A gait phase prediction model trained on benchmark datasets for evaluating a controller for prosthetic legs

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Powered lower-limb assistive devices enhance mobility for individuals with impairments.
• Accurate gait phase prediction is crucial for device control and assessing gait similarity to intact limbs.

Purpose of the Study:

• To propose and validate a deep neural network (DNN) model for continuous gait phase prediction.
• To evaluate the model's performance on diverse datasets, including those from individuals with transfemoral amputations.

Main Methods:

• Utilized a long short-term memory (LSTM)-based DNN model.
• Trained the model on historical data (250 ms) of vertical load, thigh, knee, and ankle angles.
• Validated the model on benchmark datasets (level-ground walking, stair ascent) and a powered prosthetic leg dataset.

Main Results:

• Achieved a low phase prediction error of 1.28% on benchmark datasets.
• Demonstrated effective performance (5.70% error) on a powered prosthetic leg dataset without post-processing.
• Confirmed the model's utility for evaluating prosthetic leg controller performance.

Conclusions:

• The proposed DNN model enables efficient gait phase prediction for powered lower-limb assistive devices.
• The model facilitates the evaluation of device controllers, assessing the normality of generated gait.
• This approach supports the development of more effective and responsive assistive technologies.