Self-Tuning Extended Kalman Filter Parameters to Identify Ankle's Third-Order Mechanics

Area of Science:

• Biomechanics
• Control Systems Engineering
• Computational Neuroscience

Background:

• Accurate estimation of human ankle mechanical impedance is crucial for modeling human balance control.
• Existing methods using state-augmented extended Kalman filters (AEKF) are sensitive to initial noise covariance parameters.
• A time-consuming trial-and-error approach is often required for tuning these noise parameters.

Purpose of the Study:

• To implement a parameter-estimation approach using an AEKF for inferring ankle mechanical impedance during quiet standing.
• To develop and validate a genetic algorithm (GA) for optimizing the measurement noise (Rk) and process noise covariances (Q) of the EKF.
• To improve the performance and reduce the tuning time for the EKF-based impedance estimation.

Main Methods:

• Implementation of a state-augmented extended Kalman filter (AEKF) for ankle impedance estimation.
• Application of a genetic algorithm (GA) to tune the measurement noise covariance (Rk) and process noise covariance (Q) matrices of the EKF.
• Validation using simulated data from a third-order biomechanical model and experimental data from human subjects.

Main Results:

• The proposed GA effectively tunes the noise covariance matrices (Rk and Q) for the EKF, improving parameter estimation.
• Simulated data demonstrated the efficacy of the GA-tuned EKF for parameter-estimation of a third-order biomechanical model.
• Experimental validation confirmed the algorithm's performance and suggested age as a factor influencing human balance.

Conclusions:

• The GA-based tuning of EKF noise covariances provides an efficient and effective method for estimating human ankle mechanical impedance.
• Optimized parameter estimation enhances the accuracy of biomechanical models for human balance.
• The findings highlight the potential of this approach for understanding age-related changes in balance control.