Human Locomotion Implicit Modeling-Based Real-Time Gait Phase Estimation.

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Inertial measurement unit (IMU) signals are crucial for gait phase estimation in exoskeleton adaptation.
• Current methods struggle with accuracy and robustness, especially during terrain changes.

Purpose of the Study:

• To develop a robust gait phase estimation neural network for adaptive exoskeletons.
• To enhance model generalization and performance across varied locomotion scenarios.

Main Methods:

• A neural network combining temporal convolution and transformer layers for feature extraction and information fusion.
• A channel-wise masked reconstruction pre-training strategy using gait phase state vectors and IMU signals.
• Implicit modeling of human locomotion for enhanced generalization.

Main Results:

• The proposed method outperforms baseline approaches in gait phase estimation accuracy on varied terrains and during transitions.
• Achieved low Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) for gait phase and rate estimation.
• Hardware validation confirmed reliable gait cycle and event identification in a hip exoskeleton.

Conclusions:

• The developed neural network offers robust and accurate gait phase estimation for adaptive exoskeleton systems.
• This approach enhances exoskeleton adaptability in dynamic and varied environments.
• Paves the way for more sophisticated real-time gait assistance technologies.