Online sparse Gaussian process based human motion intent learning for an electrically actuated lower extremity...

Area of Science:

• Robotics
• Human-Robot Interaction
• Machine Learning

Background:

• Accurate inference of human motion intent (HMI) is critical for effective lower extremity exoskeleton control and human-robot collaboration.
• The complex, nonlinear relationship between human-robot interaction (HRI) data and HMI is challenging to model using traditional mathematical methods.
• Gaussian Process (GP) regression offers a powerful approach for nonlinear regression but faces computational challenges with large datasets.

Purpose of the Study:

• To develop and validate an efficient online sparse GP algorithm for real-time HMI inference in lower extremity exoskeletons.
• To address the computational complexity of GP regression for large HRI datasets.
• To enable seamless human-exoskeleton collaboration by accurately predicting user's intended movements.

Main Methods:

• An online sparse Gaussian Process (GP) algorithm was developed to learn HMI.
• A dataset comprising physical HRI (torque sensor data from knee joints) and joint angular positions was collected from users wearing an exoskeleton with friction compensation.
• Grey relational analysis (GRA) was employed to reduce dataset complexity and optimize GP hyper-parameters for online regression.

Main Results:

• The proposed algorithm successfully infers human motion intent (HMI) in real time.
• Experimental validation demonstrated the algorithm's effectiveness in capturing user's intended joint movements.
• The method addresses the computational limitations of standard GP regression for large HRI datasets.

Conclusions:

• The developed online sparse GP algorithm provides an effective solution for real-time HMI inference in lower extremity exoskeletons.
• The integration of GRA significantly enhances the computational efficiency of GP regression for this application.
• This approach holds potential for broader application in similar exoskeleton systems, improving human-robot collaboration.