Automated gait event detection for exoskeleton-assisted walking using a long short-term memory model with ground...

Area of Science:

• Biomechanics
• Robotics
• Machine Learning

Background:

• Traditional gait event detection methods struggle with altered walking patterns caused by assistive devices like exoskeletons.
• Existing neural network approaches also face limitations in these scenarios.

Purpose of the Study:

• To develop and validate a novel LSTM-based model for precise gait event detection (heel strike, toe-off) in individuals using ankle exoskeletons.
• To assess the model's performance, robustness, and training efficiency compared to existing algorithms.

Main Methods:

• A hybrid LSTM model was created by combining ground reaction force (GRF) data and heel marker height data.
• Gait events were identified through peak detection on the model's output.
• The model's training requirements were evaluated using a low training-to-testing data ratio (1:79).

Main Results:

• The developed LSTM model achieved over 98% accuracy in detecting gait events within 16ms of manually identified events.
• This performance surpasses the detection rates of previous LSTM algorithms (65%-98%).
• The model demonstrated significantly reduced training data requirements compared to other machine learning algorithms.

Conclusions:

• The proposed LSTM model offers a robust and efficient solution for automated gait event detection.
• It is effective for both exoskeleton-assisted and unassisted walking in healthy subjects.
• The model's low training needs make it a practical tool for real-world applications.