Modeling and Individualizing Continuous Joint Kinematics Using Gaussian Process Enhanced Fourier Series

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Current prosthetic controllers use discrete state machines, limiting tasks and requiring extensive user-specific tuning.
• Continuous controllers offer potential for more natural gait by unifying the gait cycle and predicting movement.
• Personalized models are needed to adapt prosthetic control to individual user characteristics.

Purpose of the Study:

• To develop a continuous, personalized gait prediction model for prosthetic controllers.
• To enable prostheses to support a continuum of locomotion tasks and adapt to individual users.
• To reduce the computational burden for real-time prosthetic applications.

Main Methods:

• A Gaussian process enhanced Fourier series (GPEFS) method was proposed to model human locomotion.
• Joint trajectories were transformed into Fourier coefficient space using least squares.
• Gaussian process regression (GPR) models learned the relationship between Fourier coefficients and locomotion parameters (phase, speed, slope).

Main Results:

• The GPEFS method significantly reduced computational load compared to direct GPR fitting.
• A personalized prediction model was successfully built, adapting to individual kinematics across various speeds and slopes.
• The proposed gait prediction and personalized models demonstrated feasibility and effectiveness.

Conclusions:

• The GPEFS method provides an efficient approach for continuous gait prediction in prosthetics.
• Personalized gait prediction models enhance prosthetic adaptability and user experience.
• This method facilitates more normative biomechanics and reduces the need for extensive parameter tuning.