Iteration governor for suboptimal MPC with input constraints

This paper introduces a supervisory scheme, called the iteration governor (IG), that augments a suboptimal input-constrained MPC policy by performing online selection of the reference command and the number of optimization iterations used to generate a control input. At each time step, an auxiliary reference command is selected so that the state is contained in a region of attraction (ROA) for a corresponding auxiliary equilibrium under optimal MPC. Simultaneously, the number of optimization iterations used to generate the control input is preselected to ensure that the resulting suboptimal input steers the state towards this auxiliary equilibrium. Theoretical guarantees are provided that ensure the auxiliary reference converges to the target reference in finite-time, the state converges to the target equilibrium, and the number of online iterations never exceeds a constant that can be computed offline.