Dynamic self-organization and clustering in distributed networked systems for performance improvement

  • Authors:

  • John S. Baras;Pedram Hovareshti;Senni Perumal

  • Affiliations:

  • Institute for Systems Research and the Department of Electrical and Computer Engineering, University of Maryland, College Park, MD;Institute for Systems Research and the Department of Electrical and Computer Engineering, University of Maryland, College Park, MD;Institute for Systems Research and the Department of Electrical and Computer Engineering, University of Maryland, College Park, MD

  • Venue:
  • Allerton'09 Proceedings of the 47th annual Allerton conference on Communication, control, and computing
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

We consider two closely related dynamic self-organization problems in networked control systems. Both are forms of dynamic clustering of nodes. The structure of networked control systems is often abstracted using graph theory. In this abstraction, the nodes of the graph represent the agents and the edges between them represent the relation(s) or the possibility of communication between the corresponding agents. The topology of the communication network supporting a networked control system has critical consequences for its performance. The first problem we address is the development of a distributed self-organization algorithm, resulting into a dynamic two level hierarchy of leader and regular agents, which substantially improves the convergence speed of distributed algorithms utilized by the networked control system. For the second problem, we consider the collaborative control of a group of autonomous mobile agents (e.g. vehicles, robots) supported by a mobile wireless network, consisting of many ground and a few aerial nodes. The agents collaborate to achieve a common goal or objective, like to move in a particular area and cover it, while avoiding obstacles and collisions. Building upon our earlier work on deterministic, randomized and hybrid distributed coordination algorithms we consider the communication needs of the agents, and in particular the connectivity of their communication network as they move. We develop distributed algorithms that automatically select some agents and move them appropriately so as to maintain certain degree of desired connectivity among the moving agents. We characterize the trade-off between the gain from maintaining a certain degree of connectivity vs. the combined cost of communications and the associated dynamic re-positioning of agents. We also describe classes of efficient communication topologies and in particular their similarity to dynamic small world topologies and extensions.