Continuous Estimation of Multijoint Kinematics from Surface EMG during Daily Grasping Tasks

Abstract

The accurate and efficient estimation of complex human movement intention from measurable neural signals is meaningful for the intelligent control of prostheses. Previous researches have achieved continuous estimation of multijoint kinematics from electromyography (EMG) signals for myocontrol of robotic hands. However, the adaptive recruitment of different motor units during various tasks with different movement velocities and external loadings was not taken into account in related research. Taking hand grasping as a representative example, this paper proposes a method based on the long-short term memory (LSTM) network to estimate complex multijoint kinematics during grasping tasks with different levels of movement velocities and external loadings. This method is verified to be able to result in higher training efficiency and more accurate estimation compared with a previous method even the mapping between EMG and multijoint kinematics became more complex. The results demonstrate this method is promising to be applied to daily use rather than lab research only.