sEMG-Driven Physics-Informed Gated Recurrent Networks for Modeling Upper Limb Multi-Joint Movement Dynamics

TL;DR

This paper tackles the challenge of estimating multi-joint upper-limb dynamics from time-series sEMG for exoskeleton and rehabilitation HMIs. It proposes PiGRN, a physics-informed GRU that fuses sEMG-driven kinematic predictions with inverse dynamics constraints to predict joint angles, velocities, accelerations, external loads, and torques. Empirical results across five subjects and varying loads demonstrate RMSE from 4.02% to 11.40% and correlations from 0.87 to 0.98, with superior performance over baselines and good generalization to longer sequences. The approach reduces reliance on extensive differentiations and enhances real-time applicability, with future work expanding DOFs, datasets, musculoskeletal models, and unsupervised learning approaches.

Abstract

Exoskeletons and rehabilitation systems have the potential to improve human strength and recovery by using adaptive human-machine interfaces. Achieving precise and responsive control in these systems depends on accurately estimating joint movement dynamics, such as joint angle, velocity, acceleration, external mass, and torque. While machine learning (ML) approaches have been employed to predict joint kinematics from surface electromyography (sEMG) data, traditional ML models often struggle to generalize across dynamic movements. In contrast, physics-informed neural networks integrate biomechanical principles, but their effectiveness in predicting full movement dynamics has not been thoroughly explored. To address this, we introduce the Physics-informed Gated Recurrent Network (PiGRN), a novel model designed to predict multi-joint movement dynamics from sEMG data. PiGRN uses a Gated Recurrent Unit (GRU) to process time-series sEMG inputs, estimate multi-joint kinematics and external loads, and predict joint torque while incorporating physics-based constraints during training. Experimental validation, using sEMG data from five participants performing elbow flexion-extension tasks with 0 kg, 2 kg, and 4 kg loads, showed that PiGRN accurately predicted joint torques for 10 novel movements. RMSE values ranged from 4.02\% to 11.40\%, with correlation coefficients between 0.87 and 0.98. These results underscore PiGRN's potential for real-time applications in exoskeletons and rehabilitation. Future work will focus on expanding datasets, improving musculoskeletal models, and investigating unsupervised learning approaches.

Figures (12)

• Figure 1: The GRU framework
• Figure 2: The recursive model of GRU at time instance $t$
• Figure 3: Physics Informed Gated Recurrent Network Architecture (PiGRN), here, $N_T$ denotes length of time series data, $t$ denotes time instance. $N_C$ and $N_{dof}$ denotes number of channels of sEMG signal and total number of degrees of freedom in the MSK system , $h_{i-1}$ and $GRU_t$ denotes hidden state and GRU block at at time instance $t$. $\lambda_physics$ is physics loss weighting factor which is a hyperparameter. Losses are based on MSE i.e. Mean square error.
• Figure 4: Joint torque estimation flow chart once the PiGRN model is trained
• Figure 5: Loss curve during training