EMG-Based Estimation of Lower Limb Joint Angles and Moments Using Long Short-Term Memory Network

Area of Science:

• Biomechanics
• Machine Learning
• Human Movement Analysis

Background:

• Direct measurement of joint moments during natural human movement is challenging without altering the motion.
• Existing methods often rely on force plates, which have limited coverage and can restrict natural movement.

Purpose of the Study:

• To investigate the use of Long Short-Term Memory (LSTM) networks for predicting lower limb joint kinetics and kinematics.
• To determine if sEMG signals alone can accurately estimate joint moments and angles without force plates post-training.

Main Methods:

• Surface electromyography (sEMG) signals from 14 lower extremity muscles were recorded.
• A 112-dimensional input vector was created using RMS, Mean Absolute Value, and autoregressive model coefficients for each muscle.
• LSTM networks were trained using motion capture and force plate data, with joint kinematics and kinetics as outputs.

Main Results:

• The LSTM model achieved high accuracy in estimating joint angles and moments.
• Average R-squared scores were: knee angle (97.25%), knee moment (94.9%), ankle angle (91.44%), and ankle moment (85.44%).

Conclusions:

• The study demonstrates the feasibility of estimating joint angles and moments using only sEMG signals.
• This approach offers a practical, non-invasive method for human biomechanics analysis, removing the need for force plates and motion capture systems after model training.