Gradient-augmented Supervised Learning of Optimal Feedback Laws Using State-dependent Riccati Equations

A supervised learning approach for the solution of large-scale nonlinear\nstabilization problems is presented. A stabilizing feedback law is trained from\na dataset generated from State-dependent Riccati Equation solves. The training\nphase is enriched by the use gradient information in the loss function, which\nis weighted through the use of hyperparameters. High-dimensional nonlinear\nstabilization tests demonstrate that real-time sequential large-scale Algebraic\nRiccati Equation solves can be substituted by a suitably trained feedforward\nneural network.\n