A Varying Parameter Recurrent Neural Network for Solving Nonrepetitive Motion Problems of Redundant Robot Manipulators

Abstract

A novel varying-parameter recurrent neural network [called varying-parameter convergent-differential neural network (VP-CDNN)] is proposed and investigated to solve time-varying convex quadratic programing (QP) problems and applied to solve nonrepetitive problems of redundant robot manipulators in this brief. First, the nonrepetitive problems of redundant robot manipulators are reformulated as a QP scheme. Second, the QP scheme is reformulated as a matrix equation. Third, the proposed VP-CDNN is applied to solve the matrix equation as well as the original QP problem. To illustrate the advantages of VP-CDNN solver, comparison simulations between the VP-CDNN and the fixed-parameter convergent-differential neural network (FP-CDNN) are constructed based on a six-degrees-of-freedom robot manipulator. Two end-effector tasks employed by the VP-CDNN with linear activation function and sinh activation function verify the effectiveness and advantages of the proposed VP-CDNN and its better expansibility. The results of computer simulations and physical experiments demonstrate that the VP-CDNN solver is more effective and accurate than the FP-CDNN solver to solve nonrepetitive problems of redundant robot manipulators.