Direct data-driven design of neural reference governors

In this paper, we present a direct data-driven approach to synthesize model reference controllers for constrained nonlinear dynamical systems. To this aim, we employ a hierarchical structure composed by a receding-horizon reference governor and a data-driven low-level controller. Unlike existing approaches, here we jointly design the two blocks by solving a single optimization task, exploiting the fact that the inner controller will never be used alone. The performance of the proposed method is assessed by means of two simulation examples, involving the control of two highly nonlinear benchmark systems.