Reachability under Bounded Variations for Safety in Control Loops with Reinforcement Learning enhancing Fault Accommodation
Résumé
Ensuring both safety and fault tolerance in reinforcement learning (RL)-based control systems is critical for their deployment in aerospace systems. In this context, the aim of this work is twofold. First, a new zonotopic reachability algorithm is proposed to address the case of bounded inputs with bounded variations. It is obtained from a reformulation based on bounded delayed exogenous inputs. Second, this algorithm is used for offline and training-independent safety verification of control loops hybridizing a baseline model-based controller with a constrained reinforcement learning (CRL) agent. The agent is then trained not only to improve given performance metrics under unmodeled environmental dynamics, but also to achieve incipient fault accommodation. A numerical example based on aircraft flight dynamics illustrates the main results and shows the efficiency of the proposed scheme.