Learning-Based Event-Triggered Control for Synchronization of Passive Multiagent Systems Under Attack

In this article, we study the synchronization of a group of output passive agents that communicate with each other according to an underlying communication graph. A distributed event-triggered control framework that guarantees synchronization and reduces the required communication rate is introduced. A general Byzantine attack on a multiagent system is defined and its negative effects on synchronization are characterized. The Byzantine agents are able to intelligently falsify their data and manipulate the underlying communication graph by altering their control feedback weights. Next, a decentralized decision-making and detection framework is introduced and its steady-state and transient performances are analyzed. Furthermore, a method of identifying Byzantine neighbors and a learning-based procedure for estimating the attack parameters are introduced. Finally, learning-based control frameworks to mitigate the effects of the attack are proposed.