Motor Learning-Based Real-Time Control for Dexterous Manipulation of Prosthetic Hands

Recent studies on myoelectric-based prosthetic control have shown that surface electromyography (sEMG) can enhance prosthetic intuitiveness by improving motion detection algorithms and continuous data processing. This study aims to use a combination of feature extraction techniques and machine learning approaches to map sEMG signals to 10 upper-limb motions for real-time control. The study implements four machine learning methods (i.e., k-nearest neighbours (k-NN), artificial neural networks (ANN), support vector machines (SVM), linear discriminant analysis (LDA)) as classifiers and six time-domain features (i.e., root mean square (RMS), integrated absolute value (IAV), mean absolute value (MAV), simple square integration (SSI), waveform length (WL), average amplitude change (AAC)) to extract sEMG features to differentiate six individual fingers and four-hand griping patterns. Five subjects volunteered in the research and training datasets were recorded using seven sEMG electrodes for three static and three dynamic arm positions. The modalities were assessed with offline classification performance from the collected datasets and real-time evaluation metrics such as motion completion rate, motion detection accuracies and reach and grasp experiments. Based on the above, the control methodology differentiates independent finger motions with high accuracy, 94% completion rates with 0.23 s data processing and prediction time.