Reachability Analysis of Linear Parameter-Varying Systems with Neural Network Controllers
Résumé
A method is presented to obtain outerapproximations of forward reachable sets for Linear Parameter-Varying (LPV) systems controlled by neural networks for safety verification. The method relies on abstracting the activation functions by optimization-based sector constraints. The forward reachable sets are obtained in terms of solutions to a set of Linear Matrix Inequalities (LMIs). A numerical example is provided to demonstrate the applicability of the proposed method. A comparison with some available methods is also included.
Origine : Fichiers produits par l'(les) auteur(s)