Sequential L1 Quadratic Programming for Nonlinear Model Predictive Control

Dimitri Boiroux1, John Bagterp Jorgensen1

  • 1Technical University of Denmark

Details

11:20 - 11:40 | Thu 25 Apr | Veleiros | ThA1.3

Session: Model Predictive Control

Abstract

In this paper, we present and describe a computationally efficient sequential l1 quadratic programming (Sl1QP) algorithm for Nonlinear Model Predictive Control (NMPC). We use a tailored trust region sequential quadratic programming for the solution of the optimal control problem (OCP) involved in the NMPC algorithm. We use a multiple shooting approach for numerical integration and sensitivity computation. A second order correction ensures a faster convergence of the SQP algorithm. We exploit the structure of the OCP by using an efficient primal-dual interior point algorithm based on Riccati factorizations and a block diagonal BFGS update of the Hessian matrix. The complexity scales linearly with the prediction horizon length. We numerically evaluate and compare the performance of our algorithm on a numerical example.