Wrist and grasp myocontrol: Online validation in a goal-reaching task

Abstract

Simultaneous and proportional control of hand and wrist prostheses based upon surface electromyography (myocontrol) is still largely an open issue in the community of assistive robotics. It entails the ability of discriminating the activation levels for each degree of freedom (DOF) of the hand/wrist complex, using as few sensors as possible. Furthermore, one should avoid having the human subject train the underlying machine-learning system with all combinations of activations. To tame this problem we have proposed Linearly Enhanced Training (LET), a procedure through which a training set, composed of single-DOF activations provided by the user, is artificially completed with synthetic multi-DOF activations. In this paper, we validate the LET procedure through an online psychophysical experiment carried out on 16 intact subjects and one trans-radial amputee, in which a specific goal must be reached within a determined amount of time. Each subject tried to reach the goal in either of four different scenarios, while the LET procedure was activated or not, and while an optimisation was used or not. A comparative analysis of the results reveals that the usage of LET does not entail any statistically significant difference in the overall performance, and that the usage of the optimisations significantly improves it. Therefore, one can benefit from the drastic reduction of training time due to LET without suffering from significant reduction in performance. The optimisation showed the strong tendency to reduce the time it took to successfully accomplish a task, on average by 1.195s. A comparison of the intact subjects and the trans-radial amputee showed that half of the performance measures of the amputee lie in the 95% confidence interval determined by the able-bodied group.