NONCONVEX OPTIMIZATION AND ROBUSTNESS IN REALTIME MODEL PREDICTIVE CONTROL

Recent works in the nonlinear MPC literature have presented “realtime” optimization approaches based upon incremental updating of input parameters using local descent directions of the cost functional. The main downside to these methods is their strong dependence upon the values used to initialize the input parameters. In this note we study the robustness issues associated with nonlocal search methods in continuous-time MPC, and demonstrate a framework for robustly incorporating these approaches in a realtime setting.