Efficient robust output feedback MPC

Research on robust model predictive control (MPC) has produced a plethora of results, most of which assume that the states are measurable. When not all states are measurable one must consider using output feedback. Earlier research on robust output feedback MPC either aims at simplified systems or is very computationally demanding. This paper exploits the convenient quasi-closed loop controller and separates the error dynamics and the nominal dynamics. The errors in the prediction steps are described in terms of polytopic sets with parallel edges but variable scalings and arbitrary complexity. In Mode 2, the error set is assumed invariant and thus a maximum admissible set is computed offline as a robust invariant terminal set. A quadratic programming problem is then solved online, just as done in the case of nominal MPC. The strategy enjoys guaranteed theoretical properties and can be applied to systems with multiplicative uncertainty, additive disturbances and measurement noise.