Model Predictive Control for Industrial Robots

Abstract

Recently, some interests have been shown in applying the Model Predictive Control (MPC) method for robot position control (Camacho & Bordons,1995; Li et al., 1995; Wei & Fang,1999). The MPC approach is conceptually different from the traditional robot control methods. The MPC control action is determined by optimising a performance index, typically the error between the output prediction derived from the model and the desired output, over a time horizon. The optimal control actions are applied to the system, and the system outputs are measured over the time horizon. The above steps are repeated until the predicted tracking errors are minimised to be within the permitted range. The MPC was primarily introduced for industrial process control. Therefore, its application in robot control has not been widely reported. Usually, in a conventional MPC controller the predictive model is either an impulse or a step response model. However, these models are not suitable for non-linear systems such as an industrial robot (Camacho & Bordons,1995; Soeterboek,1992). To overcome the problems resulted from using linear models, researchers have tried to extend the MPC controller to include non-linear models. Joseph et al. derived the non-linear model of the underlying system through an analytical way (Joseph et a/.,1988). Others applied NNs as the predictive model by utilising the universal approximation