Model Predictive Control using Hybrid Feedback

Abstract Traditional Model Predictive Controllers make use of computation expensive optimization methods. The challenge of this research is to take advantage of properties of MPC and use a hybrid gradient descent method to replace the on-line optimization by a simple set of differential equations. Continuous- and discrete-time controllers are presented. Promising results are provided for the control of linear systems with smooth convex constraints with different controller tunings. This technique, even if slightly suboptimal, has a clear interpretation, is efficient and is suitable for implementation on limited embedded microcontrollers.