Nonlinear Model Predictive Control of Robot Manipulators Using Quasi-LPV Representation

Abstract

Nonlinear optimization techniques often suffer from time-consuming computational load, which impedes them to be implemented as controller of fast plans, or when a fast action like trajectory tracking is required. In this paper, a Nonlinear Model Predictive Control (NMPC) approach is used to perform the trajectory tracking problem in a robot manipulator in the presence of input saturation and un-modeled dynamics, using the Quasi-Linear Parameter Varying (Quasi-LPV) representation. In this method, instead of the nonlinear state difference equations of the system, a sequence of linearized state equations about a nominal state-control history, over the prediction horizon, is used. By so doing, standard Quadratic Programming (QP) optimization algorithms could be used for the online optimization problem, therefore, speed and efficiency of convergence to the optimal solution would be enhanced. Efficacy of this method is shown by simulation study of a 2-DOF robot manipulator.