Comparison of machine learning and deep learning-based methods for locomotion mode recognition using a single...

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Locomotion mode recognition is crucial for advanced prosthetic control, enabling switches between different walking patterns.
• Traditional machine learning methods struggle with complex data, leading to inaccuracies in motion recognition for powered prostheses.

Purpose of the Study:

• To evaluate deep learning models for locomotion mode recognition in lower-limb prostheses.
• To compare the performance of Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) against a Random Forest Classifier (RFC).

Main Methods:

• Trained and compared RNN, LSTM, and CNN models using data from an inertial measurement unit (IMU).
• Data collected from four subjects performing level walking, standing, stair ascent, and stair descent.
• Evaluated model performance against a Random Forest Classifier (RFC).

Main Results:

• Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) models demonstrated superior performance compared to other evaluated models.
• Deep learning approaches, particularly CNN and LSTM, showed significant potential for accurate locomotion mode recognition.

Conclusions:

• CNN and LSTM models are highly effective for locomotion mode recognition in prosthetic applications.
• These deep learning models offer a promising solution for real-time control of robotic prostheses, enhancing functionality and user experience.