Adaptive MPC & System Learning via LMPC (MS)
The advantage of the Learning MPC (LMPC) is that it guaranties stability for the closed loop system and local optimality of the closed loop trajectory. The goal of this project is to exploit the properties of the LMPC to build an adaptive LMPC control strategies. The idea is augment the system dynamics with an error dynamics and to use a time varying state feedback law to compensate for model mismatch. We showed that for a SISO system that the LMPC can compensate for the model mismatch. Moreover, as the input computation and the parameter estimation are coupled in one optimization, the framework provides theoretical guaranties about safety and convergence.
Prerequisite: Solid control background and programming skills (C, Python, Matlab). Good optimization or machine learning background.