Enhancing sEMG-Based Finger Motion Prediction with CNN-LSTM Regressors for Controlling a Hand Exoskeleton