Presentation description

The long-term goal of this research is to restore upper-limb function to stroke patients with chronic hemiparesis through an assistive exoskeleton.
Stroke is a leading cause of long-term disability worldwide, urging continual advancements in rehabilitation and assistive techniques. Robotic exoskeletons with electromyographic (EMG) control have emerged as promising assistive devices that use residual muscle activity to intuitively control the exoskeleton and improve patient dexterity. However, current commercially available exoskeletons only assist in the hand and/or elbow and do little to help wrist movements.
In this research, we investigated classification control algorithms to allow stroke patients to control wrist movement rotation in addition to hand movements using EMG signals. We trained classifiers on EMG corresponding to five classes: hand grasp/open, wrist pronate/supinate, and rest. In initial offline studies, our KNN model, trained on six recorded datasets, demonstrated higher accuracies compared to the CNN model (95.90 ± 3.17% and 89.93 ± 5.05%, respectively; p < 0.01 paired t-test). The KNN model also allowed for faster prediction times (seconds) compared to the CNN model (0.0032 ± 0.0001 and 0.0066 ± 0.0004 seconds, respectively; p < 0.0001, paired t-test).
The results of the offline model performance analyses suggest that adding wrist control to EMG exoskeletons is feasible. The performance of the improved exoskeleton could be further investigated through online trials that include clinical measures of hand dexterity. Future work could entail adapting the underlying technology and methodologies to implement other movements in multiple degrees of freedom such as wrist flexion. Furthermore, more dexterous exoskeletons can be adapted for robot-assisted therapy and for populations suffering from other forms of neurological and muscular motor impairments.