Simulations and experiments of ZNN for online quadratic programming applied to manipulator inverse kinematics

Zhang neural network (ZNN), a special class of recurrent neural network (RNN), has recently been introduced for time-varying convex quadratic-programming (QP) problems solving. In this paper, a drift-free robotic criterion is exploited in the form of a quadratic performance index. This repetitive-motion-planning (RMP) scheme can be reformulated into a time-varying quadratic program subject to a linear-equality constraint. As QP real-time solvers, two recurrent neural networks, i.e., Zhang neural network and gradient neural network (GNN), are then developed for the online solution of the time-varying QP problem. Computer simulations performed on a four-link robot manipulator demonstrate the superiority of the ZNN solver, compared to the GNN one. Moreover, robotic experiments conducted on a six degrees-of-freedom (DOF) motor-driven push-rod (MDPR) redundant robot manipulator substantiate the physical realizability and effectiveness of this RMP scheme using the ZNN solver.