State estimation for control: an approach for output-feedback stochastic MPC (2303.00873v1)

Abstract: The paper provides a new approach to the determination of a single state value for stochastic output feedback problems using paradigms from Model Predictive Control, particularly the distinction between open-loop and closed-loop control and between deterministic optimal control and stochastic optimal control. The State Selection Algorithm is presented and relies on given dynamics and constraints, a nominal deterministic state-feedback controller, and a sampling based method to select the best state value, based on optimizing a prescribed finite-horizon performance function, over the available candidates provided by a particle filter. The cost function is minimized over the horizon with controls determined by the nominal controller and the selected states. So, the minimization is performed not over the selection of the control other than through the choice of state value to use. The algorithm applies generally to nonlinear stochastic systems and relies on Monte Carlo sampling and averaging. However, in linear quadratic polyhedrally constrained cases the technique reduces to a quadratic program for the state value. The algorithm is evaluated in a set of computational examples, which illustrate its efficacy and limitations. Numerical aspects and the opportunity for parallelization are discussed. The examples demonstrate the algorithm operating, in closed-loop with its attendant particle filter, over the long horizon.