Divide-and-conquer muscle synergies: A new feature space decomposition approach for simultaneous multifunction myoelectric control

Simultaneous multifunctional control based on surface electromyography (sEMG) is a key issue for natural and intuitive use of upper-limb prostheses in clinical and commercial applications. However, muscle synergies make simultaneous multifunctional control technologically challenging. In this study, we proposed a new feature space decomposition approach to alleviate the difficulty brought by muscle synergies for simultaneous control of hand and wrist movements. In the feature space decomposition approach, Gaussian mixture modeling (GMM) clustering is used to split the whole feature space into a set of Gaussian clusters, each consisting of samples with similar characteristics, to “divide-and-conquer” the complex muscle synergies. Then, a hybrid simultaneous control strategy, which consists of switch control of hand movements and proportional control of wrist movements, is performed in each cluster, instead of in the whole feature space. In the experimental study, sEMG signals were recorded during static and dynamic muscle contraction involving 2-dimensional wrist rotation (flexion-extension and radial-ulnar deviation) and 3 basic hand movement patterns (relaxing, fisting and grasping). Results show that, the new feature space decomposition approach can increase the accuracy for switch control of hand movement patterns from 90.10% to 96.62%, and can improve the correlation between true and predicted values of wrist rotation angular velocity from 0.71 to 0.84 (for wrist flexion-extension) and from 0.67 to 0.82 (for wrist radial-ulnar deviation) for proportional control of wrist. The proposed feature space decomposition approach has the potential to yield simultaneous multifunctional control for sEMG-based upper-limb prosthesis.