Multi-agent Networked Systems with Adversarial Elements

Abstract

In nutshell, multi-agent networked systems involve the modeling framework of multiple heterogeneous agents (or decision makers, or players) connected in various ways, distributed over a network (or interacting networks) and interacting with limited information (on line and off line) under possibly conflicting objectives. We can actually view the agents as nodes in a graph or multiple graphs, which could be time varying (some edges in the network appearing or disappearing over time [11]), and the nodes themselves could be mobile [8]. In such settings, agents actually interact in a three-tiered architecture, with each tier corresponding to a different layer [21] , namely: Layer 1, where the agents operate and decisions are made; Layer 2, which is the information level where data, models, and actionable information reside and are exchanged; and Layer 3, which consists of the physical communication network that is used for Layers 1 and 2, and contains software and hardware entities, as well as sensors and actuators with which the teams interface with the dynamic physical environment. The underlying network for Layer 1 can be viewed as a collaboration network, where edges of the corresponding graph capture the collaboration among corresponding nodes (agents); the network for Layer 2 can be viewed as a communication/information network, where edges of the corresponding graph constitute communication links (uni- or bi-directional) among corresponding nodes (agents); and Layer 3 can be viewed as a physical network, where edges constitute the physical links.