pith. sign in

arxiv: 2606.07476 · v1 · pith:7OUZU3NPnew · submitted 2026-06-05 · 📡 eess.SY · cs.RO· cs.SY· eess.SP

Physiologically Constrained Musculoskeletal Neural Network for Multi-DoF Joint Kinematics Estimation from Partially Observed sEMG

Wending Heng , Mingming Zhang , Glen Cooper , Zhenhong Li This is my paper

Pith reviewed 2026-06-27 20:54 UTC · model grok-4.3

classification 📡 eess.SY cs.ROcs.SYeess.SP
keywords musculoskeletal neural networksEMGjoint kinematics estimationmulti-DoFpartially observed signalsmuscle activation inferencephysics-informed neural networkwrist motion
0
0 comments X

The pith

Hybrid network estimates multi-DoF joint angles from partial sEMG while recovering activations for unmeasured muscles without internal force labels.

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

The paper introduces MSK-NN, a neural architecture that pairs a CNN muscle-activation estimator with an embedded musculoskeletal forward-dynamics module. This composite model is trained end-to-end on a composite loss that penalizes kinematics error, enforces data-driven muscle synergies, and respects anatomy-guided activation trends. Because the forward-dynamics block is differentiable, the network learns to map observed surface signals to joint angles and to plausible activations of muscles that were never presented as inputs. On two-DoF wrist tasks the method records lower normalized root-mean-square error and higher R-squared values than CNN, Bi-LSTM, CNN-LSTM, and PET baselines, with the largest gains on random motion; the learned musculoskeletal parameters remain inside physiological bounds and the inferred activation of an excluded muscle tracks its independently recorded sEMG envelope.

Core claim

MSK-NN consists of a CNN-based muscle activation estimator and an embedded MSK forward dynamics module that together form a fully differentiable architecture. The network is trained without any direct supervision on muscle-tendon forces, joint torques, or other internal biomechanical variables by using a composite physics-physiology loss that combines joint kinematics error, a data-driven muscle synergy term, and an anatomy-guided trend term. When evaluated on two-DoF wrist kinematics from three rhythmic motions and one random motion, the model yields lower NRMSE and higher R2 than the listed neural baselines, keeps optimized MSK parameters inside physiological limits, and produces an activa

What carries the argument

The embedded MSK forward dynamics module, which converts estimated muscle activations into predicted joint kinematics inside the differentiable computational graph.

If this is right

  • MSK-NN produces lower NRMSE and higher R2 than CNN, Bi-LSTM, CNN-LSTM, and PET baselines on two-DoF wrist estimation, with the advantage most pronounced during random motion.
  • The optimized musculoskeletal parameters remain inside established physiological ranges after training.
  • The activation inferred for a muscle whose sEMG was withheld from the input still matches the temporal profile of its recorded envelope.
  • The same architecture handles both rhythmic motions with unconstrained speed and amplitude and fully random motion without retraining.

Where Pith is reading between the lines

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

  • The approach could be applied to other joints where anatomical constraints prevent full sEMG coverage, such as the shoulder or ankle.
  • Because no internal force labels are required, the method might be trained on data sets collected with only standard motion-capture and surface electrodes.
  • If the forward-dynamics module generalizes across subjects, the same trained network could support real-time prosthetic or exoskeleton control with fewer sensors.
  • The synergy and trend losses might be replaced or augmented with subject-specific calibration data to further tighten physiological fidelity.

Load-bearing premise

The musculoskeletal forward-dynamics equations supply a sufficiently accurate mapping from activations to kinematics that the composite loss can drive both angle prediction and recovery of unmeasured activations without ever seeing force or torque data.

What would settle it

Record an excluded muscle's sEMG envelope independently and test whether the model's inferred activation time series for that muscle shows low temporal correlation with the envelope while all other performance metrics remain high.

Figures

Figures reproduced from arXiv: 2606.07476 by Glen Cooper, Mingming Zhang, Wending Heng, Zhenhong Li.

Figure 1
Figure 1. Figure 1: The estimator takes pre-processed sEMG from mea [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
read the original abstract

This paper investigates multi-degrees of freedom (DoF) joint kinematics estimation under partially observed surface electromyography (sEMG), where only a subset of task-relevant muscles can be measured due to anatomical inaccessibility or sensor constraints. A novel musculoskeletal neural network (MSK-NN) is proposed to estimate multi-DoF joint angles while simultaneously inferring activations for both measured and unmeasured muscles. MSK-NN consists of a CNN-based muscle activation estimator and an embedded MSK forward dynamics module, forming a fully differentiable architecture. Unlike existing hybrid neural frameworks that require additional biomechanical labels (e.g., muscle-tendon forces, joint torques), MSK-NN is trained without direct supervision of internal biomechanical variables. A composite physics-physiology loss is designed by incorporating a joint kinematics loss, a data-driven muscle synergy loss, and an anatomy-guided trend loss. The proposed method is evaluated on two-DoF wrist kinematics estimation across three rhythmic motions with unconstrained speed and amplitude, and one random motion. Compared with CNN, Bi-LSTM, CNN-LSTM, and PET baselines, MSK-NN achieves lower normalized root mean square error (NRMSE) and higher coefficient of determination (R2), especially for the random motion. More importantly, the optimized MSK parameters remain within physiological limits, and the estimated activation of an input-excluded muscle exhibits strong temporal agreement with its recorded sEMG envelope, demonstrating the capability of musculoskeletal (MSK)-NN to recover physiologically plausible activations.

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

3 major / 2 minor

Summary. The paper proposes MSK-NN, a hybrid architecture combining a CNN-based muscle activation estimator with an embedded differentiable musculoskeletal (MSK) forward dynamics module. It estimates 2-DoF wrist joint kinematics from partially observed sEMG (one muscle excluded) while inferring activations for both measured and unmeasured muscles. Training uses only a composite loss (joint kinematics, data-driven muscle synergy, and anatomy-guided trend) with no direct supervision on internal biomechanical variables such as forces or torques. On rhythmic and random wrist motions, MSK-NN reports lower NRMSE and higher R² than CNN, Bi-LSTM, CNN-LSTM, and PET baselines; optimized MSK parameters remain within physiological bounds, and the excluded-muscle activation estimate shows temporal agreement with its recorded sEMG envelope.

Significance. If the embedded MSK dynamics prove sufficiently accurate and the composite loss yields valid internal inferences, the approach would advance hybrid physics-informed neural modeling for multi-DoF kinematics estimation under sensor constraints, reducing reliance on full biomechanical labels. The differentiable embedding and absence of force/torque supervision are notable strengths relative to prior hybrid frameworks.

major comments (3)
  1. [Abstract / Evaluation] Abstract and evaluation description: quantitative gains (lower NRMSE, higher R² especially on random motion) are reported without error bars, exact train/test splits, cross-validation details, or statistical tests, leaving open whether improvements are robust or sensitive to data partitioning.
  2. [Abstract / Methods (MSK module and loss)] The central claim that unmeasured activations are physiologically recovered rests on the embedded MSK forward dynamics supplying an accurate enough mapping from activations to kinematics. Training uses only kinematics + synergy + trend losses with no terms on forces, torques, or other internal states; no standalone validation, sensitivity analysis, or ablation of the MSK module's accuracy is described, so modeling error could admit multiple solutions that match observed angles while producing incorrect internals.
  3. [Abstract / Loss design] The data-driven muscle synergy loss introduces dependence on observed training patterns, yet the paper asserts that the MSK module supplies external biomechanical grounding sufficient to recover plausible activations for the excluded muscle. No quantitative check (e.g., comparison of recovered vs. measured synergy structure or parameter sensitivity) is provided to show that improvements do not reduce to quantities defined solely by fitted parameters.
minor comments (2)
  1. [Methods] Notation for the composite loss components and the exact form of the anatomy-guided trend loss should be defined explicitly with equation numbers for reproducibility.
  2. [Methods] Implementation details of the MSK forward dynamics module (muscle-tendon parameters, Hill-type model equations, numerical integration scheme) are absent and should be supplied or referenced.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of statistical rigor and validation of the hybrid model. We address each point below and will revise the manuscript accordingly where feasible.

read point-by-point responses

  1. Referee: [Abstract / Evaluation] Abstract and evaluation description: quantitative gains (lower NRMSE, higher R² especially on random motion) are reported without error bars, exact train/test splits, cross-validation details, or statistical tests, leaving open whether improvements are robust or sensitive to data partitioning.

    Authors: We agree that additional statistical details are needed to demonstrate robustness. In the revised version we will report mean and standard deviation of NRMSE and R² across multiple random seeds, explicitly state the train/test split (subject-wise or trial-wise) and any cross-validation scheme, and include paired statistical tests comparing MSK-NN against each baseline. revision: yes

  2. Referee: [Abstract / Methods (MSK module and loss)] The central claim that unmeasured activations are physiologically recovered rests on the embedded MSK forward dynamics supplying an accurate enough mapping from activations to kinematics. Training uses only kinematics + synergy + trend losses with no terms on forces, torques, or other internal states; no standalone validation, sensitivity analysis, or ablation of the MSK module's accuracy is described, so modeling error could admit multiple solutions that match observed angles while producing incorrect internals.

    Authors: The absence of force/torque labels precludes direct supervision or standalone validation of internal states; this is an inherent limitation of the dataset. We will nevertheless add an ablation that removes the MSK module (replacing it with a direct regression head) and a sensitivity study on MSK parameters to quantify their contribution to the observed gains and to the excluded-muscle activation estimates. revision: partial

  3. Referee: [Abstract / Loss design] The data-driven muscle synergy loss introduces dependence on observed training patterns, yet the paper asserts that the MSK module supplies external biomechanical grounding sufficient to recover plausible activations for the excluded muscle. No quantitative check (e.g., comparison of recovered vs. measured synergy structure or parameter sensitivity) is provided to show that improvements do not reduce to quantities defined solely by fitted parameters.

    Authors: We will add a quantitative comparison of the learned synergy matrix against the empirical synergy computed from the measured sEMG channels, together with a parameter-sensitivity analysis that perturbs the MSK module while keeping the synergy loss fixed. These additions will clarify the distinct contribution of the biomechanical constraints. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The MSK-NN embeds an external musculoskeletal forward-dynamics module whose mapping from activations to kinematics is treated as given (not derived within the paper). Training proceeds via a composite loss on observed joint angles plus a data-driven synergy term and an anatomy-guided trend term; the inference of unmeasured activations and the physiological bounds on optimized parameters are direct consequences of this differentiable embedding and the loss, without any step that defines a quantity in terms of itself or renames a fitted input as a prediction. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work appear in the derivation.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim depends on the accuracy of the embedded musculoskeletal forward dynamics and the sufficiency of the composite loss to enforce physiological plausibility. No new entities are postulated.

free parameters (1)
  • MSK model parameters
    Optimized during training subject to physiological limits; their specific values are not reported in the abstract.
axioms (2)
  • domain assumption The musculoskeletal forward dynamics module accurately represents the mapping from muscle activations to joint kinematics for the wrist.
    Invoked by embedding the module directly in the differentiable architecture.
  • domain assumption Muscle synergy patterns extracted from data plus anatomy-guided trends provide adequate constraints to infer activations of unmeasured muscles.
    Used to construct the composite loss terms.

pith-pipeline@v0.9.1-grok · 5823 in / 1446 out tokens · 52458 ms · 2026-06-27T20:54:43.412936+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

45 extracted references · 1 canonical work pages

  1. [1]

    Continuous Motion Intention Prediction Using sEMG for Upper-Limb Rehabilitation: A Systematic Review of Model-Based and Model-Free Approaches,

    Z. Weiet al., “Continuous Motion Intention Prediction Using sEMG for Upper-Limb Rehabilitation: A Systematic Review of Model-Based and Model-Free Approaches,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 32, pp. 1487–1504, 2024

    2024

  2. [2]

    Adaptive Cooperative Control Strategy for a Wrist Exoskeleton Using Model-Based Joint Impedance Estimation,

    Y . Zhaoet al., “Adaptive Cooperative Control Strategy for a Wrist Exoskeleton Using Model-Based Joint Impedance Estimation,” IEEE/ASME Transactions on Mechatronics, vol. 28, no. 2, pp. 748–757, Apr. 2023

    2023

  3. [3]

    Adaptive Compliance Control for a Wearable Lower Limb Rehabilitation Robot Based on Online Estimation of User’s Joint Stiffness,

    Z. Zhenget al., “Adaptive Compliance Control for a Wearable Lower Limb Rehabilitation Robot Based on Online Estimation of User’s Joint Stiffness,”IEEE/ASME Transactions on Mechatronics, vol. 30, no. 6, pp. 5590–5602, Dec. 2025

    2025

  4. [4]

    Surface emg in clinical assessment and neuroreha- bilitation: Barriers limiting its use,

    I. Campaniniet al., “Surface emg in clinical assessment and neuroreha- bilitation: Barriers limiting its use,”Frontiers in Neurology, vol. 11, p. 934, Sep. 2020

    2020

  5. [5]

    An electromyography-assisted musculoskeletal simulation with concurrent optimization of muscle excitations and knee joint kinematics,

    A. Esrafilianet al., “An electromyography-assisted musculoskeletal simulation with concurrent optimization of muscle excitations and knee joint kinematics,”Journal of Biomechanical Engineering, vol. 147, no. 9, p. 091002, 2025

    2025

  6. [6]

    An efficient framework for personalizing emg- driven musculoskeletal models based on reinforcement learning,

    J. Bermanet al., “An efficient framework for personalizing emg- driven musculoskeletal models based on reinforcement learning,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 32, pp. 4174–4185, 2024

    2024

  7. [7]

    Emg-driven control in lower limb prostheses: a topic-based systematic review,

    A. Cimolatoet al., “Emg-driven control in lower limb prostheses: a topic-based systematic review,”Journal of NeuroEngineering and Rehabilitation, vol. 19, p. 43, 2022

    2022

  8. [8]

    Bio-robotics research for non-invasive myoelectric neural interfaces for upper-limb prosthetic control: A 10-year perspective review,

    N. Jianget al., “Bio-robotics research for non-invasive myoelectric neural interfaces for upper-limb prosthetic control: A 10-year perspective review,”National Science Review, vol. 10, no. 5, p. nwad048, Apr. 2023

    2023

  9. [9]

    Simultaneously encoding movement and semg-based stiffness for robotic skill learning,

    C. Zenget al., “Simultaneously encoding movement and semg-based stiffness for robotic skill learning,”IEEE Transactions on Industrial Informatics, vol. 17, no. 9, pp. 6248–6257, 2021

    2021

  10. [10]

    An electromyography signals-based human-robot collaboration method for human skill learning and imitation,

    T. Zhanget al., “An electromyography signals-based human-robot collaboration method for human skill learning and imitation,”Journal of Manufacturing Systems, vol. 64, pp. 330–343, Jul. 2022

    2022

  11. [11]

    A Bi-Directional LSTM Network for Estimating Contin- uous Upper Limb Movement From Surface Electromyography,

    C. Maet al., “A Bi-Directional LSTM Network for Estimating Contin- uous Upper Limb Movement From Surface Electromyography,”IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 7217–7224, Oct. 2021

    2021

  12. [12]

    A CNN-LSTM Hybrid Model for Wrist Kinematics Estimation Using Surface Electromyography,

    T. Baoet al., “A CNN-LSTM Hybrid Model for Wrist Kinematics Estimation Using Surface Electromyography,”IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–9, 2021

    2021

  13. [13]

    A parallel and efficient transformer deep learning network for continuous estimation of hand kinematics from electromyographic signals,

    C. Linet al., “A parallel and efficient transformer deep learning network for continuous estimation of hand kinematics from electromyographic signals,”Scientific Reports, vol. 15, no. 1, p. 36150, Oct. 2025

    2025

  14. [14]

    Deep learning for emg-based human-machine interaction: A review,

    D. Z. Xionget al., “Deep learning for emg-based human-machine interaction: A review,”IEEE/CAA Journal of Automatica Sinica, vol. 8, no. 3, pp. 512–533, Mar 2021

    2021

  15. [15]

    Neuromusculoskeletal Modeling: Estimation of Muscle Forces and Joint Moments and Movements from Measurements of Neural Command,

    T. S. Buchananet al., “Neuromusculoskeletal Modeling: Estimation of Muscle Forces and Joint Moments and Movements from Measurements of Neural Command,”Journal of Applied Biomechanics, vol. 20, no. 4, pp. 367–395, Nov. 2004

    2004

  16. [16]

    Computationally Efficient Personalized EMG-Driven Musculoskeletal Model of Wrist Joint,

    Y . Zhaoet al., “Computationally Efficient Personalized EMG-Driven Musculoskeletal Model of Wrist Joint,”IEEE Transactions on Instru- mentation and Measurement, vol. 72, pp. 1–10, 2023

    2023

  17. [17]

    Physics-Informed Deep Learning for Musculoskeletal Modeling: Predicting Muscle Forces and Joint Kinematics From Sur- face EMG,

    J. Zhanget al., “Physics-Informed Deep Learning for Musculoskeletal Modeling: Predicting Muscle Forces and Joint Kinematics From Sur- face EMG,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 31, pp. 484–493, 2023

    2023

  18. [18]

    A physics-informed low-shot adversarial learning for semg-based estimation of muscle force and joint kinematics,

    Y . Shiet al., “A physics-informed low-shot adversarial learning for semg-based estimation of muscle force and joint kinematics,”IEEE Journal of Biomedical and Health Informatics, vol. 28, no. 3, pp. 1309– 1320, Mar. 2024

    2024

  19. [19]

    Predicting multi-joint kinematics of the upper limb from emg signals across varied loads with a physics-informed neural network,

    R. Kumaret al., “Predicting multi-joint kinematics of the upper limb from emg signals across varied loads with a physics-informed neural network,”arXiv preprint arXiv:2312.09418, 2023

    work page arXiv 2023

  20. [20]

    Boosting Personalized Musculoskeletal Modeling With Physics-Informed Knowledge Transfer,

    J. Zhanget al., “Boosting Personalized Musculoskeletal Modeling With Physics-Informed Knowledge Transfer,”IEEE Transactions on Instrumentation and Measurement, vol. 72, pp. 1–11, 2023

    2023

  21. [21]

    Physics-embedded neural networks for semg-based con- tinuous motion estimation,

    W. Henget al., “Physics-embedded neural networks for semg-based con- tinuous motion estimation,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025, pp. 16 538–16 543

    2025

  22. [22]

    Matrix factorization algorithms for the identification of muscle synergies: Evaluation on simulated and experimental data sets,

    M. C. Treschet al., “Matrix factorization algorithms for the identification of muscle synergies: Evaluation on simulated and experimental data sets,”Journal of Neurophysiology, vol. 95, no. 4, pp. 2199–2212, 2006

    2006

  23. [23]

    Can measured synergy excitations accurately construct unmeasured muscle excitations?

    N. A. Biancoet al., “Can measured synergy excitations accurately construct unmeasured muscle excitations?”Journal of Biomechanical Engineering, vol. 140, no. 1, p. 011011, 2018

    2018

  24. [24]

    EMG-driven musculoskeletal model calibration with estimation of unmeasured muscle excitations via synergy extrapolation,

    D. Aoet al., “EMG-driven musculoskeletal model calibration with estimation of unmeasured muscle excitations via synergy extrapolation,” Frontiers in Bioengineering and Biotechnology, vol. 10, p. 962959, Sep. 2022

    2022

  25. [25]

    Comparison of synergy extrapolation and static optimization for estimating multiple unmeasured muscle activations dur- ing walking,

    D. Ao and B. J. Fregly, “Comparison of synergy extrapolation and static optimization for estimating multiple unmeasured muscle activations dur- ing walking,”Journal of NeuroEngineering and Rehabilitation, vol. 21, no. 1, p. 194, Nov. 2024

    2024

  26. [26]

    Muscle synergy-informed neuromusculoskeletal modelling to estimate knee contact forces in children with cerebral palsy,

    M. F. Rabbiet al., “Muscle synergy-informed neuromusculoskeletal modelling to estimate knee contact forces in children with cerebral palsy,”Biomechanics and Modeling in Mechanobiology, vol. 23, pp. 1077–1090, 2024

    2024

  27. [27]

    Adjustment of Muscle Mechanics Model Parameters to Simulate Dynamic Contractions in Older Adults,

    D. G. Thelen, “Adjustment of Muscle Mechanics Model Parameters to Simulate Dynamic Contractions in Older Adults,”Journal of Biome- chanical Engineering, vol. 125, no. 1, pp. 70–77, Feb. 2003

    2003

  28. [28]

    An EMG-driven musculoskeletal model for estimation of wrist kinematics using mirrored bilateral movement,

    Y . Zhaoet al., “An EMG-driven musculoskeletal model for estimation of wrist kinematics using mirrored bilateral movement,”Biomedical Signal Processing and Control, vol. 81, p. 104480, Mar. 2023

    2023

  29. [29]

    An EMG-driven musculoskeletal model to estimate muscle forces and knee joint moments in vivo,

    D. G. Lloyd and T. F. Besier, “An EMG-driven musculoskeletal model to estimate muscle forces and knee joint moments in vivo,”Journal of Biomechanics, vol. 36, no. 6, pp. 765–776, Jun. 2003

    2003

  30. [30]

    Muscle and tendon: Properties, models, scaling, and application to biomechanics and motor control,

    F. E. Zajac, “Muscle and tendon: Properties, models, scaling, and application to biomechanics and motor control,”Critical Reviews in Biomedical Engineering, vol. 17, no. 4, pp. 359–411, 1989

    1989

  31. [31]

    OpenSim: Open-Source Software to Create and Analyze Dynamic Simulations of Movement,

    S. L. Delpet al., “OpenSim: Open-Source Software to Create and Analyze Dynamic Simulations of Movement,”IEEE Transactions on Biomedical Engineering, vol. 54, no. 11, pp. 1940–1950, Nov. 2007

    1940

  32. [32]

    An EMG-Driven Musculoskeletal Model for Estimating Continuous Wrist Motion,

    Y . Zhaoet al., “An EMG-Driven Musculoskeletal Model for Estimating Continuous Wrist Motion,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 28, no. 12, pp. 3113–3120, Dec. 2020

    2020

  33. [33]

    Anatomy, Biomechanics, and Loads of the Wrist Joint,

    J. Eschweileret al., “Anatomy, Biomechanics, and Loads of the Wrist Joint,”Life, vol. 12, no. 2, p. 188, Jan. 2022

    2022

  34. [34]

    The importance of abductor pollicis longus in wrist motions: A physiological wrist simulator study,

    D. S. Shahet al., “The importance of abductor pollicis longus in wrist motions: A physiological wrist simulator study,”Journal of Biomechan- ics, vol. 77, pp. 218–222, Aug. 2018

    2018

  35. [35]

    How muscle architecture and moment arms affect wrist flexion-extension moments,

    R. V . Gonzalezet al., “How muscle architecture and moment arms affect wrist flexion-extension moments,”Journal of Biomechanics, vol. 30, no. 7, pp. 705–712, Jul. 1997

    1997

  36. [36]

    Dynamics of wrist rotations,

    S. K. Charles and N. Hogan, “Dynamics of wrist rotations,”Journal of Biomechanics, vol. 44, no. 4, pp. 614–621, 2011

    2011

  37. [37]

    In VivoEstimation of Human Forearm and Wrist Dynamic Properties,

    K. Parket al., “In VivoEstimation of Human Forearm and Wrist Dynamic Properties,”IEEE Transactions on Neural Systems and Re- habilitation Engineering, vol. 25, no. 5, pp. 436–446, May 2017

    2017

  38. [38]

    Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters,

    P. De Leva, “Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters,”Journal of Biomechanics, vol. 29, no. 9, pp. 1223–1230, Sep. 1996

    1996

  39. [39]

    Standards for surface electromyog- raphy: The european project surface emg for non-invasive assessment of muscles (seniam),

    D. F. Stegeman and H. J. Hermens, “Standards for surface electromyog- raphy: The european project surface emg for non-invasive assessment of muscles (seniam),”Roessingh Research and Development, pp. 12–108, 2007

    2007

  40. [40]

    Decoupled weight decay regularization,

    I. Loschilov and F. Hutter, “Decoupled weight decay regularization,” International Conference on Learning Representations (ICLR), 2019. 11

    2019

  41. [41]

    Interpretation of the correlation coefficient: A basic review,

    R. Taylor, “Interpretation of the correlation coefficient: A basic review,” Journal of Diagnostic Medical Sonography, vol. 6, no. 1, pp. 35–39, 1990

    1990

  42. [42]

    Benchmarking of dynamic simulation predictions in two software platforms using an upper limb musculoskeletal model,

    K. R. Saulet al., “Benchmarking of dynamic simulation predictions in two software platforms using an upper limb musculoskeletal model,” Computer Methods in Biomechanics and Biomedical Engineering, vol. 18, no. 13, pp. 1445–1458, October 2015

    2015

  43. [43]

    A novel tcn-lstm hybrid model for semg-based continuous estimation of wrist joint angles,

    J. Duet al., “A novel tcn-lstm hybrid model for semg-based continuous estimation of wrist joint angles,”Sensors, vol. 24, no. 5631, pp. 1–17, 2024

    2024

  44. [44]

    Dual Transformer Network for Predicting Joint Angles and Torques From Multi-Channel EMG Signals in the Lower Limbs,

    Z. Wanget al., “Dual Transformer Network for Predicting Joint Angles and Torques From Multi-Channel EMG Signals in the Lower Limbs,” IEEE Journal of Biomedical and Health Informatics, pp. 1–13, 2025

    2025

  45. [45]

    Assessing stroke-induced abnormal muscle coactivation in the upper limb using the surface emg co-contraction index: A sys- tematic review,

    Y . Wanget al., “Assessing stroke-induced abnormal muscle coactivation in the upper limb using the surface emg co-contraction index: A sys- tematic review,”Journal of Electromyography and Kinesiology, vol. 81, p. 102985, 2025

    2025