Design of a deep neural network-based integral sliding mode control for nonlinear systems under fully unknown dynamics

In this letter a novel deep neural network based integral sliding mode (DNN-ISM) control is proposed for controlling perturbed systems with fully unknown dynamics. In particular, two DNNs with an arbitrary number of hidden layers are exploited to estimate the unknown drift term and the control effectiveness matrix of the system, which are instrumental to design the ISM controller. The DNNs weights are adjusted according to adaptation laws derived directly from Lyapunov stability analysis, and the proposal is satisfactorily assessed in simulation relying on benchmark examples.