-allocation policy rather than a trivial morphological identity. Table VI therefore conflates model generalization with per-hardware tuning. Please report the calibration protocol, the resulting matrices (or representative entries), tuning effort, sensitivity, and whether any test-object trials influenced the coefficients; if they did, the 'simple manual calibration' claim must be weakened.
  2. [Sec. V-B, Eqs. (22)-(23)] FEM-AT is not a normalized average/standard deviation as described. Eq. (22) accumulates over all O×N×S samples but divides only by S, so the reported value is O·N times the actual per-sample mean (O=20, N=25 → 500×). Eq. (23) sums per-pad standard deviations instead of computing a combined standard deviation. This explains why FEM-AT values in Tables III/IV (≈437 N) are far above any plausible per-pad contact force (≤20 N). The relative ordering across methods may be unaffected, but the claim of being 'closest to human operation' in absolute Newtons is misleading. Please redefine FEM-AT as a true mean and combined standard deviation over all samples, or explicitly relabel it as an unnormalized aggregate score.
minor comments (5)
  1. [Sec. VII (Conclusion)] Typo: 'including deformable, slippery, and deformable ones' should likely read 'deformable, slippery, and irregular ones.'
  2. [Sec. V-B (DCD)] DCD is described only verbally as 'mean normalized Chamfer Distance.' Please give the explicit formula and the normalization factor (e.g., object scale), since Tables III/IV report values in units of 10^-3.
  3. [Sec. V-E (Table V)] Success rates are reported without confidence intervals; with 15 attempts per object, binomial uncertainty is substantial (e.g., 14/15 vs 15/15 are statistically indistinguishable). Please report intervals or at least the raw numerators/denominators for each condition.
  4. [Sec. VI-A (Table VII)] Seen-object success rates are non-monotonic as training data are expanded (100% → 84.44% → 71.67% → 85.33%). The text claims a 'steady increase' for unseen objects and overall performance, which is true, but the drop on seen objects is unexplained. Please clarify whether all prior demonstrations are retained, whether training is re-run from scratch, and whether results vary with random seeds.
  5. [Sec. I / Sec. VI-D(iii)] The introduction promises 'only simple manual calibration,' while Sec. VI-D(iii) states that the force-position mapping 'requires manual parameter tuning specific to the hardware.' Please align these statements and quantify the tuning effort in the revision.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the model is trained by supervised imitation against human demonstration states and evaluated on external grasp outcomes; per-hand mapping tuning is calibration, not a definitional shortcut.

full rationale

The derivation chain is: glove-collected human demonstrations → graph-structured kinesthetic/tactile states → TK-STGN trained with weighted MSE (Eqs. 18–20) to predict next joint angles, contact forces, and steady-state indicators → force-position hybrid mapping (Eq. 21) converting predictions into robot commands. None of the predicted targets (θ*, h*, b*) are used to define the predictor itself, and success is judged by physical lifting outcomes, robot-side force measurements, and mesh deformation (DCD), which are not model outputs by construction. The force-position matrices Γθ and Γh are 'determined experimentally' and require per-hardware manual tuning (Sec. III-D, Sec. VI-D(iii)); this is an acknowledged calibration step, not a fitted quantity renamed as a prediction. The paper does not claim these coefficients are learned from demonstrations, and the generalization claim is specifically about the trained state predictor, with the mapping as a separate hardware interface. There are no self-citations bearing on the central argument, no imported uniqueness theorems, and no ansatz smuggled via citation. The under-specification of the tuning protocol is a reproducibility/correctness limitation, not circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The method relies on several domain assumptions about hand structure and tactile sufficiency; the main fitted quantities are the mapping coefficients and network hyperparameters.

free parameters (2)
  • Force-position mapping coefficients Γθ and Γh = Not disclosed; determined experimentally per robotic hand
    Sparse matrices in Eq. (21) that convert predicted joint angle and force errors into actuator commands. The paper states they are 'determined experimentally' and vary with the robotic hand (Sec. III-D).
  • Network hyperparameters (L=2, K=3, M=4, lr=1e-4, batch 32) = As listed in Table II
    Chosen by hand, not fitted to data. They affect model capacity but are not the central fitted quantities.
assumptions (4)
  • domain assumption The topological structure of the hand graph is fixed during grasping (S_n = D^{-1/2} A D^{-1/2} is constant)
    Assumed in Sec. III-C, Eq. (5): 'we assume that the topological relationships of nodes remain unchanged during the grasping process.' This simplification may fail if contact dynamics alter effective connectivity.
  • domain assumption Finger nodes are connected only sequentially to neighbors, while palm nodes are fully connected
    Stated in Sec. III-B based on relative independence of finger proprioception and dense palm muscle coupling. An incorrect coupling structure would mislead graph convolution.
  • domain assumption Finger abduction/adduction (except thumb) can be neglected; polar-coordinate representation with flexion/extension and thumb abduction/adduction suffices for grasping
    Sec. III-B and Limitation ii in Sec. VI-D. If tasks require fine abduction/adduction, the reduced representation would be insufficient.
  • domain assumption Normal force magnitude/distribution from 25 sparse pads is a reasonable tactile observation for grasping
    Sec. III-A states they treat the acquired magnitude/distribution as a downscaled but reasonable observation. Sparse sampling and lack of shear/temperature sensing are acknowledged as limitations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Grasp Like Humans: Learning Generalizable Multi-Fingered Grasping from Human Proprioceptive Sensorimotor Integration." pith.science (2026). https://pith.science/paper/SD5VBAXW

@misc{pith2026250908354,
  author       = {Pith},
  title        = {Pith review of: Grasp Like Humans: Learning Generalizable Multi-Fingered Grasping from Human Proprioceptive Sensorimotor Integration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SD5VBAXW}},
  note         = {Machine review of arXiv:2509.08354}
}
read the original abstract

Tactile and kinesthetic perceptions are crucial for human dexterous manipulation, enabling reliable grasping of objects via proprioceptive sensorimotor integration. For robotic hands, even though acquiring such tactile and kinesthetic feedback is feasible, establishing a direct mapping from this sensory feedback to motor actions remains challenging. In this paper, we propose a novel glove-mediated tactile-kinematic perception-prediction framework for grasp skill transfer from human intuitive and natural operation to robotic execution based on imitation learning, and its effectiveness is validated through generalized grasping tasks, including those involving deformable objects. Firstly, we integrate a data glove to capture tactile and kinesthetic data at the joint level. The glove is adaptable for both human and robotic hands, allowing data collection from natural human hand demonstrations across different scenarios. It ensures consistency in the raw data format, enabling evaluation of grasping for both human and robotic hands. Secondly, we establish a unified representation of multi-modal inputs based on graph structures with polar coordinates. We explicitly integrate the morphological differences into the designed representation, enhancing the compatibility across different demonstrators and robotic hands. Furthermore, we introduce the Tactile-Kinesthetic Spatio-Temporal Graph Networks (TK-STGN), which leverage multidimensional subgraph convolutions and attention-based LSTM layers to extract spatio-temporal features from graph inputs to predict node-based states for each hand joint. These predictions are then mapped to final commands through a force-position hybrid mapping.

Figures

Figures reproduced from arXiv: 2509.08354 by the authors.

Figure 1
Figure 1. Pipeline of our framework. (a) In the demonstration procedure, the human operator wears the data glove to complete the grasping operation, and the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The data glove. (a) The palm side of the glove. It is covered by the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Unified representation based on the graph and polar coordinates. (a) [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (15 more)
Figure 5
Figure 5. Figure 5: The structure of TK-STGN and imitation learning procedure. The [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Seen and unseen objects used in the training and testing process. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 6
Figure 6. Figure 6: A demonstration of grasping by a human hand. (a) Human grasping [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 8
Figure 8. Figure 8: The grasp testing system. The left and right figures show the initial [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Deformation assessment specimens and reconstructed meshes. [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: The successful grasping processes using our TK-STGN. We present the grasping processes for three objects from two perspectives. Besides, [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Final states of ideal, suboptimal and unsuccessful grasps using different grasping approaches. The annotations below the subfigures identify the [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Representative mesh reconstruction comparisons before and after grasping across different approaches (one specimen shown per approach from six [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Distribution of critical contact forces among 50 successful grasps for [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Distribution of critical contact forces among 50 successful grasps for [PITH_FULL_IMAGE:figures/full_fig_p014_14.png]
Figure 15
Figure 15. Figure 15: The range of random approach angles. For side grasps, the [PITH_FULL_IMAGE:figures/full_fig_p015_15.png]
Figure 16
Figure 16. Figure 16: Region for random placement of the grasped object (top-down view). [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]
Figure 17
Figure 17. Figure 17: Successful grasps under random setups. The labeled locations represent tactile sampling pads that are randomly masked. [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]
Figure 18
Figure 18. Figure 18: Successful grasps with different robotic hands. Key dimensions and the number of actuated DoF of each hand are labeled. [PITH_FULL_IMAGE:figures/full_fig_p016_18.png]
Figure 19
Figure 19. Figure 19: Grasp-based task execution processes and the corresponding contact force distribution. [PITH_FULL_IMAGE:figures/full_fig_p017_19.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

55 extracted references · 53 canonical work pages

  1. [1]

    Multi- fingered grasping based on multimodal reinforcement learning,

    H. Liang, L. Cong, N. Hendrich, S. Li, F. Sun, and J. Zhang, “Multi- fingered grasping based on multimodal reinforcement learning,”IEEE Robot. Autom. Lett., vol. 7, no. 2, pp. 1174–1181, 2021

    work page 2021

  2. [2]

    Grasping living objects with adversarial behaviors using inverse reinforcement learning,

    Z. Hu, Y . Zheng, and J. Pan, “Grasping living objects with adversarial behaviors using inverse reinforcement learning,”IEEE Trans. Robot., vol. 39, no. 2, pp. 1151–1163, 2023

    work page 2023

  3. [3]

    Learning dexterous in-hand manipulation,

    O. M. Andrychowicz, B. Baker, M. Chociej, R. Jozefowicz, B. McGrew, J. Pachocki, A. Petron, M. Plappert, G. Powell, A. Rayet al., “Learning dexterous in-hand manipulation,”Int. J. Robot. Res., vol. 39, no. 1, pp. 3–20, 2020

    work page 2020

  4. [4]

    Simplified configuration design of anthropomorphic hand imitating specific human hand grasps,

    X. Tian, Q. Zhan, Y . Zhang, J. Zou, L. Jiang, and Q. Xu, “Simplified configuration design of anthropomorphic hand imitating specific human hand grasps,”IEEE Robot. Autom. Lett., vol. 8, no. 1, pp. 152–159, 2022

    work page 2022

  5. [5]

    Tacgnn: Learning tactile-based in-hand manipulation with a blind robot using hierarchical graph neural network,

    L. Yang, B. Huang, Q. Li, Y .-Y . Tsai, W. W. Lee, C. Song, and J. Pan, “Tacgnn: Learning tactile-based in-hand manipulation with a blind robot using hierarchical graph neural network,”IEEE Robot. Autom. Lett., vol. 8, no. 6, pp. 3605–3612, 2023

    work page 2023

  6. [6]

    An internal model for sensorimotor integration,

    D. M. Wolpert, Z. Ghahramani, and M. I. Jordan, “An internal model for sensorimotor integration,”Science, vol. 269, no. 5232, pp. 1880–1882, 1995

    work page 1995

  7. [7]

    The neural basis of somatosensory temporal discrimination threshold as a paradigm for time processing in the sub-second range: an updated review,

    C. M. Ord ´as and F. Alonso-Frech, “The neural basis of somatosensory temporal discrimination threshold as a paradigm for time processing in the sub-second range: an updated review,”Neurosci. Biobehav. Rev., p. 105486, 2023

    work page 2023

  8. [8]

    Unidexgrasp++: Improving dexterous grasping policy learning via geometry-aware curriculum and iterative generalist-specialist learning,

    W. Wan, H. Geng, Y . Liu, Z. Shan, Y . Yang, L. Yi, and H. Wang, “Unidexgrasp++: Improving dexterous grasping policy learning via geometry-aware curriculum and iterative generalist-specialist learning,” inProc. IEEE Int. Conf. Comput. Vision, 2023, pp. 3891–3902

    work page 2023

Show all 55 references
  1. [9]

    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,” inProc. Robot. Sci. Syst., 2023

    2023

  2. [10]

    Multi-fingered in-hand manipulation with various object properties using graph convolutional networks and distributed tactile sensors,

    S. Funabashi, T. Isobe, F. Hongyi, A. Hiramoto, A. Schmitz, S. Sugano, and T. Ogata, “Multi-fingered in-hand manipulation with various object properties using graph convolutional networks and distributed tactile sensors,”IEEE Robot. Autom. Lett., vol. 7, no. 2, pp. 2102–2109, 2022

    2022

  3. [11]

    Grasp it like a pro 2.0: A data-driven approach exploiting basic shape decomposition and human data for grasping unknown objects,

    A. Palleschi, F. Angelini, C. Gabellieri, L. Pallottino, A. Bicchi, M. Garabiniet al., “Grasp it like a pro 2.0: A data-driven approach exploiting basic shape decomposition and human data for grasping unknown objects,”IEEE Trans. Robot., vol. 39, no. 5, pp. 4016–4036, 2023

    2023

  4. [12]

    Recent advances in robot learning from demonstration,

    H. Ravichandar, A. S. Polydoros, S. Chernova, and A. Billard, “Recent advances in robot learning from demonstration,”Annu. Rev. Contr. Robot. Autonom. Syst., vol. 3, pp. 297–330, 2020

    2020

  5. [13]

    Mimicplay: Long-horizon imitation learning by watching human play,

    C. Wang, L. Fan, J. Sun, R. Zhang, L. Fei-Fei, D. Xu, Y . Zhu, and A. Anandkumar, “Mimicplay: Long-horizon imitation learning by watching human play,” inProc. Mach. Learn. Res., 2023

    2023

  6. [14]

    Learning robot manipulation skills from human demonstration videos using two-stream 2-d/3-d resid- ual networks with self-attention,

    X. Xu, K. Qian, X. Jing, and W. Song, “Learning robot manipulation skills from human demonstration videos using two-stream 2-d/3-d resid- ual networks with self-attention,”IEEE Trans. Cogn. Dev. Syst., vol. 15, no. 3, pp. 1000–1011, 2022

    2022

  7. [15]

    Causal confusion in imitation learning,

    P. de Haan, D. Jayaraman, and S. Levine, “Causal confusion in imitation learning,” inAdv. Neural Inf. Proces. Syst., vol. 32, 2019

    2019

  8. [16]

    Grasp it like a pro: Grasp of unknown objects with robotic hands based on skilled human expertise,

    C. Gabellieri, F. Angelini, V . Arapi, A. Palleschi, M. G. Catalano, G. Grioli, L. Pallottino, A. Bicchi, M. Bianchi, and M. Garabini, “Grasp it like a pro: Grasp of unknown objects with robotic hands based on skilled human expertise,”IEEE Robot. Autom. Lett., vol. 5, no. 2, p...

    2020

  9. [17]

    A wearable robotic hand for hand-over-hand imitation learning,

    D. Wei and H. Xu, “A wearable robotic hand for hand-over-hand imitation learning,” inProc. IEEE Int. Conf. Rob. Autom.IEEE, 2024, pp. 18 113–18 119

    2024

  10. [18]

    Learning the signatures of the human grasp using a scalable tactile glove,

    S. Sundaram, P. Kellnhofer, Y . Li, J.-Y . Zhu, A. Torralba, and W. Ma- tusik, “Learning the signatures of the human grasp using a scalable tactile glove,”Nature, vol. 569, no. 7758, pp. 698–702, 2019

    2019

  11. [19]

    Stable in-grasp manipulation with a low-cost robot hand by using 3-axis tactile sensors with a cnn,

    S. Funabashi, T. Isobe, S. Ogasa, T. Ogata, A. Schmitz, T. P. Tomo, and S. Sugano, “Stable in-grasp manipulation with a low-cost robot hand by using 3-axis tactile sensors with a cnn,” inProc. IEEE Int. Conf. Intell. Rob. Syst., 2020, pp. 9166–9173

    2020

  12. [20]

    Hands for dexterous manipulation and robust grasping: A difficult road toward simplicity,

    A. Bicchi, “Hands for dexterous manipulation and robust grasping: A difficult road toward simplicity,”IEEE Trans. Robot. Autom., vol. 16, no. 6, pp. 652–662, 2000

    2000

  13. [21]

    Relaxed-rigidity constraints: kine- matic trajectory optimization and collision avoidance for in-grasp ma- nipulation,

    B. Sundaralingam and T. Hermans, “Relaxed-rigidity constraints: kine- matic trajectory optimization and collision avoidance for in-grasp ma- nipulation,”Auton. Robot., vol. 43, pp. 469–483, 2019

    2019

  14. [22]

    Multi-fingered tactile servoing for grasping adjustment under partial observation,

    H. Liu, B. Huang, Q. Li, Y . Zheng, Y . Ling, W. Lee, Y . Liu, Y .-Y . Tsai, and C. Yang, “Multi-fingered tactile servoing for grasping adjustment under partial observation,” inProc. IEEE Int. Conf. Intell. Rob. Syst., 2022, pp. 7781–7788

    2022

  15. [23]

    Living object grasping using two-stage graph reinforcement learning,

    Z. Hu, Y . Zheng, and J. Pan, “Living object grasping using two-stage graph reinforcement learning,”IEEE Robot. Autom. Lett., vol. 6, no. 2, pp. 1950–1957, 2021

    1950

  16. [24]

    Sim-to-real transfer for robotic manipulation with tactile sensory,

    Z. Ding, Y .-Y . Tsai, W. W. Lee, and B. Huang, “Sim-to-real transfer for robotic manipulation with tactile sensory,” inProc. IEEE Int. Conf. Intell. Rob. Syst., 2021, pp. 6778–6785

    2021

  17. [25]

    Universal manipulation interface in-the-wild robot teach- ing 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 teach- ing without in-the-wild robots,” inProc. Robot. Sci. Syst., 2024

    2024

  18. [26]

    Unidexgrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal-conditioned policy,

    Y . Xu, W. Wan, J. Zhang, H. Liu, Z. Shan, H. Shen, R. Wang, H. Geng, Y . Weng, J. Chenet al., “Unidexgrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal-conditioned policy,” inProc. IEEE Comput. Soc Conf. Comput. Vision Pattern Recogn...

    2023

  19. [27]

    Learning human-like functional grasping for multi-finger hands from few demonstrations,

    W. Wei, P. Wang, S. Wang, Y . Luo, W. Li, D. Li, Y . Huang, and H. Duan, “Learning human-like functional grasping for multi-finger hands from few demonstrations,”IEEE Trans. Robot., vol. 40, pp. 3897–3916, 2024

    2024

  20. [28]

    Fingerslam: Closed-loop unknown object localization and reconstruction from visuo-tactile feed- back,

    J. Zhao, M. Bauza, and E. H. Adelson, “Fingerslam: Closed-loop unknown object localization and reconstruction from visuo-tactile feed- back,” inProc. IEEE Int. Conf. Rob. Autom., 2023, pp. 8033–8039

    2023

  21. [29]

    Detection of slip from vision and touch,

    G. Yan, A. Schmitz, T. P. Tomo, S. Somlor, S. Funabashi, and S. Sugano, “Detection of slip from vision and touch,” inProc. IEEE Int. Conf. Rob. Autom., 2022, pp. 3537–3543

    2022

  22. [30]

    Swingbot: Learning physical features from in-hand tactile exploration for dynamic swing-up manipulation,

    C. Wang, S. Wang, B. Romero, F. Veiga, and E. Adelson, “Swingbot: Learning physical features from in-hand tactile exploration for dynamic swing-up manipulation,” inProc. IEEE Int. Conf. Intell. Rob. Syst., 2020, pp. 5633–5640

    2020

  23. [31]

    Encoding of tactile information in hand via skin- integrated wireless haptic interface,

    K. Yao, J. Zhou, Q. Huang, M. Wu, C. K. Yiu, J. Li, X. Huang, D. Li, J. Su, S. Houet al., “Encoding of tactile information in hand via skin- integrated wireless haptic interface,”Nat. Mach. Intell., vol. 4, no. 10, pp. 893–903, 2022

    2022

  24. [32]

    Skin-inspired piezoelectric tactile sensor array with crosstalk-free row+ column elec- trodes 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 elec- trodes for spatiotemporally distinguishing diverse stimuli,”Adv. Sci., vol. 8, no. 3, p. 2002817, 2021

    2021

  25. [33]

    Towards integrated tactile sensorimotor control in anthropomorphic soft robotic hands,

    N. F. Lepora, C. Ford, A. Stinchcombe, A. Brown, J. Lloyd, M. G. Catalano, M. Bianchi, and B. Ward-Cherrier, “Towards integrated tactile sensorimotor control in anthropomorphic soft robotic hands,” inProc. IEEE Int. Conf. Rob. Autom., 2021, pp. 1622–1628

    2021

  26. [34]

    Learning to grasp without seeing,

    A. Murali, Y . Li, D. Gandhi, and A. Gupta, “Learning to grasp without seeing,” inSpringer. Proc. Adv. Robot., 2018, pp. 375–386

    2018

  27. [35]

    Learning to grasp under uncertainty using pomdps,

    N. P. Garg, D. Hsu, and W. S. Lee, “Learning to grasp under uncertainty using pomdps,” inProc. IEEE Int. Conf. Rob. Autom., 2019, pp. 2751– 2757

    2019

  28. [36]

    Tactile-based grasping stability prediction based on human grasp demonstration for robot manipulation,

    Z. Zhao, W. He, and Z. Lu, “Tactile-based grasping stability prediction based on human grasp demonstration for robot manipulation,”IEEE Robot. Autom. Lett., vol. 9, no. 3, pp. 2646–2653, 2024

    2024

  29. [37]

    Stretchable glove for accurate and robust hand pose reconstruction based on comprehensive motion data,

    M. Park, T. Park, S. Park, S. J. Yoon, S. H. Koo, and Y .-L. Park, “Stretchable glove for accurate and robust hand pose reconstruction based on comprehensive motion data,”Nat. Commun., vol. 15, no. 1, p. 5821, 2024

    2024

  30. [38]

    Tactile and kinesthetic communication glove with fusion of triboelectric sensing and pneumatic actuation,

    R. Wang, L. Jiang, J. Li, Z. Dai, M. Liu, P. Lv, X. Li, and M. Zhu, “Tactile and kinesthetic communication glove with fusion of triboelectric sensing and pneumatic actuation,”Nano Energy, vol. 131, p. 110273, 2024

    2024

  31. [39]

    Fluid reality: High-resolution, untethered haptic gloves using electroosmotic pump arrays,

    V . Shen, T. Rae-Grant, J. Mullenbach, C. Harrison, and C. Shultz, “Fluid reality: High-resolution, untethered haptic gloves using electroosmotic pump arrays,” inProc. Annual ACM Symp. User Interface Softw. Technol., 2023, pp. 1–20

    2023

  32. [40]

    A robotic grasping state perception framework with multi-phase tactile information and ensemble learning,

    G. Yan, A. Schmitz, S. Funabashi, S. Somlor, T. P. Tomo, and S. Sugano, “A robotic grasping state perception framework with multi-phase tactile information and ensemble learning,”IEEE Robot. Autom. Lett., vol. 7, no. 3, pp. 6822–6829, 2022. 20 IEEE TRANSACTIONS ON ROBOTICS

    2022

  33. [41]

    Geometric transformation: Tactile data augmentation for robotic learning,

    G. Yan, J. Yuyeol, S. Funabashi, T. P. Tomo, S. Somlor, A. Schmitz, and S. Sugano, “Geometric transformation: Tactile data augmentation for robotic learning,” inProc. IEEE Int. Conf. Dev. Learn., 2023, pp. 346–353

    2023

  34. [42]

    Tactilegcn: A graph convolutional network for predicting grasp stability with tactile sensors,

    A. Garcia-Garcia, B. S. Zapata-Impata, S. Orts-Escolano, P. Gil, and J. Garcia-Rodriguez, “Tactilegcn: A graph convolutional network for predicting grasp stability with tactile sensors,” inProc. Int. Jt. Conf. Neural Networks, 2019, pp. 1–8

    2019

  35. [43]

    Learning object-action relations from bimanual human demonstration using graph networks,

    C. R. Dreher, M. W ¨achter, and T. Asfour, “Learning object-action relations from bimanual human demonstration using graph networks,” IEEE Robot. Autom. Lett., vol. 5, no. 1, pp. 187–194, 2019

    2019

  36. [44]

    Tactile features of human finger contact motor primitives,

    Q. Tian, J. Liu, K. Liu, and S. Guo, “Tactile features of human finger contact motor primitives,”IEEE Trans. Haptics, vol. 16, no. 4, pp. 848– 860, 2023

    2023

  37. [45]

    Easycalib: Simple and low-cost in-situ calibration for force reconstruction with vision-based tactile sensors,

    M. Li, L. Zhang, Y . H. Zhou, T. Li, and Y . Jiang, “Easycalib: Simple and low-cost in-situ calibration for force reconstruction with vision-based tactile sensors,”IEEE Robot. Autom. Lett., vol. 9, no. 9, pp. 7803–7810, 2024

    2024

  38. [46]

    Gray’s anatomy: the anatomical basis of clinical practice,

    S. Standring, H. Ellis, J. Healy, D. Johnson, A. Williams, P. Collins, and C. Wigley, “Gray’s anatomy: the anatomical basis of clinical practice,” Am. J. Neuroradiol., vol. 26, no. 10, p. 2703, 2005

    2005

  39. [47]

    Benchmarking in manipulation research: Using the yale-cmu- berkeley object and model set,

    B. Calli, A. Walsman, A. Singh, S. Srinivasa, P. Abbeel, and A. M. Dollar, “Benchmarking in manipulation research: Using the yale-cmu- berkeley object and model set,”IEEE Robot. Autom. Mag., vol. 22, no. 3, pp. 36–52, 2015

    2015

  40. [48]

    Instant neural graphics primitives with a multiresolution hash encoding,

    T. M ¨uller, A. Evans, C. Schied, and A. Keller, “Instant neural graphics primitives with a multiresolution hash encoding,”ACM Trans. Graph., vol. 41, no. 4, pp. 1–15, 2022

    2022

  41. [49]

    Segment and track anything,

    Y . Cheng, L. Li, Y . Xu, X. Li, Z. Yang, W. Wang, and Y . Yang, “Segment and track anything,”arXiv preprint arXiv:2305.06558, 2023

    2023 arXiv

  42. [50]

    Colmap: A memory-efficient occupancy grid mapping frame- work,

    A. Fisher, R. Cannizzaro, M. Cochrane, C. Nagahawatte, and J. L. Palmer, “Colmap: A memory-efficient occupancy grid mapping frame- work,”Robot. Auton. Syst., vol. 142, p. 103755, 2021

    2021

  43. [51]

    See, hear, and feel: Smart sensory fusion for robotic manipulation,

    H. Li, Y . Zhang, J. Zhu, S. Wang, M. A. Lee, H. Xu, E. Adelson, L. Fei- Fei, R. Gao, and J. Wu, “See, hear, and feel: Smart sensory fusion for robotic manipulation,” inProc. Mach. Learn. Res., 2022

    2022

  44. [52]

    A dexterous hand-arm teleoperation system based on hand pose estimation and active vision,

    S. Li, N. Hendrich, H. Liang, P. Ruppel, C. Zhang, and J. Zhang, “A dexterous hand-arm teleoperation system based on hand pose estimation and active vision,”IEEE T. Cybern., vol. 54, no. 3, pp. 1417–1428, 2022

    2022

  45. [53]

    Survey on mapping human hand motion to robotic hands for teleoperation,

    R. Li, H. Wang, and Z. Liu, “Survey on mapping human hand motion to robotic hands for teleoperation,”IEEE Trans. Circuits Syst. Video Technol., vol. 32, no. 5, pp. 2647–2665, 2021

    2021

  46. [54]

    Telefmg: A wearable force-myography device for natural teleoperation of multi-finger robotic hands,

    A. Mizrahi and A. Sintov, “Telefmg: A wearable force-myography device for natural teleoperation of multi-finger robotic hands,”IEEE Robot. Autom. Lett., vol. 9, no. 3, pp. 2933–2940, 2024

    2024

  47. [55]

    Gendexgrasp: Generalizable dexterous grasping,

    P. Li, T. Liu, Y . Li, Y . Geng, Y . Zhu, Y . Yang, and S. Huang, “Gendexgrasp: Generalizable dexterous grasping,” inProc. IEEE Int. Conf. Rob. Autom., 2023, pp. 8068–8074. Ce Guoreceived the B.E. degree in automation and the M.E. degree in control science and engineering from...

    2023

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.