An online re-linearization scheme suited for Model Predictive and Linear Quadratic Control

This technical note documents the equations for primal-dual interior-point quadratic programming problem solver used for MPC. The algorithm exploits the special structure of the MPC problem and is able to reduce the computational burden such that the computational burden scales with prediction horizon length in a linear way rather than cubic, which would be the case if the structure was not exploited.

It is also shown how models used for design of model-based controllers, e.g. linear quadratic and model predictive, can be linearized both at equilibrium and non-equilibrium points, making the presented exstension of the controller formulation equivalent to that of the extended Kalman filter compared to an ordinary Kalman filter.