Myoelectric Control of a Virtual Hand Based on Third-Order Cumulants Eugenio C. Orosco 1 , Richard Godinez 2 , Natalia M. Lopez 3 , Sridhar Arjunan 4 , Dinesh K. Kumar 4 , Teodiano F. Bastos-Filho 2 , Fernando di Sciacio 1 1 Instituto de Automática, Universidad Nacional de San Juan, Argentina 2 PPGEE. Universidad Federal de Espirito Santo (UFES).Vitoria. Brazil 3 Gabinete de tecnología médica, Universidad Nacional de San Juan, Argentina 4 Royal Melbourne Institute of Technology (RMIT) University Abstract—. This work presents a myoelectric control scheme of a virtual hand based on higher order statistics. The main objective of the experiments is to benefit the general amputee population, offering a training tool to potential prosthesis users. A simple and direct input control is based on the residual SEMG signals from the upper limb of amputee people. The third-order cumulant features are calculated recursively and a non-threshold control scheme is implemented online in a virtual hand. These SEMG features open and close the virtual hand assisted by the corresponding user's visual feedback. Seven experiments were concluded satisfactorily by an amputee volunteer. Keywords— sEMG; virtual hand; HOS; upper limb, amputee I. INTRODUCTION Myoelectric control is widely used in assistive technologies, including multifunction prosthesis, wheelchairs, grasping control, virtual keyboards, diagnoses and clinical applications, such as functional neuromuscular stimulation. The main goal of this work is to control a virtual hand using sEMG signals from the upper limb and user’s visual feedback of amputee volunteers. A wide variety of different features for sEMG applications have been extensively reported in the literature [1]. Features can be classified according to several criteria, e.g., their specific feature domain (time, frequency, or time-frequency domain); the linear or nonlinear nature of the feature extraction mapping; or whether the feature extraction method is based on second or high-order statistics. High-order statistics, and the probabilistic models based on them, allow modeling non-Gaussian and nonlinear signals [2- 6]. In past years, several authors have proposed the use of HOS concepts to analyze and classify the sEMG signals, e.g., high- order cumulant sequences, high-order spectra or polyspectra, skewness, kurtosis, and other concepts [7-12]. In this paper, the third-order cumulant sequence from multiple sEMG channels is proposed as a HOS-based feature method applied to the myoelectric control of a virtual hand, an online application. The third-order cumulant is estimated from two sEMG channels, corresponding to biceps brachii and brachioradialis. The advantage of using the third-order cumulant instead of the second-order one is that the third-order cumulants of the non-EMG signals are zeros, while the second- order cumulants are not zeros. This property allows to do featuring detection without a threshold, avoiding the cumulative residuals of the noise [12], and to implement a direct control scheme. The control scheme uses zero lags time shifting third-order cumulants estimation. The joint velocities and coordinates are based on the difference of the third-order cumulants of the two sEMG channels. Finally, the virtual hand was developed according to the physiological limitation of a human hand and it is executed online in Windows environment. The paper is organized as follows: Section II explains the experimental protocol, the sEMG data collection and segmentation, the cumulant theory and estimation, the virtual hand development and the online myoelectric control methods. In the Section III, the results of two experiments are presented. Finally, the conclusions are exposed. II. MATERIALS AND METHODS A. Experimental protocol A protocol is designed to experiment with the upper limb SEMG signals for pronator and brachioradialis muscles from an amputee volunteer and a virtual hand, as shown in Fig. 1. The experiments are a first test with amputee people and this protocol is developed in an effort to directly benefit to the general amputee population with applications that do not require a previous training and learning. The movements of interest in this work are pronation, supination, and rest position. The volunteer is encouraged to perform a free sequence of movements prompted by their visual feedback based on the virtual hand. Also, the user has no previous contact with the system. The volunteer is healthy, with no history of muscle weakness, neurological diseases or drug therapy and he has a congenital malformation, i.e., unilateral phocomelia below his elbow. The volunteer approved and signed an informed consent form, according to the experiments to be performed. The tasks were adapted to the user's capabilities. Due to the fact of the level of amputation of the volunteer, the movements of pronation and supination are matched to hand opening and closing, i.e., if the user makes a pronation movement, then the virtual hand opens. The trials were executed in seven experiments, in which the effects of muscular fatigue were taken into account as a practical situation and this will be compensated in future works. Each trial has 120s maximum duration. Figure 1. Experiments executed in a laptop with an amputee volunteer.