Dependent-Gaussian-Process-Based Learning of Joint Torques Using Wearable Smart Shoes for Exoskeleton.

Area of Science:

• Biomechanics
• Robotics
• Machine Learning

Background:

• Estimating lower limb joint torques is crucial for developing advanced lower-limb exoskeleton controllers.
• Current methods often require complex sensor setups or are intrusive during human gait analysis.

Purpose of the Study:

• To present a novel dependent Gaussian process (DGP)-based learning algorithm for accurate joint-torque estimations.
• To utilize data from wearable smart shoes for nonintrusive and low-cost gait analysis.

Main Methods:

• Developed a dependent Gaussian process (DGP) model for data fusion and exploring correlations between joint kinematics and torques.
• Designed dynamic specific composite kernel functions to model multi-scale features and temporal gait variations.
• Employed joint kinematics during training, allowing predictions using only smart shoe data during the prediction phase.

Main Results:

• The DGP model demonstrated accurate joint-torque estimations, even under trained and untrained speed levels across five subjects.
• DGP models showed significant improvements over traditional Gaussian process (GP) models, with higher r² values.
• The smart shoe-based approach proved to be low-cost, nonintrusive, and comfortable for wearers during outdoor activities.

Conclusions:

• The proposed DGP algorithm offers a flexible and accurate solution for time-varying gait-pattern learning and joint-torque estimation.
• This approach enables effective control strategies for lower-limb exoskeletons using readily available wearable sensor data.
• The findings highlight the potential of smart shoes and advanced machine learning for enhancing human-robot interaction in gait rehabilitation and assistance.