Fostering Dexterous sEMG-Driven Control for Hand Exoskeletons Through Real-Time CNN-LSTM Regression

In recent years, more and more people are experiencing forms of disability, leading to increasing interest in the field of assistive and rehabilitative robotics. In particular, given the importance of the hand in everyday life, many hand exoskeletons have been developed over the years. At the same time, increasingly sophisticated control systems have also been implemented. This paper presents the real-time implementation of a regressor for hand exoskeletons based on a CNN-LSTM structure and exploiting surface sEMG signals obtained with a Myo Armband. The regressor aims to provide a reference angle for the low-level control system implemented on the exoskeleton. The results show that the proposed regressor achieves real-time performance comparable to offline implementation.