A Neural Network Estimation of Ankle Torques From Electromyography and Accelerometry

Area of Science:

• Biomechanics and Robotics
• Machine Learning in Healthcare

Background:

• Accurate human joint torque estimation is crucial for clinical applications, therapy planning, and wearable robotic device design.
• Predicting future joint torques is essential for developing anticipatory robot control systems.

Purpose of the Study:

• To develop and evaluate a method for mapping joint torque estimates and predictions from motion capture data to wearable sensor data.
• To compare the performance of various neural network architectures, including Long Short-Term Memory (LSTM) networks, for this mapping task.

Main Methods:

• Utilized dense feedforward, convolutional, neural ordinary differential equation, and Long Short-Term Memory (LSTM) neural networks.
• Trained models to learn the mapping for ankle plantarflexion and dorsiflexion torque during various activities (standing, walking, running, sprinting).
• Investigated both single-point torque estimation and prediction of future torque sequences.

Main Results:

• Long Short-Term Memory (LSTM) neural networks demonstrated superior performance compared to dense feedforward, neural ordinary differential equation, and convolutional neural networks.
• Predictions of future ankle torques up to 0.4 seconds ahead showed high positive correlations with actual measured torques.
• The developed method enables torque estimation using only wearable sensors after initial model training with motion capture data.

Conclusions:

• Sequential data processing with LSTM networks is highly effective for human joint torque estimation and prediction.
• The proposed method offers a practical solution for non-invasive, real-time joint torque assessment using wearable technology.
• This approach has significant potential for advancing personalized rehabilitation and human-robot interaction.