An SOR-like method for fast model predictive control

Abstract

In this paper, we propose an iterative first-order gradient method for fast and efficient implementation of Model Predictive Control (MPC). The proposed method utilizes a matrix-splitting scheme common to successive over-relaxation (SOR) iterative methods and may be considered a generalization of many existing fast gradient-based methods for MPC. We comment on parameter choices for global convergence and demonstrate the efficiency of the ensuing algorithms using a simulation example.