Power-sum Activated Neural Dynamics for Lower Limb Motion Intention Recognition

It can be found that the exact solution of the equations has quite similarity to the pursuing of the minimum error in the field of control. Therefore, many control problems can be explained from the perspective of solving equations. As a powerful computing tool, neural dynamics model has been widely applied in addressing time-varying equations and optimization problems because of its remarkable convergence and robustness. For neural dynamics model, the selection of activation function plays a significant role in its performance. In this paper, a special nonlinearly activated (i.e. power-sum activation function) neural dynamics model is exploited to the experimental simulation of intention recognition of lower limb motion. Compared with the linear activation function, the corresponding analyses of convergence and robustness are given respectively, demonstrating the superior characteristics of the neural dynamics model with power-sum activation function.