Distributing the control of a temporal network among multiple agents

Abstract

Agents collaborating on a set of tasks subject to temporal constraints must coordinate their activities to ensure that all of the temporal constraints are ultimately satisfied. Simple Temporal Networks (STNs) can be used to concisely represent temporal constraints; however, most algorithms for manipulating such networks presume that a single agent controls the network. Although recent research considers the controllability of networks in which Nature independently controls some temporal intervals, it nonetheless presumes that a single agent controls the rest of the network.This paper makes the following contributions. First, it argues for STNs augmented to accommodate the real-time execution of tasks. Although borrowing from existing approaches, it differs by sharply distinguishing between constraints in the network and the distribution of control over that network. Second, it introduces a more general conception of distributing control of a temporal network, one that is able to accommodate not only networks partially controlled by Nature, but also networks controlled by multiple agents. Third, it construes an existing algorithm for partitioning temporal networks into independent subnetworks as an algorithm for distributing control of a temporal network among multiple agents, each agent having sole control over one subnetwork. It then presents a more general algorithm that allows one of the subnetworks to remain dependent on the rest, thereby enabling the overall network to be less constrained. Restrictions on the control of the dependent subnetwork, specified in terms of necessary and sufficient bounds, guarantee an effective distribution of control over the network.