Pith. sign in

REVIEW 4 major objections 5 minor 31 references

Pose-Aware Modeling to Mitigate Pose-Related Artifacts in Tactile Gloves

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Adding hand pose information to a tactile glove's force estimator cuts pose-induced artifacts and lowers minimum detectable force by up to 18% across three glove designs.

desk verdict Deserves a serious referee, but the headline MDF gains are not fully attributable to pose information until capacity-matched controls and a pose-shuffle ablation are added. read the letter →

arxiv 2607.22964 v1 pith:VCR2YS2O submitted 2026-07-25 cs.RO cs.HC

Tianhong Catherine Yu , Ziyi Kou , Mia Huang , Taylor Niehues , Yiyue Luo , Li Guan , Dingtian Zhang This is my paper
classification cs.ROcs.HC
keywords tactileglovespose-relatedartifactsforceestimationresiduallearningminimumdetectablesoftsensorshandposecontactdetection
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

Tactile gloves use soft, flexible sensors that respond to contact force but also to the wearer's hand pose, so a bent finger can masquerade as a light touch. This paper argues that these pose-related artifacts can be corrected algorithmically by feeding synchronized hand joint angles into a force-estimation model as a learned residual correction, with no hardware changes. Across three different glove designs and fifteen users, the pose-aware model lowers the minimum detectable force by 10.4%, 12.2%, and 18.3% and improves touch detection, MAE, and R-squared consistently. The gains are concentrated in the low-force range that matters for dexterous manipulation, and they persist on pseudo-unseen free movements, suggesting the correction is not just memorizing the training poses.

What carries the argument

The load-bearing identity is F_hat_t = F_base_t(T_t) - F_residual_t(T_t, P_t): a tactile-only force estimate minus a pose-conditioned residual. The residual branch fuses tactile and pose features via concatenation plus element-wise product so the model captures context-dependent pose-tactile interactions. Hand pose is encoded with a sinusoidal encoding and a Bi-LSTM over the four index-finger joint angles. An auxiliary decoder reconstructs zero-load tactile signals from pose, tying the pose features to the artifact structure. This formulation is glove-agnostic: any tactile-to-force pipeline that already produces a tactile feature and a force estimate can be augmented with the residual branch

What would settle it

Measure PRA magnitude under identical joint angles with two different glove fits (e.g., tight and loose) for the same user; if the pose-to-artifact mapping changes drastically with fit, then joint angles alone are insufficient and the MDF reduction would not transfer across glove fit or hand shape. A more direct falsifier: attach a strain gauge near the fingertip sensor and check whether joint angles predict the strain-induced signal; if they explain little variance, the residual branch has no reliable signal to exploit.

Watch

Extended reading notes

Core claim

The central discovery is that pose-related artifacts in tactile glove signals are systematic and predictable from the hand's joint angles, even though the strain field is never measured. The method instantiates this as a residual identity: corrected force = tactile-only force estimate minus a signed residual predicted jointly from tactile and pose features. An auxiliary loss reconstructs zero-load tactile signals from pose, forcing the residual branch to capture the artifact rather than a spurious pose-to-force mapping. Across three gloves and 15 users, the pose-aware model reduces MDF by 10.4%, 12.2%, and 18.3% and improves all reported metrics, including on pseudo-unseen movements.

Load-bearing premise

The central premise is that the four measured index-finger joint angles capture the mechanical strain the sensors experience; the paper itself notes that exact strain also depends on sensor mechanics, hand shape, and glove fit, and can only be measured with sensor modifications.

Editorial extensions

If this is right

  • On all three gloves, the pose-aware model lowers the minimum detectable force (10.4%, 12.2%, 18.3%) and improves touch detection error by 15-24%, so gloves can reliably register lighter contacts.
  • The method requires no glove modification; any existing glove can be upgraded by adding a pose stream (mocap, egocentric vision, or wearable sensors) and retraining the residual branch.
  • Pose-aware correction suppresses false positives during in-air movements, delayed touch onset, and lingering force after unload - the three failure modes that degrade contact timing in policy learning.
  • Gains persist on pseudo-unseen free pose-force combinations, indicating the residual generalizes beyond prescribed poses.
  • The residual formulation avoids the degradation seen with naive pose-tactile concatenation, implying that explicit artifact modeling, not just extra input, is what drives the improvement.

Reading between the lines

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

  • If kinematic joint angles are a sufficient proxy for the strain field, the same residual-correction recipe could extend to other flexible wearable sensors (e.g., e-textiles at wrist, knee, or palm) where pose-induced deformation confounds the readout - though the paper validates only the index fingertip.
  • The paper's own admission that strain depends on sensor mechanics, hand shape, and glove fit suggests a hybrid design: kinematic residual correction plus a sparse set of on-sensor strain gauges could provide the missing information for zero-shot transfer across users and gloves.
  • A natural testable extension is to feed pose streams of varying quality (e.g., egocentric hand tracking with jitter) into the same framework and measure how MDF gains degrade; the paper notes this as future work, but it would establish the method's practical envelope.
  • Because the auxiliary zero-load loss is supervised only during in-air movements, the model should be probed with the same pose sequence at different loads to verify that the residual does not accidentally absorb force-dependent effects; the paper's signed-residual form assumes additivity of pose and force effects.
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 / 5 minor

Summary. The paper defines pose-related artifacts (PRAs) in soft tactile gloves as sensor signal changes caused by hand pose rather than contact force, and proposes a pose-aware force estimation framework that augments a tactile-only branch with a residual correction branch fed by hand pose. The method is validated on three glove designs with 15 users, reporting consistent improvements in touch detection balanced accuracy, F1, MDF, MAE, and R², with MDF reductions of 10.4%, 12.2%, and 18.3%. The authors also provide qualitative examples showing reduced false positives during in-air movements and reduced latency at touch onset/offset.

Significance. If the causal claim is sustained, the paper offers a practical, hardware-free mitigation for a known limitation of soft tactile gloves, with potential value for teleoperation, learning from demonstration, and tactile data collection. The strengths are the multi-glove, multi-user dataset, the explicit PRA characterization (Observations 1–3), the real-time demonstration, and the consistency of improvements across all reported metrics and gloves. The authors also honestly disclose limitations such as reliance on marker-based motion capture and a single flat end effector. The main gap is that the reported experiments do not yet isolate pose information as the cause of the improvements, which is load-bearing for the paper's central claim.

major comments (4)
  1. [§VI.A, Table I, Fig. 5] The central attribution claim is not yet supported because the tactile-only baseline is not capacity-matched. The pose-aware model adds a pose encoder (Bi-LSTM), fusion layer (concatenation plus elementwise product), and an auxiliary decoder supervised by the L_PRA_tactile loss in Eq. (5), while the baseline omits all of these. The gains in Table I could therefore come from extra parameters or the auxiliary regularizer rather than from pose information. Moreover, the scripted pose–force protocol in Sec. IV can create a correlation between pose and force labels; without a control in which pose channels are randomly shuffled or replaced with noise, the residual branch may learn a pose-to-force proxy rather than a pose-to-artifact correction. A capacity-matched tactile-only model and a pose-shuffle control are needed.
  2. [§VI.B, 'Statistical Analysis'] The headline MDF reductions are reported without per-user confidence intervals or a statistical test; the paper explicitly states that per-user MDF analysis was not performed due to limited data. With n=5 per glove, the MDF point estimates and their differences could be unstable. Please provide per-user MDF distributions, bootstrap confidence intervals, or a leave-one-session-out analysis. The Wilcoxon tests reported for BA, F1, MAE, and R² do not cover the metric highlighted in the abstract and title.
  3. [§I, §III, §VI.A] The mechanistic claim that 'explicitly accounts for pose-induced sensor deformations' is not directly tested. The paper notes in Sec. I that exact strain depends on unmeasured sensor mechanics, hand shape, and glove fit, and Sec. VI.A uses only four index-finger joint angles. Fig. 8 shows that pose carries predictive information for zero-load tactile signals, but there is no analysis of whether the learned residual actually tracks measured PRAs. A direct check—for example, correlating the predicted residual with the zero-load PRA magnitude across poses, or ablating pose-driven residual corrections—would provide the missing causal link and support generalization beyond the specific gloves tested.
  4. [§IV, §VI.B] The 'pseudo-unseen' generalization evaluation is based on free-movement substages within the same sessions, users, and glove fittings, and the chronological split (first 80% / last 20%) means the test segment comes from the same distribution and the same physical mounting. This is a reasonable temporal-split test, but the text should avoid implying broader generalization. The claim 'generalizes beyond prescribed movements' is accurate only for within-session movement diversity; it does not address new users, new glove instances, or new pose distributions. Consider clarifying this boundary in the results and discussion.
minor comments (5)
  1. [Author affiliation] Typo: 'Cornell Univeristy' should be 'Cornell University'.
  2. [§I] Missing space after period: 'contact forces.PRA is a fundamental challenge'.
  3. [References] Reference [9] is given as a bare URL; format it consistently with the other references (title, publisher, year).
  4. [Fig. 6] The caption says error bars and shaded areas indicate 95% confidence intervals, but the method of computation (e.g., bootstrap vs. normal approximation) is not stated. Please specify for reproducibility.
  5. [§VI.A] MDF is computed with 10 gF force bins; please state the bin-edge convention (e.g., [0,10), [10,20), ...) and how the 90% crossing is interpolated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pose-aware improvement is an empirical result on held-out temporal data, not a derivation reduced to its inputs.

full rationale

The paper's central claim is an empirical performance comparison, not a derivation. The pose-aware model computes F_hat = F_base - F_residual (Eq. 1), trained with force losses and an auxiliary zero-load tactile reconstruction loss (Eqs. 2-5), and is evaluated on a chronological 80/20 split per session ('we split the data chronologically: the first 80% of continuous samples for training and the last 20% for testing'). MDF, MAE, BA, and F1 are all measured on this held-out test data, so the reported reductions are not equal by construction to any training input. The 2.5gF touch threshold is a calibration choice ('select 2.5gF as the touch detection threshold as the force ground truths of all zero-load actions fall below 2.5gF') and the auxiliary loss uses zero-load frames, but these are training/calibration design choices rather than fitted outputs relabeled as predictions; the model must still generalize to unseen temporal data. Self-citations (e.g., [8], [20]) support background claims and are corroborated by the paper's own experiments, so they are not load-bearing. The lack of a capacity-matched or pose-shuffled ablation is a legitimate attribution/correctness concern, but it is not circularity under the stated rules.

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

The paper's empirical result rests on a set of hand-chosen hyperparameters and data-derived thresholds; the most consequential for the headline MDF numbers are the glove-dependent auxiliary loss weight λ, the 2.5gF touch-detection threshold, and the 10gF force bin used in the MDF definition. No new physical entities or conserved quantities are introduced.

free parameters (4)
  • Auxiliary loss weight λ = Not reported (glove-dependent, selected by sweeping)
    Weight for the zero-load PRA tactile loss in Eq. (4)-(5); the paper says λ is glove-dependent and selected by sweeping a range, but values are not reported and the validation split is unspecified.
  • Touch detection threshold = 2.5 gF
    Chosen because 'the force ground truths of all zero-load actions fall below 2.5gF' (Sec. VI.A); this is a data-derived threshold used in BA/F1/MDF evaluation.
  • MDF force bin size = 10 gF
    MDF is defined as the smallest force where balanced accuracy reaches 90%, with BA computed per 10gF bin; bin width affects the reported MDF values.
  • Per-session zero-load median normalization = Per-session channel median
    Tactile signals are normalized by subtracting the channel median of zero-load stages (Sec. VI.A); this is a data-dependent preprocessing choice that affects all results.
assumptions (4)
  • domain assumption Hand pose (joint angles) is a sufficient proxy for pose-induced sensor strain; four index-finger joint angles capture the deformation that produces PRAs.
    Stated in Sec. I and III; the paper notes exact strain depends on sensor mechanics, hand shapes, glove fit and cannot be measured without sensor modifications. The pose encoder uses only DIP, PIP, MCP flexion/extension and MCP abduction/adduction (Sec. VI.A). If joint angles fail to encode the deformation field, the residual branch cannot correct the artifact.
  • domain assumption The temporal 80/20 split within the same users is an adequate evaluation protocol; results generalize to the same user's later behavior.
    Sec. VI.A; all training and testing are within the same five users per glove, so cross-user or cross-session generalization is untested. The paper frames this as standard wearable sensing practice.
  • domain assumption Force ground truth from the ATI Nano17 load cell with a flat end effector, after dynamic taring, is a valid normal-force reference.
    Sec. IV and VI.A; a flat end effector does not capture the diversity of real manipulation, and the authors acknowledge this as future work.
  • ad hoc to paper The modified MAPE loss (denominator 1+Fi) is a reasonable objective that emphasizes low-force accuracy.
    Eq. (2); the base of 1 is stated to stabilize training and emphasize small forces; it is a design choice rather than a derived optimum.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pose-Aware Modeling to Mitigate Pose-Related Artifacts in Tactile Gloves." pith.science (2026). https://pith.science/paper/VCR2YS2O

@misc{pith2026260722964,
  author       = {Pith},
  title        = {Pith review of: Pose-Aware Modeling to Mitigate Pose-Related Artifacts in Tactile Gloves},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VCR2YS2O}},
  note         = {Machine review of arXiv:2607.22964}
}
read the original abstract

Tactile gloves digitize contact and force during hand-object interactions, enabling robotics applications in dexterous manipulation, teleoperation, and learning from demonstration. To preserve hand dexterity and capture the nuances of natural interactions, these gloves and the integrated tactile sensors are designed to be soft, flexible, and comfortable. However, such flexible sensors are sensitive not only to contact forces but also unavoidably to hand pose changes, resulting in pose-related artifacts (PRAs). PRAs are especially problematic in the low-force range, resulting in misdetections or late-onset detections of contact, which raises the minimum detectable force (MDF) of the glove. In this work, we characterize the PRAs in relation to pose and force. Building on these insights, we introduce a glove-agnostic algorithmic framework that leverages hand pose information, which is increasingly available, to mitigate PRAs without glove modifications. Our pose-aware force estimation model augments tactile-to-force pipelines with a residual prediction branch that explicitly accounts for pose-induced sensor deformations. We validate our approach across 3 glove designs and 15 users, reducing MDF by 10.4%, 12.2%, and 18.3%, with consistent improvements across all evaluated metrics. This method provides a practical path to improving the usability of tactile gloves in data collection and diverse robotic applications.

Figures

Figures reproduced from arXiv: 2607.22964 by the authors.

Figure 1
Figure 1. In tactile sensing gloves (such as Glove 2 shown at the top left), [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Examples of real-time PRAs mitigation for Glove 2 using our [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. (A) Different index finger poses result in different sensor responses [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: (A) Data collection setup. (B) Illustrations of the fingertip taxel [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: The pose-aware modeling framework takes tactile signals and joint angles as inputs, fuses pose and tactile information through a series of encoders [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Comparison of pose-aware and tactile-only models: (A) Force estimation MAE, (B) touch detection balanced accuracy, and (C) minimal detectable force (MDF, the smallest force where balanced accuracy reaches 90%). Results show improved performance with pose-aware models. …
Figure 7
Figure 7. Figure 7: In comparison to tactile-only estimated force, the pose-aware estimated force mitigates PRAs, especially for the low-force range, as shown in the bottom subplots where the blue shaded areas are the no-touch range (< 2.5 gF). (A) During in-air hand pose changes, tactile…
Figure 8
Figure 8. Figure 8: Reconstruction of tactile input (one channel from each glove) for [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 1 linked inside Pith

  1. [5]

    Posture-informed muscular force learning for robust hand pressure estimation,

    K. Seo, J. Seo, H. Jeong, S. Kim, and S. H. Yoon, “Posture-informed muscular force learning for robust hand pressure estimation,”Advances in Neural Information Processing Systems, vol. 37, pp. 87 831–87 873, 2024

    2024

  2. [1]

    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

  3. [2]

    Learning human–environment in- teractions using conformal tactile textiles,

    Y . Luo, Y . Li, P. Sharma, W. Shou, K. Wu, M. Foshey, B. Li, T. Pala- cios, A. Torralba, and W. Matusik, “Learning human–environment in- teractions using conformal tactile textiles,”Nature Electronics, vol. 4, no. 3, pp. 193–201, 2021

    2021

  4. [3]

    Capturing complex hand movements and object interactions using machine learning-powered stretchable smart textile gloves,

    A. Tashakori, Z. Jiang, A. Servati, S. Soltanian, H. Narayana, K. Le, C. Nakayama, C.-l. Yang, Z. J. Wang, J. J. Eng,et al., “Capturing complex hand movements and object interactions using machine learning-powered stretchable smart textile gloves,”Nature Machine Intelligence, vol. 6, no. 1, pp. 106–118, 2024

    2024

  5. [4]

    Vtdexmanip: A dataset and benchmark for visual-tactile pretraining and dexterous manipulation with reinforcement learning,

    Q. Liu, Y . Cui, Z. Sun, G. Li, J. Chen, and Q. Ye, “Vtdexmanip: A dataset and benchmark for visual-tactile pretraining and dexterous manipulation with reinforcement learning,” inThe Thirteenth Interna- tional Conference on Learning Representations

  6. [6]

    Learning to jointly understand visual and tactile signals,

    Y . Li, Y . Du, C. Liu, F. Williams, M. Foshey, B. Eckart, J. Kautz, J. B. Tenenbaum, A. Torralba, and W. Matusik, “Learning to jointly understand visual and tactile signals,” inThe Twelfth International Conference on Learning Representations, 2023

    2023

  7. [7]

    Advances in soft strain and pressure sensors,

    D. Van Nguyen, P. Song, F. Manshaii, J. Bell, J. Chen, and T. Dinh, “Advances in soft strain and pressure sensors,”ACS Nano, vol. 19, no. 7, p. 6663–6704, Feb 2025

    2025

  8. [8]

    Fits like a flex-glove: Automatic design of personalized fpcb-based tactile sensing gloves,

    D. Murphy, Y . Li, C. E. Owens, L. Stanton, P. P. Liang, Y . Luo, A. Torralba, and W. Matusik, “Fits like a flex-glove: Automatic design of personalized fpcb-based tactile sensing gloves,” inProceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems, 2025, pp. 1–8

    2025

Show all 31 references
  1. [9]

    Available: https://pressureprofile.com/ interpreting-tactileglove-data-for-practical-use-cases

    [Online]. Available: https://pressureprofile.com/ interpreting-tactileglove-data-for-practical-use-cases

  2. [10]

    Capturing forceful interaction with deformable objects using a deep learning-powered stretchable tactile array,

    C. Jiang, W. Xu, Y . Li, Z. Yu, L. Wang, X. Hu, Z. Xie, Q. Liu, B. Yang, X. Wang,et al., “Capturing forceful interaction with deformable objects using a deep learning-powered stretchable tactile array,”Nature Communications, vol. 15, no. 1, p. 9513, 2024

    2024

  3. [11]

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

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

    2025

  4. [12]

    Fingertip forces and completion time for index finger and thumb touchscreen gestures,

    D. S. Asakawa, G. H. Crocker, A. Schmaltz, and D. L. Jindrich, “Fingertip forces and completion time for index finger and thumb touchscreen gestures,”Journal of Electromyography and Kinesiology, vol. 34, pp. 6–13, 2017

    2017

  5. [13]

    Stretchable and all-directional strain-insensitive electronic glove for robotic skins and human–machine interfacing,

    S. Sharma, G. B. Pradhan, S. Jeong, S. Zhang, H. Song, and J. Y . Park, “Stretchable and all-directional strain-insensitive electronic glove for robotic skins and human–machine interfacing,”ACS nano, vol. 17, no. 9, pp. 8355–8366, 2023

    2023

  6. [14]

    Available: https://optitrack.com/

    [Online]. Available: https://optitrack.com/

  7. [15]

    Analysis of the hands in egocentric vision: A survey,

    A. Bandini and J. Zariffa, “Analysis of the hands in egocentric vision: A survey,”IEEE transactions on pattern analysis and machine intelligence, vol. 45, no. 6, pp. 6846–6866, 2020

    2020

  8. [16]

    Reconstructing hands in 3d with transformers,

    G. Pavlakos, D. Shan, I. Radosavovic, A. Kanazawa, D. Fouhey, and J. Malik, “Reconstructing hands in 3d with transformers,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 9826–9836

    2024

  9. [17]

    A systematic review on custom data gloves,

    V . Belcamino, A. Carf `ı, and F. Mastrogiovanni, “A systematic review on custom data gloves,”IEEE Transactions on Human-Machine Sys- tems, vol. 54, no. 5, pp. 520–535, 2024

    2024

  10. [18]

    Pressurevision++: Estimating fingertip pressure from diverse rgb images,

    P. Grady, J. A. Collins, C. Tang, C. D. Twigg, K. Aneja, J. Hays, and C. C. Kemp, “Pressurevision++: Estimating fingertip pressure from diverse rgb images,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 8698–8708

    2024

  11. [19]

    Egopressure: A dataset for hand pressure and pose estimation in egocentric vision,

    Y . Zhao, T. Kwon, P. Streli, M. Pollefeys, and C. Holz, “Egopressure: A dataset for hand pressure and pose estimation in egocentric vision,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 27 727–27 738

    2025

  12. [20]

    Tactile embeddings for multi-task learning,

    Y . Luo, M. Wonsick, J. Hodgins, and B. Okorn, “Tactile embeddings for multi-task learning,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 3348–3355

    2024

  13. [21]

    Grasp like humans: Learning generalizable multifingered grasping from human proprioceptive sensorimotor integration,

    C. Guo, X. Chen, Z. Zeng, Z. Guo, Y . Li, H. Xiao, D. Hu, and H. Lu, “Grasp like humans: Learning generalizable multifingered grasping from human proprioceptive sensorimotor integration,”IEEE Transactions on Robotics, vol. 41, pp. 5700–5719, 2025

    2025

  14. [22]

    Learning-enhanced electronic skin for tactile sensing on deformable surface based on electrical impedance tomography,

    H. Dong, X. Wu, D. Hu, Z. Liu, F. Giorgio-Serchi, and Y . Yang, “Learning-enhanced electronic skin for tactile sensing on deformable surface based on electrical impedance tomography,”IEEE Transac- tions on Instrumentation and Measurement, 2025

    2025

  15. [23]

    Data-efficient control policy search using residual dynamics learning,

    M. Saveriano, Y . Yin, P. Falco, and D. Lee, “Data-efficient control policy search using residual dynamics learning,” in2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017, pp. 4709–4715

    2017

  16. [24]

    Resid- ual rotation correction using tactile equivariance,

    Y . Zhu, Z. Ye, B. Hu, H. Zhao, Y . Qi, D. Wang, and R. Platt, “Resid- ual rotation correction using tactile equivariance,”arXiv preprint arXiv:2511.07381, 2025

    2025

  17. [25]

    Sim- to-real of soft robots with learned residual physics,

    J. Gao, M. Y . Michelis, A. Spielberg, and R. K. Katzschmann, “Sim- to-real of soft robots with learned residual physics,”IEEE Robotics and Automation Letters, 2024

    2024

  18. [26]

    Tensortouch: Calibration of tactile sensors for high resolution stress tensor and deformation for dexterous manipulation,

    W. K. Do, M. Strong, A. Swann, B. Lei, and M. Kennedy III, “Tensortouch: Calibration of tactile sensors for high resolution stress tensor and deformation for dexterous manipulation,”arXiv preprint arXiv:2506.08291, 2025

    2025 arXiv

  19. [27]

    Available: https://pressureprofile.com/ body-pressure-mapping/tactile-glove

    [Online]. Available: https://pressureprofile.com/ body-pressure-mapping/tactile-glove

  20. [28]

    Temporal convolutional networks for action segmentation and detection,

    C. Lea, M. D. Flynn, R. Vidal, A. Reiter, and G. D. Hager, “Temporal convolutional networks for action segmentation and detection,” in proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 156–165

    2017

  21. [29]

    On the continuity of rotation representations in neural networks,

    Y . Zhou, C. Barnes, J. Lu, J. Yang, and H. Li, “On the continuity of rotation representations in neural networks,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 5745–5753

    2019

  22. [30]

    Disentangling and unifying graph convolutions for skeleton-based action recogni- tion,

    Z. Liu, H. Zhang, Z. Chen, Z. Wang, and W. Ouyang, “Disentangling and unifying graph convolutions for skeleton-based action recogni- tion,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 143–152

    2020

  23. [31]

    Revisiting skeleton- based action recognition,

    H. Duan, Y . Zhao, K. Chen, D. Lin, and B. Dai, “Revisiting skeleton- based action recognition,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 2969–2978

    2022

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.