Robust Nonlinear Model Predictive Control With Partial State Information

Abstract

A nonlinear controller methodology is developed based on nonlinear model predictive control and set-membership estimation, resulting in a controller which is robust to model uncertainties and bounded noise sources. The model predictive control makes direct use of the estimated bounds of the states and model parameters through integration as constraints in the optimization problem. This results in lower computation times and the ability to satisfy constraints with partial state information. Stability of the integrated estimation and control algorithm is guaranteed through the use of a contractive terminal cost. Real time implementation is developed through a unique sequential quadratic programming solver, which develops an initial feasible solution very quickly, and continues to optimize as time allows. The estimation and control algorithm is demonstrated in software simulation, hardware simulation, and in uninhabited aircraft flight tests. A formation flight example consisting of two aircraft is presented, showing how the controller can maintain a formation within given constraints and avoid collisions while in the presence of noise. Real time flight control results for an autonomous airplane demonstrate the capability of the algorithms to control a nonlinear aircraft in real time around a suddenly appearing threat, thus avoiding collisions.