Stabilizing non-linear model predictive control using linear parameter-varying embeddings and tubes

This paper proposes a model predictive control (MPC) approach for non-linear systems where the non-linear dynamics are embedded inside a linear parameter-varying (LPV) representation. The non-linear MPC problem is therefore replaced by an LPV MPC problem, without using linearization. Compared to general non-linear MPC, advantages of this approach are that it allows for the tractable construction of a terminal set and cost, and that only a single convex program must be solved online. The key idea that enables proving recursive feasibility and stability, is to restrict the state evolution of the non-linear system to a time-varying sequence of state constraint sets. Because in LPV embeddings, there exists a relationship between the scheduling and state variables, these state constraints are used to construct a corresponding future scheduling tube. Compared to non-time-varying state constraints, tighter bounds on the future scheduling trajectories are obtained. Computing a scheduling tube in this setting requires applying a non-linear function to the sequence of constraint sets. Outer approximations of this non-linear projection-based scheduling tube can be found, e.g., via interval analysis. The computational properties of the approach are demonstrated on numerical examples.