Cohesive Transitions in Decentralized Networks with Delays

Abstract

This dissertation involves development of controls for cohesive transitions in non-ideal networks with delays. Most existing network control strategies focus on improving the convergence rate of networks to consensus, say from one operating value to another. However, maintaining consensus during the transitions, (i.e. achieving cohesive transitions), can be as important as achieving the final consensus value, for instance, in spacing control of connected vehicles, and flexible-object transport using decentralized robot networks. It is shown that increasing the convergence rate to consensus does not necessarily lead to cohesive networks. Recent work on delayed self reinforcement (DSR) has shown to improve the cohesion during transitions in decentralized networks using delayed versions of already available local neighbor information. However the DSR-based approach assumes ideal networks with no agent or network delays. Agent delays represent potential delays due to time taken during actuation or due to local agent dynamics, and network delays can occur due to the time lost during sensing or communication between neighboring agents. The presence of such delays can cause instability, and further, can lead to loss of cohesion with the DSR-based approach. The thesis develops stability conditions, as sufficient delay margins on the net- work and agent delays in the DSR-based approach, and methods to maintain for cohesive transitions in networks with delays.