Who needs QP for linear MPC anyway? (Q1614310)

An approach which offers an attractive alternative to the quadratic programming (QP) based model predictive control (QPMPC) is proposed. The numerical advantages of Newton-Raphson MPC (NRMPC) over QPMPC are outlined, and the extension of NRMPC (ENRMPC) is derived. The ENRMPC accepts univariate optimization outside the ellipsoidal set thereby removing most of the conservatism incurred through the use of ellipsoidal rather than maximal polyhedral sets. The results of Monte Carlo simulations on the large number of constrained state-space models show a performance indistinguishable from the QPMPC one and a computational cost comparable to the NRMPC one.