Stability guarantees for real-time model predictive control

Abstract

The application of predictive controllers to high-speed embedded systems requiring computation times in the millisecond range opens up new possibilities but also poses new challenges for control. Computational constraints limiting the solution time of standard convex solvers generally invalidate the theoretical guarantees on constraint satisfaction and stability of optimal MPC controllers. This is addressed by new certified control approaches that provide reliable implementations. This talk will discuss two important aspects in this context. A real-time MPC approach with robust stability guarantees for arbitrary time limits is presented together with its high-speed implementation in the toolbox FORCES. Afterwards, we report on first steps toward addressing the issue of real-time constraint satisfaction by means of a new soft constrained MPC method with closed-loop stability guarantees.