pith. sign in

arxiv: 2604.22499 · v1 · submitted 2026-04-24 · 💻 cs.LG · cs.RO

Decoding High-Dimensional Finger Motion from EMG Using Riemannian Features and RNNs

Martin Colot , C\'edric Simar , Guy Cheron , Ana Maria Cebolla Alvarez , Gianluca Bontempi This is my paper

Pith reviewed 2026-05-08 12:26 UTC · model grok-4.3

classification 💻 cs.LG cs.RO
keywords EMGfinger kinematicsRiemannian covarianceGRUregressionprostheticsreal-time decodingcross-subject generalization
0
0 comments X

The pith

A Temporal Riemannian Regressor using GRU and multi-band covariance features decodes 15 finger joint angles from 8-channel EMG with lower error than prior methods and runs near 10 predictions per second on a Raspberry Pi.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper builds a complete pipeline for continuous regression of high-dimensional finger kinematics directly from forearm EMG signals collected with low-cost hardware. It releases the EMG-FK dataset of synchronized 8-channel EMG and 15 joint angles from 20 participants and introduces the TRR model that feeds sequences of Riemannian covariance features into a lightweight GRU. The approach reports lower average absolute errors than existing methods in both intra-subject and cross-subject tests on EMG-FK and the emg2pose benchmark while demonstrating embedded real-time execution. A sympathetic reader would care because accurate, calibration-light finger decoding could enable more natural prosthetic hands, virtual-reality interfaces, and teleoperated systems without laboratory-grade equipment.

Core claim

The authors establish that sequences of multi-band Riemannian covariance features extracted from 8-channel EMG and processed by a GRU-based regressor can decode 15 finger joint angles with an intra-subject average absolute error of 9.79 deg and a cross-subject error of 16.71 deg, outperforming state-of-the-art methods on both their new EMG-FK dataset and the public emg2pose benchmark while achieving nearly 10 predictions per second on a Raspberry Pi 5.

What carries the argument

The Temporal Riemannian Regressor (TRR), a lightweight GRU that ingests sequences of multi-band Riemannian covariance features to perform continuous EMG-to-finger-kinematics regression.

If this is right

  • High-dimensional continuous finger control becomes feasible with only an 8-channel consumer EMG armband and a single webcam.
  • Real-time inference at nearly 10 predictions per second on a Raspberry Pi enables portable, battery-powered prosthetic or interface devices.
  • Improved cross-subject performance reduces the need for extensive per-user training data or calibration sessions.
  • Lower computational demands than prior approaches support broader deployment in embedded EMG-based control systems.

Where Pith is reading between the lines

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

  • The same Riemannian feature pipeline could be tested on other biosignals such as EEG for related decoding tasks.
  • The automatic synchronization procedure might be adapted to collect ground-truth labels for additional multimodal sensor combinations.
  • Deployment on actual prosthetic hardware would provide a direct test of whether the reported accuracy translates to intuitive user control.
  • Further reduction of the residual error could open finer motor tasks such as individual finger typing in virtual environments.

Load-bearing premise

The automatic synchronization between the 8-channel EMG armband and webcam-tracked joint angles produces sufficiently accurate ground-truth labels without introducing systematic bias or noise.

What would settle it

An independent recording of the same hand motions that uses optical motion capture with sub-degree precision and shows that the reported absolute errors rise substantially or that cross-subject generalization collapses.

Figures

Figures reproduced from arXiv: 2604.22499 by Ana Maria Cebolla Alvarez, C\'edric Simar, Gianluca Bontempi, Guy Cheron, Martin Colot.

Figure 1
Figure 1. Figure 1: Pipeline for data acquisition and processing view at source ↗
Figure 2
Figure 2. Figure 2: Automatic procedure for synchronization of EMG and j view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the EMG-FK and emg2pose datasets (joint-angles and EMG signals). The EMG from EMG-FK in this plot are processed using CAR. The EMG channel displayed in red represents the average value of all the recorded channels in EMG-FK before CAR. The bottom left plot shows the total explained variance ratio obtained with PCA on the joint angles of the two datasets, and on emg2pose with only the 15 joint… view at source ↗
Figure 4
Figure 4. Figure 4: Evaluation of TRR on two datasets in comparison with state-of-the-art approaches, in intra- and cross-subject configurations. For statistical analysis, we compute the average rank of the models (higher is better). Models that are not significantly different are connected by a black line (p = 0.05, Nemenyi test [53]). The critical distance (CD) indicates when models are considered statistically different. o… view at source ↗
Figure 5
Figure 5. Figure 5: B, we compare the predictions of TRR in intra￾subject configuration with those from TRR in cross-subject configuration. In this setting, the difference is more important, but the overall shape of the cross-subject prediction still appears to match most of the real kinematics. This observation suggests that, although there is still room for improvement of cross-subject regressors, TRR manages to capture rel… view at source ↗
Figure 5
Figure 5. Figure 5: Visualization of the predicted joint-angles. view at source ↗
Figure 6
Figure 6. Figure 6: Evaluation of TRR on the different fingers joints, using all subjects from our EMG-FK dataset. The absolute error provides an estimation of the error angle in degrees. EMG envelope using the model described in Section II-E4. The results, reported in view at source ↗
Figure 7
Figure 7. Figure 7: Results of the ablation study. The results in red corr view at source ↗
Figure 8
Figure 8. Figure 8: Evolution of the Raspberry Pi 5 CPU’s temperature over time with different regression models including TRR and its simplified version (single frequency band), vemg2pose, and a simple MLP on TDF to serve as baseline. 5) Limitations toward myoelectric prosthesis control: While the results of this study demonstrate that TRR outper￾forms state-of-the-art methods in predicting high-degree-of￾freedom finger join… view at source ↗
Figure 10
Figure 10. Figure 10: Evaluation of TRR with different temporal offset between EMG and joint-angle kinematics view at source ↗
read the original abstract

Continuous estimation of high-dimensional finger kinematics from forearm surface electromyography (EMG) could enable natural control for hand prostheses, AR/XR interfaces, and teleoperation. However, the complexity of human hand gestures and the entanglement of forearm muscles make accurate recognition intrinsically challenging. Existing approaches typically reduce task complexity by relying on classification-based machine learning, limiting the controllable degrees of freedom and compromising on natural interaction. We present an end-to-end framework for continuous EMG-to-kinematics regression using only consumer-grade hardware. The framework combines an 8-channel EMG armband, a single webcam, and an automatic synchronization procedure, enabling the collection of the EMG Finger-Kinematics dataset (EMG-FK), a 10-h dataset of synchronized EMG and 15 finger joint angles from 20 participants performing rich, unconstrained right-hand motions. We also introduce the Temporal Riemannian Regressor (TRR), a lightweight GRU-based model that uses sequences of multi-band Riemannian covariance features to decode finger motion. Across EMG-FK and the public emg2pose benchmark, TRR outperforms state-of-the-art methods in both intra- and cross-subject evaluation. On EMG-FK, it reaches an average absolute error of $9.79 \deg \pm 1.48$ in intra-subject and $16.71 \deg \pm 3.97$ in cross-subject. Finally, we demonstrate real-time deployment on a Raspberry Pi 5 and intuitive control of a robotic hand; TRR runs at nearly 10 predictions/s and is roughly an order of magnitude faster than state-of-the-art approaches. Together, these contributions lower the barrier to reproducible, real-time EMG-based decoding of high-dimensional finger motion, and pave the way toward more natural and intuitive control of embedded EMG-based systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims to introduce an end-to-end framework for continuous high-dimensional finger kinematics regression from 8-channel forearm EMG using only consumer hardware. It collects the EMG-FK dataset (10 hours, 20 subjects, 15 joint angles) via an armband plus single-webcam tracking with automatic synchronization, proposes the Temporal Riemannian Regressor (TRR) that feeds multi-band Riemannian covariance features into a lightweight GRU, reports superior intra-subject (9.79° ± 1.48 MAE) and cross-subject (16.71° ± 3.97 MAE) performance versus prior methods on EMG-FK and emg2pose, and demonstrates real-time inference at ~10 Hz on a Raspberry Pi 5 with robotic-hand control.

Significance. If the webcam-derived ground-truth labels prove sufficiently accurate, the work would meaningfully lower barriers to accessible, high-DoF EMG decoding for prosthetics, AR/XR, and teleoperation by combining a large public dataset, a practical Riemannian-GRU architecture, and embedded deployment. The cross-subject generalization and order-of-magnitude speed-up over prior approaches are potentially valuable contributions to the field.

major comments (2)
  1. [Dataset Collection / EMG-FK construction] Dataset collection and label generation (described in the abstract and methods): the paper provides no quantitative validation of the webcam-based 3D joint-angle tracking or the automatic synchronization step against a gold-standard motion-capture system. Without reported per-joint errors, repeatability statistics, or synchronization residual distributions, it remains possible that systematic label noise or bias inflates the reported MAE values; this assumption is load-bearing for every intra- and cross-subject performance claim and the “outperforms SOTA” statement.
  2. [Experimental Evaluation] Evaluation protocol (results and experimental sections): the abstract and summary report mean absolute errors with standard deviations but do not describe data-split procedures, subject-wise cross-validation details, or statistical significance tests comparing TRR against baselines. Without these, it is impossible to assess whether the claimed improvements are robust or could arise from post-hoc choices.
minor comments (1)
  1. [Abstract] Notation: the abstract uses “deg” rather than the conventional “°” symbol for degrees; this should be standardized throughout.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback. The comments highlight important aspects of reproducibility and validation that we will address in the revision. Below we respond point-by-point to the major comments.

read point-by-point responses

  1. Referee: [Dataset Collection / EMG-FK construction] Dataset collection and label generation (described in the abstract and methods): the paper provides no quantitative validation of the webcam-based 3D joint-angle tracking or the automatic synchronization step against a gold-standard motion-capture system. Without reported per-joint errors, repeatability statistics, or synchronization residual distributions, it remains possible that systematic label noise or bias inflates the reported MAE values; this assumption is load-bearing for every intra- and cross-subject performance claim and the “outperforms SOTA” statement.

    Authors: We acknowledge that direct quantitative validation against optical motion capture would strengthen confidence in the absolute error values. Our design choice was to prioritize an accessible, low-cost pipeline using only consumer hardware, which is central to the contribution. In the revised manuscript we will expand the Dataset section with additional details on the automatic synchronization procedure (including any post-hoc consistency checks performed), cite supporting literature on the accuracy of single-webcam 3D hand tracking, and add a dedicated Limitations paragraph that explicitly discusses potential label noise. Because new motion-capture experiments are not feasible within this revision cycle, we cannot supply per-joint mocap residuals; however, all compared methods are evaluated on identical labels, so relative performance claims remain valid. revision: partial

  2. Referee: [Experimental Evaluation] Evaluation protocol (results and experimental sections): the abstract and summary report mean absolute errors with standard deviations but do not describe data-split procedures, subject-wise cross-validation details, or statistical significance tests comparing TRR against baselines. Without these, it is impossible to assess whether the claimed improvements are robust or could arise from post-hoc choices.

    Authors: We agree that the evaluation protocol must be described more explicitly. In the revised Experimental Evaluation section we will add: (i) precise train/validation/test split ratios and subject-wise partitioning for both intra-subject (per-subject 5-fold cross-validation) and cross-subject (leave-one-subject-out) settings, (ii) implementation details for all baselines to ensure fair comparison, and (iii) statistical significance testing (paired t-tests across subjects with reported p-values) to quantify whether TRR improvements are robust. These additions will be placed before the results tables so readers can immediately assess the reliability of the reported gains. revision: yes

standing simulated objections not resolved
  • Quantitative validation of webcam-based 3D joint-angle tracking against a gold-standard motion-capture system (new experiments required)

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's central claims rest on empirical supervised regression performance of a GRU model trained on a newly collected EMG-FK dataset (EMG signals paired with webcam-derived joint angles) and evaluated on emg2pose. Riemannian covariance features are drawn from established prior signal-processing literature rather than defined in terms of the target kinematics or the model's outputs. No equation or procedure reduces a claimed prediction to a fitted input by construction, nor does any load-bearing premise collapse to a self-citation chain. The reported MAE values and real-time benchmarks are obtained from standard train/test splits and hardware timing, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 0 axioms · 0 invented entities

The central claim rests on the assumption that Riemannian geometry applied to EMG covariance matrices yields informative features for kinematics regression and that the collected dataset is representative of real-world use.

free parameters (2)
  • multi-band frequency ranges
    Choice of frequency bands for covariance computation is a modeling decision that must be selected or tuned.
  • GRU hidden size and sequence length
    Architectural hyperparameters that affect capacity and temporal modeling.

pith-pipeline@v0.9.0 · 5641 in / 1162 out tokens · 24331 ms · 2026-05-08T12:26:13.136583+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

59 extracted references · 53 canonical work pages

  1. [1]

    MediaPipe Hands: On-device Real-time Hand Tracking,

    F. Zhang, V . Bazarevsky, A. V akunov, A. Tkachenka, G. Sun g, C.-L. Chang, and M. Grundmann, “Mediapipe hands: On-device real-time hand tracking,” arXiv preprint arXiv:2006.10214 , 2020, https://doi.org/10.48550/arXiv.2006.10214

    work page doi:10.48550/arxiv.2006.10214 2006

  2. [2]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    A. Spurr, A. Dahiya, X. Wang, X. Zhang, and O. Hilliges, “Self-supervised 3d hand pose estimation from monocular rg b via contrastive learning,” in Proceedings of the IEEE/CVF interna- tional conference on computer vision , 2021, pp. 11 230–11 239, https://doi.org/10.1109/iccv48922.2021.01104

    work page doi:10.1109/iccv48922.2021.01104 2021

  3. [3]

    Hand tracking for immersive virtual rea lity: opportu- nities and challenges,

    G. Buckingham, “Hand tracking for immersive virtual rea lity: opportu- nities and challenges,” Frontiers in Virtual Reality , vol. 2, p. 728461, 2021, https://doi.org/10.3389/frvir.2021.728461

    work page doi:10.3389/frvir.2021.728461 2021

  4. [4]

    Virtual/augmented reality for re habilitation applications using electromyography as control/biofeedb ack: systematic literature review,

    C. L. Toledo-Peral, G. V ega-Martinez, J. A. Mercado-Gut i´ errez, G. Rodriguez-Reyes, A. V era-Hernandez, L. Leija-Salas, an d J. Gutierrez-Martinez, “Virtual/augmented reality for re habilitation applications using electromyography as control/biofeedb ack: systematic literature review,” Electronics, vol. 11, no. 14, p. 2271, 2022, https://doi.org/10.339...

    work page doi:10.3390/electronics11142271 2022

  5. [5]

    Teleo perated robotic arm movement using electromyography signal with wearable myo armband,

    H. F. Hassan, S. J. Abou-Loukh, and I. K. Ibraheem, “Teleo perated robotic arm movement using electromyography signal with wearable myo armband,” Journal of King Saud University- Engineering Sciences , vol. 32, no. 6, pp. 378–387, 2020, https://doi.org/10.1016/j.jksues.2019.05.001

    work page doi:10.1016/j.jksues.2019.05.001 2020

  6. [6]

    Maestro: An emg-driven assistive hand exoskeleton for spinal cord injury patients,

    Y . Y un, S. Dancausse, P . Esmatloo, A. Serrato, C. A. Merri ng, P . Agar- wal, and A. D. Deshpande, “Maestro: An emg-driven assistive hand exoskeleton for spinal cord injury patients,” in 2017 IEEE international conference on robotics and automation (ICRA) . IEEE, 2017, pp. 2904– 2910, https://doi.org/10.1109/ICRA.2017.7989337

    work page doi:10.1109/icra.2017.7989337 2017

  7. [7]

    Sign lan- guage recognition using the electromyographic signal: a sy stem- atic literature review,

    A. Ben Haj Amor, O. El Ghoul, and M. Jemni, “Sign lan- guage recognition using the electromyographic signal: a sy stem- atic literature review,” Sensors, vol. 23, no. 19, p. 8343, 2023, https://doi.org/10.3390/s23198343

    work page doi:10.3390/s23198343 2023

  8. [8]

    A revi ew of myoelectric control for prosthetic hand manipulation,

    Z. Chen, H. Min, D. Wang, Z. Xia, F. Sun, and B. Fang, “A revi ew of myoelectric control for prosthetic hand manipulation,” Biomimetics, vol. 8, no. 3, p. 328, 2023, https://doi.org/10.3390/biomi metics8030328

    work page doi:10.3390/biomi 2023

  9. [9]

    Design of upper limb prosth esis using real-time motion detection method based on emg signal proce ssing,

    N. N. Unanyan and A. A. Belov, “Design of upper limb prosth esis using real-time motion detection method based on emg signal proce ssing,” Biomedical Signal Processing and Control , vol. 70, p. 103062, 2021, https://doi.org/10.1016/j.bspc.2021.103062

    work page doi:10.1016/j.bspc.2021.103062 2021

  10. [10]

    A Review on EMG Signal Classification and Applications

    E. L. W. Ting, A. Chai, and L. P . Chin, “A review on emg signal classification and applications,” International Journal of Signal Processing Systems , vol. 10, no. 1, pp. 1–6, 2022, https://doi.org/10.18178/ijsps.10.1.1-6

    work page doi:10.18178/ijsps.10.1.1-6 2022

  11. [11]

    A Comprehensive Survey of Machine Learning Algorithms in Hand Gesture Recognition for Myoelectric Prosthetic Control: Current Trends, Challenges and Future Directions

    P . Tamilvanan, V . K. Elumalai, and B. Elumalai, “A compr ehensive survey of machine learning algorithms in hand gesture recog nition for myoelectric prosthetic control: Current trends, challeng es and future directions,” Archives of Computational Methods in Engineering , pp. 1– 30, 2025, https://doi.org/10.1007/s11831-025-10480-9

    work page doi:10.1007/s11831-025-10480-9 2025

  12. [12]

    Machine learning for hand pose classification from phasic and tonic emg signals during bimanual activities in v ir- tual reality,

    C. Simar, M. Colot, A.-M. Cebolla, M. Petieau, G. Cheron , and G. Bontempi, “Machine learning for hand pose classification from phasic and tonic emg signals during bimanual activities in v ir- tual reality,” Frontiers in Neuroscience , vol. 18, p. 1329411, 2024, https://doi.org/10.3389/fnins.2024.1329411

    work page doi:10.3389/fnins.2024.1329411 2024

  13. [13]

    Li near non-conservative unsupervised domain adaptation for cros s-subject emg gesture recognition,

    M. Colot, C. Simar, A. M. C. Alvarez, and G. Bontempi, “Li near non-conservative unsupervised domain adaptation for cros s-subject emg gesture recognition,” Biomedical Signal Processing and Control , vol. 111, p. 108283, 2026, https://doi.org/10.1016/j.bspc.20 25.108283

    work page doi:10.1016/j.bspc.20 2026

  14. [14]

    emg2pose: A large and diverse benchmark for surface electromyographic hand p ose esti- mation,

    S. Salter, R. Warren, C. Schlager, A. Spurr, S. Han, R. Bh asin, Y . Cai, P . Walkington, A. Bolarinwa, R. J. Wang et al. , “emg2pose: A large and diverse benchmark for surface electromyographic hand p ose esti- mation,” Advances in Neural Information Processing Systems , vol. 37, pp. 55 703–55 728, 2024, https://doi.org/10.52202/079017 -1770

    work page doi:10.52202/079017 2024

  15. [15]

    Young, Martin Lebrat, and Markus Greiner

    P . Kaifosh and T. R. Reardon, “A generic non-invasive ne uromotor interface for human-computer interaction,” Nature, vol. 645, no. 8081, pp. 702–711, 2025, https://doi.org/10.1038/s41586-025- 09255-w

    work page doi:10.1038/s41586-025- 2025

  16. [16]

    Current rates of prosth etic usage in upper-limb amputees–have innovations had an impac t on device acceptance?

    S. Salminger, H. Stino, L. H. Pichler, C. Gstoettner, A. Sturma, J. A. Mayer, M. Szivak, and O. C. Aszmann, “Current rates of prosth etic usage in upper-limb amputees–have innovations had an impac t on device acceptance?” Disability and rehabilitation , vol. 44, no. 14, pp. 3708– 3713, 2022, https://doi.org/10.1080/09638288.2020.186 6684

    work page doi:10.1080/09638288.2020.186 2022

  17. [17]

    Handrix: animating the human h and,

    G. ElKoura and K. Singh, “Handrix: animating the human h and,” in Proceedings of the 2003 ACM SIGGRAPH/Eurographics symposium on Computer animation , 2003, pp. 110–119, https://doi.org/10.2312/SCA03/110-119

    work page doi:10.2312/sca03/110-119 2003

  18. [18]

    Accurate continuous prediction of 14 de grees of freedom of the hand from myoelectrical signals through conv olutive deep learning,

    R. C. Sˆ ımpetru, M. Osswald, D. I. Braun, D. S. Oliveira, A. L. Cakici, and A. Del V ecchio, “Accurate continuous prediction of 14 de grees of freedom of the hand from myoelectrical signals through conv olutive deep learning,” in 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) . IEEE, 2022, pp. 702–7...

    work page doi:10.1109/embc48229.2 2022

  19. [19]

    T rans- lating semg signals to continuous hand poses using recurren t neu- ral networks,

    F. Quivira, T. Koike-Akino, Y . Wang, and D. Erdogmus, “T rans- lating semg signals to continuous hand poses using recurren t neu- ral networks,” in 2018 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI) . IEEE, 2018, pp. 166–169, https://doi.org/10.1109/BHI.2018.8333395

    work page doi:10.1109/bhi.2018.8333395 2018

  20. [20]

    Effec t of user practice on prosthetic finger control with an intuitive myoe lec- tric decoder,

    A. Krasoulis, S. Vijayakumar, and K. Nazarpour, “Effec t of user practice on prosthetic finger control with an intuitive myoe lec- tric decoder,” Frontiers in neuroscience , vol. 13, p. 891, 2019, https://doi.org/10.3389/fnins.2019.00891

    work page doi:10.3389/fnins.2019.00891 2019

  21. [21]

    W earable high-density emg sleeve for complex hand gesture classifica tion and continuous joint angle estimation,

    N. Tacca, C. Dunlap, S. P . Donegan, J. O. Hardin, E. Meyer s, M. J. Darrow, S. Colachis IV , A. Gillman, and D. A. Friedenberg, “W earable high-density emg sleeve for complex hand gesture classifica tion and continuous joint angle estimation,” Scientific Reports , vol. 14, no. 1, p. 18564, 2024, https://doi.org/10.1038/s41598-024-64458 -x. 11

    work page doi:10.1038/s41598-024-64458 2024

  22. [22]

    Emg-based deco d- ing of manipulation motions in virtual reality: Towards imm er- sive interfaces,

    A. Dwivedi, Y . Kwon, and M. Liarokapis, “Emg-based deco d- ing of manipulation motions in virtual reality: Towards imm er- sive interfaces,” in 2020 IEEE International Conference on Sys- tems, Man, and Cybernetics (SMC) . IEEE, 2020, pp. 3296–3303, https://doi.org/10.1109/SMC42975.2020.9283270

    work page doi:10.1109/smc42975.2020.9283270 2020

  23. [23]

    Finger angle estimation from array emg system using linear regres- sion model with independent component analysis,

    S. Stapornchaisit, Y . Kim, A. Takagi, N. Y oshimura, and Y . Koike, “Finger angle estimation from array emg system using linear regres- sion model with independent component analysis,” Frontiers in neuro- robotics, vol. 13, p. 75, 2019, https://doi.org/10.3389/fnbot.201 9.00075

    work page doi:10.3389/fnbot.201 2019

  24. [24]

    Multi–joint angles estimation of forearm motion using a re gres- sion model,

    Z. Qin, S. Stapornchaisit, Z. He, N. Y oshimura, and Y . Ko ike, “Multi–joint angles estimation of forearm motion using a re gres- sion model,” Frontiers in Neurorobotics , vol. 15, p. 685961, 2021, https://doi.org/10.3389/fnbot.2021.685961

    work page doi:10.3389/fnbot.2021.685961 2021

  25. [25]

    Simultaneous es - timation of hand joints’ angles toward semg-driven human–r obot interaction,

    W. Hai, T. Qing, S. Na, and Z. Xiaodong, “Simultaneous es - timation of hand joints’ angles toward semg-driven human–r obot interaction,” Ieee Access , vol. 10, pp. 109 385–109 394, 2022, https://doi.org/10.1109/ACCESS.2022.3212046

    work page doi:10.1109/access.2022.3212046 2022

  26. [26]

    semg-based hand moveme nt regression by prediction of joint angles with recurrent neu ral networks,

    P . Koch, K. Mohammad-Zadeh, M. Maass, M. Dreier, O. Thom sen, T. J. Parbs, H. Phan, and A. Mertins, “semg-based hand moveme nt regression by prediction of joint angles with recurrent neu ral networks,” in 2021 43rd Annual International Conference of the IEEE Engin eering in Medicine & Biology Society (EMBC) . IEEE, 2021, pp. 6519–6523, https://doi.org/10...

    work page doi:10.1109/embc46164.2021.9630042 2021

  27. [27]

    Estim at- ing finger joint angles on surface emg using manifold learnin g and long short-term memory with attention mechanism,

    C. Avian, S. W. Prakosa, M. Faisal, and J.-S. Leu, “Estim at- ing finger joint angles on surface emg using manifold learnin g and long short-term memory with attention mechanism,” Biomed- ical Signal Processing and Control , vol. 71, p. 103099, 2022, https://doi.org/10.1016/j.bspc.2021.103099

    work page doi:10.1016/j.bspc.2021.103099 2022

  28. [28]

    Neuropose: 3d hand pose t racking using emg wearables,

    Y . Liu, S. Zhang, and M. Gowda, “Neuropose: 3d hand pose t racking using emg wearables,” in Proceedings of the W eb Conference 2021 , 2021, pp. 1471–1482, https://doi.org/10.1145/3442381.3 449890

    work page doi:10.1145/3442381.3 2021

  29. [29]

    Sensing the full dynamics of the human hand with a neural interface and deep learning,

    R. C. Sˆ ımpetru, A. Arkudas, D. I. Braun, M. Osswald, D. S . de Oliveira, B. Eskofier, T. M. Kinfe, and A. Del V ecchio, “Sensing the full dynamics of the human hand with a neural interface and deep learning,” BioRxiv, pp. 2022–07, 2022, https://doi.org/10.1101/2022.07.29. 502064

    work page doi:10.1101/2022.07.29 2022

  30. [30]

    Continuous gest ure recognition from semg sensor data with recurrent neural networks and adv ersarial domain adaptation,

    I. Sosin, D. Kudenko, and A. Shpilman, “Continuous gest ure recognition from semg sensor data with recurrent neural networks and adv ersarial domain adaptation,” in 2018 15Th international conference on control, automation, robotics and vision (ICARCV) . IEEE, 2018, pp. 1436– 1441, https://doi.org/10.1109/ICARCV .2018.8581206

    work page doi:10.1109/icarcv 2018

  31. [31]

    Continuous and sim ultaneous estimation of finger kinematics using inputs from an EMG- to-muscle activation model,

    J. G. Ngeo, T. Tamei, and T. Shibata, “Continuous and sim ultaneous estimation of finger kinematics using inputs from an EMG- to-muscle activation model,” Journal of NeuroEngineering and Rehabilitation, vol. 11, no. 1, p. 122, 2014. [Online]. Available: https://doi.org,/10.1186/1743-0003-11-122

    work page doi:10.1186/1743-0003-11-122 2014

  32. [32]

    A deep kalman filter network for hand kinematics estimation us ing semg,

    T. Bao, Y . Zhao, S. A. R. Zaidi, S. Xie, P . Y ang, and Z. Zhan g, “A deep kalman filter network for hand kinematics estimation us ing semg,” Pattern Recognition Letters , vol. 143, pp. 88–94, 2021. [Online]. Available: https://doi.org/10.1016/j.patrec.2021.01.001

    work page doi:10.1016/j.patrec.2021.01.001 2021

  33. [33]

    semg-based regression of hand kinematics with temporal convolutional networks on a low-power edge microcontrolle r,

    M. Zanghieri, S. Benatti, A. Burrello, V . J. Kartsch Mor inigo, R. Meattini, G. Palli, C. Melchiorri, and L. Benini, “semg-based regression of hand kinematics with temporal convolutional networks on a low-power edge microcontrolle r,” in 2021 IEEE International Conference on Omni-Layer Intellig ent Systems (COINS) . IEEE, 2021, pp. 1–6. [Online]. Availab...

    work page doi:10.1109/coins51742.2021.9524188 2021

  34. [34]

    Estimating finger joint angles by surface emg signal using f eature extraction and transformer-based deep learning model,

    N. A. S. Putro, C. Avian, S. W. Prakosa, M. I. Mahali, and J .-S. Leu, “Estimating finger joint angles by surface emg signal using f eature extraction and transformer-based deep learning model,” Biomedical Signal Processing and Control , vol. 87, p. 105447, 2024. [Online]. Available: https://doi.org/10.1016/j.bspc.2023.105447

    work page doi:10.1016/j.bspc.2023.105447 2024

  35. [35]

    Continuous estimation of finger kin ematics based on fusion inception and transformer network from surf ace emg,

    C. Lin and X. Zhang, “Continuous estimation of finger kin ematics based on fusion inception and transformer network from surf ace emg,” Frontiers in Neurorobotics, vol. 18, p. 1305605, 2024. [Online]. Available: https://doi.org/10.3389/fnbot.2024.1305605

    work page doi:10.3389/fnbot.2024.1305605 2024

  36. [36]

    Real-time decomposition of surface electromyography into hand joint angles using a parallel an d efficient transformer network,

    C. Lin, X. Zhang, and C. Zhao, “Real-time decomposition of surface electromyography into hand joint angles using a parallel an d efficient transformer network,” Scientific Reports , vol. 15, p. 36150, 2025. [Online]. Available: https://doi.org/10.1038/s41598-0 25-16268-y

    work page doi:10.1038/s41598-0 2025

  37. [37]

    Cla ssification of covariance matrices using a riemannian-based kernel for bci applica- tions,

    A. Barachant, S. Bonnet, M. Congedo, and C. Jutten, “Cla ssification of covariance matrices using a riemannian-based kernel for bci applica- tions,” Neurocomputing, vol. 112, p. 172–178, 07 2013

    2013

  38. [38]

    Hyperscanning eeg and classifica tion based on riemannian geometry for festive and violent mental state discrimi- nation,

    C. Simar, A.-M. Cebolla, G. Chartier, M. Petieau, G. Bon tempi, A. Berthoz, and G. Cheron, “Hyperscanning eeg and classifica tion based on riemannian geometry for festive and violent mental state discrimi- nation,” Frontiers in Neuroscience , vol. 14, 2020. [Online]. Available: https://www.frontiersin.org/articles/10.3389/fnins.2020.588357

    work page doi:10.3389/fnins.2020.588357 2020

  39. [39]

    A general m odel based on riemannian manifold for stable decoding movement trajec tory from ECoG signals,

    R. Eyvazpour, B. Farrokhi, and A. Erfanian, “A general m odel based on riemannian manifold for stable decoding movement trajec tory from ECoG signals,” iScience, vol. 29, no. 2, p. 114521, 2026, first published online 2025-12-23; issue/eCollection date d 2026-02. [Online]. Available: https://doi.org/10.1016/j.isci.2025.114521

    work page doi:10.1016/j.isci.2025.114521 2026

  40. [40]

    Domain randomization for transferring deep neural networks from simulation to the real world,

    N. Jaquier and S. Calinon, “Gaussian mixture regressio n on symmetric positive definite matrices manifolds: Application to wrist motion estimation with sEMG,” in 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2017, pp. 59–64. [Online]. Available: https://doi.org/10.1109/IROS.2017.8202138

    work page doi:10.1109/iros.2017.8202138 2017

  41. [41]

    C ombining electromyography and tactile myography to improve hand and wrist activity detection in prostheses,

    N. Jaquier, M. Connan, C. Castellini, and S. Calinon, “C ombining electromyography and tactile myography to improve hand and wrist activity detection in prostheses,” Technologies, vol. 5, no. 4, p. 64,

  42. [42]

    Available: https://doi.org/10.3390/tec hnologies5040064

    [Online]. Available: https://doi.org/10.3390/tec hnologies5040064

    work page doi:10.3390/tec

  43. [43]

    Fast and accurate decodi ng of finger movements from ECoG through riemannian features and modern machine learning techniques,

    L. Y ao, B. Zhu, and M. Shoaran, “Fast and accurate decodi ng of finger movements from ECoG through riemannian features and modern machine learning techniques,” Journal of Neural Engineering, vol. 19, no. 1, p. 016037, 2022. [Online]. Available: https://doi.org/10.1088/1741-2552/ac4ed1

    work page doi:10.1088/1741-2552/ac4ed1 2022

  44. [44]

    Emg-finger-kinematics (emg-fk) dataset,

    M. Colot, C. Simar, g. cheron, A. M. CEBOLLA AL V AREZ, and G. Bontempi, “Emg-finger-kinematics (emg-fk) dataset,” Ap r. 2026. [Online]. Available: https://doi.org/10.5281/zenodo.19453843

    work page doi:10.5281/zenodo.19453843 2026

  45. [45]

    Use of a wearable electromyography armband to detect lift-lower tasks and classify hand loads,

    S. Taori and S. Lim, “Use of a wearable electromyography armband to detect lift-lower tasks and classify hand loads,” Applied Ergonomics, vol. 119, p. 104285, 2024, https://doi.org/10.1016/j.ape rgo.2024.104285

    work page doi:10.1016/j.ape 2024

  46. [46]

    Classification algorithms trained on simple (symm etric) lifting data perform poorly in predicting hand loads during complex (free- dynamic) lifting tasks,

    ——, “Classification algorithms trained on simple (symm etric) lifting data perform poorly in predicting hand loads during complex (free- dynamic) lifting tasks,” Applied Ergonomics, vol. 125, p. 104427, 2025, https://doi.org/10.1016/j.apergo.2024.104427

    work page doi:10.1016/j.apergo.2024.104427 2025

  47. [47]

    Leap hand: Low-cost, efficient, and anthropomor- phic hand for robot learning.arXiv preprint arXiv:2309.06440, 2023

    K. Shaw, A. Agarwal, and D. Pathak, “Leap hand: Low-cost , effi- cient, and anthropomorphic hand for robot learning,” arXiv preprint arXiv:2309.06440, 2023, https://doi.org/10.48550/arXiv.2309.06440

    work page doi:10.48550/arxiv.2309.06440 2023

  48. [48]

    The influ - ence of training with visual biofeedback on the predictabil ity of myoelectric control usability,

    J. L. Nawfel, K. B. Englehart, and E. J. Scheme, “The influ - ence of training with visual biofeedback on the predictabil ity of myoelectric control usability,” IEEE Transactions on Neural Sys- tems and Rehabilitation Engineering , vol. 30, pp. 878–892, 2022, https://doi.org/10.1109/TNSRE.2022.3162421

    work page doi:10.1109/tnsre.2022.3162421 2022

  49. [49]

    Concurrent adaptation of human and machine improves simul taneous and proportional myoelectric control,

    J. M. Hahne, S. D¨ ahne, H.-J. Hwang, K.-R. M¨ uller, and L . C. Parra, “Concurrent adaptation of human and machine improves simul taneous and proportional myoelectric control,” IEEE Transactions on Neural Systems and Rehabilitation Engineering , vol. 23, no. 4, pp. 618–627, 2015, https://doi.org/10.1109/TNSRE.2015.2401134

    work page doi:10.1109/tnsre.2015.2401134 2015

  50. [50]

    Synchronization of wearable motion capture and emg measurement systems,

    R. V . Schulte, E. C. Prinsen, L. Schaake, and J. H. Buurke , “Synchronization of wearable motion capture and emg measurement systems,” in 2022 International Conference on Rehabilitation Robotics (ICORR) . IEEE, 2022, pp. 1–6, https://doi.org/10.1109/ICORR55369.2022.9896587

    work page doi:10.1109/icorr55369.2022.9896587 2022

  51. [51]

    Emg-based deco d- ing of manipulation motions in virtual reality: Towards imm er- sive interfaces,

    H. Manjunatha, S. S. Jujjavarapu, and E. T. Esfahani, “C lassifi- cation of motor control difficulty using emg in physical huma n- robot interaction,” in 2020 IEEE International Conference on Sys- tems, Man, and Cybernetics (SMC) . IEEE, 2020, pp. 2708–2713, https://doi.org/10.1109/SMC42975.2020.9283016

    work page doi:10.1109/smc42975.2020.9283016 2020

  52. [52]

    A review of classification algorithms for EEG-based brain–computer interfaces: a 10 year update,

    F. Lotte, L. Bougrain, A. Cichocki, M. Clerc, M. Congedo , A. Rako- tomamonjy, and F. Yger, “A review of classification algorith ms for eeg-based brain–computer interfaces: a 10 year update, ” Jour- nal of neural engineering , vol. 15, no. 3, p. 031005, 2018, https://doi.org/10.1088/1741-2552/aab2f2

    work page doi:10.1088/1741-2552/aab2f2 2018

  53. [53]

    A review of processing meth- ods and classification algorithm for eeg signal,

    Y . Xie and I. L. Oniga, “A review of processing meth- ods and classification algorithm for eeg signal,” 2020, https://doi.org/10.2478/cjece-2020-0004

    work page doi:10.2478/cjece-2020-0004 2020

  54. [54]

    Statistical comparisons of classifiers ov er multiple data sets,

    J. Demˇ sar, “Statistical comparisons of classifiers ov er multiple data sets,” Journal of Machine learning research , vol. 7, no. Jan, pp. 1–30, 2006, https://www.scopus.com/inward/record.uri?eid=2-s2.0-29644438050&partnerID=40&md5=81

    2006

  55. [55]

    Pytorch: An imperative style, high-performance deep learning library ,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. C hanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga et al. , “Pytorch: An imperative style, high-performance deep learning library ,” Advances in neural information processing systems , vol. 32, 2019

    2019

  56. [56]

    TensorFlow: Large-scale machine learning on heterogeneo us systems,

    M. Abadi, A. Agarwal, P . Barham, E. Brevdo, Z. Chen, C. Ci tro, G. S. Corrado, A. Davis, J. Dean, M. Devin, S. Ghemawat, I. Goodfel low, A. Harp, G. Irving, M. Isard, Y . Jia, R. Jozefowicz, L. Kaiser , M. Kudlur, J. Levenberg, D. Man´ e, R. Monga, S. Moore, D. Murr ay, C. Olah, M. Schuster, J. Shlens, B. Steiner, I. Sutskever, K. Talwar, P . Tucker, V ....

    2015

  57. [57]

    pyriema nn,

    A. Barachant, Q. Barth´ elemy, J.-R. King, A. Gramfort, S. Chevallier, P . L. C. Rodrigues, E. Olivetti, V . Goncharenko, G. Wagner vo m Berg, G. Reguig, A. Lebeurrier, E. Bj¨ areholt, M. S. Y amamoto, P . C lisson, M.-C. Corsi, I. Carrara, A. Mellot, B. Junqueira Lopes, B. Ga isford, A. Mian, A. Andreev, G. Cattan, and B. Aristimunha, “pyriema nn,” 1

  58. [58]

    2020 , keywords =

    [Online]. Available: https://doi.org/10.5281/zen odo.593816

    work page doi:10.5281/zen

  59. [59]

    Mne-py thon,

    E. Larson, A. Gramfort, D. A. Engemann, J. Leppakangas, C. Brodbeck, M. Jas, T. L. Brooks, J. Sassenhagen, D. McCloy, M. Luessi, J. -R. King, R. H¨ ochenberger, C. Brunner, R. Goj, G. Favelier, M. v an Vliet, M. Wronkiewicz, S. Appelhoff, A. Rockhill, and ..., “Mne-py thon,” Nov. 2025, https://doi.org/10.5281/zenodo.17675410. 13 Fig. 9. Evaluation of TRR ...

    work page doi:10.5281/zenodo.17675410 2025