Strategies for calibration and training to individualize signal generation in nnyoelectric control of assistive devices

摘要

Bioelectric signals are often used for control purposes in rehabilitation engineering. As an example, the antagonistic myoelectric control is the de-facto standard for the control of hand prostheses. Lately, examinations have started to control wheelchairs via the EMG of two earmuscles. Typically, the generated signals show individual amplitudes and unintended coactivations hindering a direct interpretation of the user intention. This article discusses effects and influencing factors affecting the quality of the myoelectric signal and provides a signal processing pipeline to improve the estimation of the user intention. Standardized training paradigms are introduced to individually adapt parameters. Regression models help in minimizing the effects of co-activations. The functionality of the approach is shown using simulated and real-world data of two-channel EMG-measurements of forearm and ear.