Enhanced Upper Limb Exoskeleton Control for Stroke Rehabilitation Using Combined Electromyography and Force Sensors

Most exoskeletons for upper limbs used in stroke rehabilitation rely solely on muscle activity signals, which are often weak and unstable. This study aimed to develop and improve the control method for these exoskeletons. Two Myo Armband sensors and six force sensors were evaluated to collect electromyography (EMG) signals and reaction forces from the forearm and upper arm regions of ten healthy participants. The data was used to control an external exoskeleton for upper limb rehabilitation, including basic hand and elbow movements. Three methodological techniques were adopted for classifying and analyzing electromyography (EMG) data: K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Linear Discriminant Analysis (LDA). Offline analysis revealed that SVM exhibited the highest accuracy, achieving an average of 95.209% for hand motions utilizing only Myo Armband data and 99.406% for elbow movements among the ten participants. A new method that uses an SVM classifier to combine EMG and force data got 99.1% accuracy. This study demonstrates the superiority of the SVM algorithm in classifying movements and enhancing control of upper limb exoskeletons.