Neural Koopman Lyapunov control

Learning and synthesizing stabilizing controllers for unknown nonlinear control systems is a challenging problem real-world industrial applications. Koopman operator theory allows one to analyze through the lens of linear bilinear systems. The key idea these methods lies in transformation coordinates system into observables, which are that allow representation original (control system) as higher dimensional (bilinear control) system. However, systems, model obtained by applying based learning not necessarily stabilizable. Simultaneous identification stabilizable lifted well associated observables still an open problem. In this paper, we propose framework construct such models identify their from data simultaneously embedding underlying affine Control Lyapunov Function (CLF) using learner falsifier. Our proposed approach thereby provides provable guarantees asymptotic stability Numerical simulations provided validate efficacy our class feedback control-affine