Enhancing Control Strategies for Hand Exoskeletons Through Modeling and Electromyogram-based Control

Functional losses following hand impairments necessitate innovative solutions, leading to the development of hand exoskeletons. This study addresses the challenge of aligning exoskeleton hand control with users' motion intentions by incorporating electromyography (EMG) signals from forearm muscles. The focus is on enhancing control strategies through EMG-based modeling within a virtual environment. This approach allows for cost-effective design and control evaluation, streamlining the complex modeling process. The conceptual design of the exoskeleton hand is executed in Solidworks and embedded into Simscape Multibody to facilitate efficient controller design in MATLAB. A hierarchical controller is implemented, featuring high-level perception and low-level execution layers. At the highest level, kinematic estimation of the hand is computed based on the relationship between forearm EMG signals, handgrip force, and various isometric handgrip patterns. A feed-forward Artificial Neural Network (ANN) is employed for this purpose, trained, and tested using non-invasively collected EMG data from healthy subjects. Experimental results demonstrate that the time domain features extracted using waveform length (WL) yield significant outcomes and are selected to be used to estimate the joint angles in the feed-forward ANN. The ANN has been validated and demonstrates acceptable estimations. This suggests that utilizing WL features in the ANN makes it a viable method for myoelectric control, suitable for integration with a PID controller in low-level control for continuous operation of the hand exoskeleton.