Approximating Reachable Sets for Neural Network-Based Models in Real Time via Optimal Control

In this brief, we present a data-driven framework for real-time estimation of reachable sets control systems where the plant is modeled using neural networks (NNs). We utilize running example quadrotor model that learned trajectory data via NNs. The NN offline can be excited online to obtain linear approximations reachability analysis. use dynamic mode decomposition (DMD)-based approach lifting model. models thus obtained optimal theory polytopic in real time. tuned arbitrary degrees accuracy. proposed extended other nonlinear NNs estimate dynamics. demonstrate effectiveness an illustrative simulation