The performance of model predictive control (MPC) largely depends on the accuracy of the prediction model and on the constraints the system is subject to. In several applications, obtaining an accurate knowledge of these elements might be expensive in terms of money and resources, if at all possible. Therefore, by starting from an initial guess, their accuracy can be enhanced by combining MPC with active learning approaches. Even though learning-based MPC approaches are appealing due to the potential performance improvement, the inclusion of online learning can also lead the system to instability, constraint violations, arbitrary performance deterioration, as well as lack of recursive feasibility of the considered MPC scheme. In this talk, we discuss novel learning-based MPC frameworks that actively incentivize learning of the underlying system dynamics and of the constraints, while ensuring closed-loop guarantees.
Join at: imt.lu/aula1