Uncertainty-Aware Probabilistic Fusion Post-Processing for Continuous Wrist Motion Estimation in Myoelectric Control.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Engineering

Background:

• Continuous wrist angle estimation using surface electromyography (sEMG) faces challenges due to signal variability and prediction instability.
• Existing regression models often produce temporally fluctuating and non-robust predictions because of the non-stationary nature of sEMG signals.

Purpose of the Study:

• To develop an uncertainty-aware probabilistic fusion post-processing framework for enhanced continuous wrist motion estimation.
• To decouple regression and uncertainty modeling for improved compatibility with various regression models.

Main Methods:

• Implemented a local Gaussian Process Regression (LGPR) model to estimate predictive uncertainty from sliding feature windows.
• Developed a Bayesian-inspired Gaussian formulation for fusing instantaneous regression outputs with LGPR predictions.
• Introduced a closed-form adaptive gain to dynamically adjust smoothing based on predictive variance.

Main Results:

• The proposed framework demonstrated superior performance in key indicators like task completion time, trajectory smoothness, and trajectory tracking error.
• Experimental results validated the method in both open-loop wrist joint motion estimation and closed-loop myoelectric control tasks.
• The uncertainty-aware approach significantly improved prediction robustness compared to existing methods.

Conclusions:

• The uncertainty-aware probabilistic fusion framework offers a robust solution for continuous wrist motion estimation from sEMG.
• This method enhances the reliability and performance of myoelectric control systems.
• The plug-in nature of the framework allows for broad applicability with different regression models.