Safety Verification of Neural Network Controlled Systems

Arthur Clavière (1, 2) , Eric Asselin (1) , Christophe Garion (3) , Claire Pagetti (2)

1 Collins Aerospace France
2 DTIS, ONERA, Université de Toulouse [Toulouse]
3 ISAE-SUPAERO - Institut Supérieur de l'Aéronautique et de l'Espace

In this paper, we propose a system-level approach for verifying the safety of systems combining a continuoustime physical system with a discrete-time neural network based controller. We define a generic modelling approach and an associated reachability analysis that soundly approximates the reachable states of the overall system. We illustrate our approach through a real-world use case.

Mots clés: NEURAL NETWORKS; FORMAL METHODS; CERTIFICATION

Domaines: Intelligence artificielle [cs.AI]; Systèmes embarqués

Arthur Clavière, Eric Asselin, Christophe Garion, Claire Pagetti. Safety Verification of Neural Network Controlled Systems. 51st Annual IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN-W), Jun 2021, Taipei, Taiwan. ⟨hal-02975455v2⟩