Gait Phase Classification and Assist Torque Prediction for a Lower Limb Exoskeleton System Using Kernel Recursive...

Area of Science:

• Robotics
• Biomechanical Engineering
• Machine Learning

Background:

• Exoskeleton robot control relies on accurate gait phase classification.
• Individual gait variations present challenges for consistent control.
• Existing methods struggle to adapt to unique human-robot coupling gait features.

Purpose of the Study:

• To enhance gait phase classification in exoskeleton robots using only hip and knee joint angles.
• To develop an adaptive assist torque predictor for personalized exoskeleton control.
• To evaluate the performance of the Kernel Recursive Least Squares (KRLS) algorithm against established methods.

Main Methods:

• A Kernel Recursive Least Squares (KRLS) algorithm was employed to build gait phase classification and assist torque prediction models.
• Gait data from 10 healthy volunteers interacting with a custom exoskeleton robot were collected.
• The KRLS model's performance was compared with Multi-Layer Perceptron Neural Network (MLPNN) and Support Vector Machine (SVM) algorithms.

Main Results:

• The KRLS model achieved an average testing accuracy of 86% for gait phase classification.
• KRLS demonstrated a 3% higher classification accuracy compared to MLPNN and SVM.
• KRLS-based assist torque prediction was twice as effective as MLPNN, showing robustness and generalization.

Conclusions:

• KRLS offers a superior approach for gait phase classification and assist torque prediction in exoskeleton robots.
• The KRLS algorithm effectively adapts to unique gait features in human-robot interactions.
• KRLS provides stable, robust, and generalizable performance for enhanced exoskeleton control.