Human-In-The-Loop Optimal Containment Control for Multi-Agent Systems Under DoS Attack

This paper addresses a novel critic neural network (NN) based framework for the optimal containment control of the Human-in-the-loop (HiTL) multi-agent systems (MASs) under the denial of service (DoS) attack. In this system, the leaders are controlled by the control signal transmitted by a human operator, with the control signal unknown to all followers. Moreover, the communication network environment of the MASs is under the DoS attack. This framework consists of three parts. First, an DoS attack compensator is designed to estimate the neighbors' state of the followers during the attack. Then, a distributed trajectory generator is proposed to generate the state reference for each follower based on the estimated neighbors' state. Finally, an intelligent learning controller is developed to track the state reference, which is learned via critic NN. In the critic designs, a novel weight update law with dynamic regression extension and mixing (DREM) is presented approach to update the weights of the critic NN to relax persistent of excitation (PE) conditions. Furthermore, theoretical analysis is given to prove the relaxation of PE condition. Finally, the feasibility of the proposed method is verified by simulation.