Lower limb movement intention recognition for rehabilitation robot aided with projected recurrent neural network

Abstract For the lower limb rehabilitation robot, how to better realize intention recognition is key issue in practical application. Recognition of patient’s movement a challenging research work, which needs be studied from shallow deep. Specifically, it necessary ensure that normal person can accurately recognized, and then improve model patients. Therefore, before studying intention, essential consider first, also for safety considerations. In recent years, new Hill-based muscle has been demonstrated capable directly estimating joint angle an open-loop form. On this basis, by introducing recurrent neural network (RNN), whole prediction process achieve more accuracy closed-loop However, traditional RNN algorithms, activation function must convex, brings some limitations solution problems. Especially, when convergence speed limited applications, as error continues decrease, performance will greatly affected. To end, projected (PRNN) proposed, relaxes condition convex used saturation constraint case. addition, corresponding theoretical proof given, PRNN method with successfully applied experiment compared model.