Resilient Cooperative Control of Networked Multi-Agent Systems

Abstract

Networked multi-agent systems consist of a set of agents that exchange information through a medium referred to as the network. The agents in the networked system are tasked with achieving certain group objectives. These group objectives are typically decomposed into constituent objectives that require coordination among the individual agents through distributed algorithms. Two fundamental constituent objectives are consensus and synchronization. Due to the large-scale, distributed, and dynamic nature of many networked multi-agent systems, the most effective and applicable distributed consensus and synchronization algorithms are those that use purely local strategies. When using algorithms based on purely local strategies, the agents make decisions and act based only on their sensor measurements, calculations, dynamics, and direct interactions with neighbors in the network. No global information is shared or assumed to be known. Instead, information is disseminated within the network in an iterative or diffusive manner. While the last decade has seen a surge of research in the cooperative control of networked multi-agent systems, the issue of security has just recently begun to be explored. Much of the existing research focuses on detection and identification of compromised nodes and network attacks, which requires nonlocal information. Such global information may not be available, rendering these techniques inapplicable. This dissertation introduces consensus and synchronization algorithms using purely local strategies that are resilient to the adversarial influence of compromised nodes in the network. Various threat models are defined to model the behavior of compromised agents, along with scope of threat assumptions that define the scope of interactions allowed between compromised and uncompromised nodes. The efficacy of the consensus and synchronization algorithms is analyzed under the assumptions defined by the adversary models. For this analysis, a recently introduced graph theoretic metric, network robustness, is refined and shown to be the key property for characterizing the network topological conditions required for the consensus and synchronization algorithms to succeed. Several important properties of robust networks are provided and several algorithms for determining the robustness of a network are given.