Controlling 3D Model of Human Hand Exploiting Synergistic Activation of The Upper Limb Muscles

Surface electromyography (EMG) based control of hand exoskeleton has been developed in numerous methods, but a proportional control of multiple degrees of freedom (DOF) exoskeleton is still rarely seen in recent years. This paper proposes an alternative to the most recent solutions for proportionally estimating multiple DOFs of the hand. High-density electromyography (HD-EMG) was used to estimate the joint angle of five fingers in sophisticated gestures. The gestures were the ten alphabets of American Sign Language (A B C D F I K L O W). The association between HD-EMG and kinematics was mapped by using a neural network (NN) and k-Nearest Neighbour (kNN). The estimation accuracy was around 70-95 % (R index) for the eleven DOFs in four normal-bodied subjects’ hands. Furthermore, kNN performed better than NN, even in the case of input feature reduction. The regression results also were able to distinguish 9 of 10 alphabets (except O), with the false interpretation because of the identical muscle activity and kinematics between O and C. The approach described in this paper provides a practical solution to proportionally and simultaneously control multiple DOFs for the hand exoskeleton.