On MPC based trajectory tracking

This work proposes and investigates a tracking scheme for linear, constrained system based on a combination of model predictive control, virtual references and unknown input observers. In contrast to existing results the proposed approach allows to track a larger class of trajectories exactly: it does not require a reference model, does not need to assumes constant/periodic references or that the reference converges to a steady state. The scheme guarantees under mild conditions recursive feasibility independent of the reference and asymptotically exact tracking. It is computationally tractable, since only a convex quadratically constrained quadratic program or a convex quadratic program needs to be solved at each time step. We outline the applicability and the efficacy of the proposed approach using two examples.