IMU-Based Real-Time Estimation of Gait Phase Using Multi-Resolution Neural Networks.

Area of Science:

• Biomechanics
• Robotics
• Machine Learning

Background:

• Accurate gait phase estimation is crucial for analyzing human locomotion and developing assistive technologies.
• Current methods may lack robustness across varied walking conditions and individual gait patterns.

Purpose of the Study:

• To develop and validate a real-time gait phase estimator using wearable inertial measurement units (IMUs).
• To assess the model's generalizability across different participants, walking speeds, and abnormal gait patterns.

Main Methods:

• A multi-rate convolutional neural network (CNN) was trained using data from thigh- and shank-mounted IMUs.
• The model was evaluated using one-subject-out cross-validation and tested on various walking speeds and conditions, including asymmetric walking and stop-start scenarios.
• Statistical tests (e.g., Kolmogorov-Smirnov) were employed to confirm performance robustness.

Main Results:

• The gait phase estimator achieved a spatial root mean square error of 5.00±1.65% and a temporal mean absolute error of 2.78±0.97% at heel strike.
• Cross-validation demonstrated no significant performance degradation when excluding specific walking conditions or testing on new participants.
• No significant error increase was observed for abnormal walking conditions not included in the training set.

Conclusions:

• The proposed IMU-based gait phase estimator exhibits strong generalizability across diverse participants and walking scenarios.
• This technology holds potential for clinical gait analysis, particularly for patient populations with pathological gaits.
• The findings support the use of this estimator in advancing robot-assisted walking and rehabilitation.