Distributed model-free deep reinforcement learning control for unknown nonlinear multi-agent systems

Abstract

This paper addresses the consensus problem of a class of unknown nonlinear multi-agent systems (MASs) via a novel distributed model-free deep reinforcement learning (DRL) control method. First, the DRL-based feedback linearization approach is developed to learn an approximated linearized control protocol in a model-free fashion. Then, a distributed observer is further designed to estimate the unavailable information of the exosystem. Based on the obtained nominal linear model and developed distributed observer, the distributed model-free control protocol is synthesized such that the consensus of nonlinear MASs can be achieved. Finally, the validity of the proposed control scheme is verified by simulations.