Toward a Control Theory for Networks
Sandip Roy, Ali Saberi, Anton A. Stoorvogel · International Journal of Robust and Nonlinear Control · 2006
Modelling and algorithm-development tasks in such diverse application areas as autonomous vehicle coordination, sensor networking, and air traffic flow management have led to flourishing interest in network dynamics and control. The design problems of interest in these application areas are very often, at their core, decentralized control problems, and hence we believe strongly that the controls community will (and should) be at the forefront of these new and exciting research directions. What is becoming increasingly apparent, however, is that these new decentralized control problems are difficult ones: they cannot be addressed by the traditional methodology of decentralized control in which the network is viewed as a disturbance impacting robust agents, nor can they be addressed by ad hoc adaptation of centralized-control techniques. Instead, solutions to these exciting control problems must exploit the network communications/sensing, and hence the problems require us to understand in a systematic way the role played by network topology in decentralized control. In our opinion, there is as yet no suitable methodology for designing controllers that take advantage of the network topology, and even very few promising approaches for design. Though the general design of decentralized controllers is undoubtedly a daunting task, our strong belief is that some common characteristics of the modern network applications will permit us to design good controllers for these applications, and in turn give significant insight into the more general design problem. Such modern networks as mobile sensor systems and autonomous-vehicle teams have in common that (1) they are built of agents with highly simplified and often identical internal dynamics, (2) require inter-agent communication to complete a range of tasks, and (3) are subject to severe constraints and topological variation. For this class of communicating-agent networks, there is much hope for understanding the role played by the network topology in permitting control, and for developing high-performance controllers for a range of algorithmic tasks. The articles in this special issue take several approaches toward, and study several aspects of, controller analysis and design for communicating-agent networks. We believe strongly that, together, these articles capture and explore many key challenges in this exciting new direction in control theory.