Managing Resource Limited Multi-Agent Systems Using Real-Time Interventions

Abstract

In transportation management, building management and water management systems, dynamic autonomous and self-interested agents compete for limited shared resources. When left to their own devices, the collective behavior of the individuals in these systems does not meet social goals. Therefore, it is necessary to have a regulatory authority to manage resource usage, prevent wrong actions and to fulfill social goals. However, the task of regulatory entities in managing and distributing resources is difficult because agents' preferences are not visible to them. Hence, it is difficult to predict agents' future behavioral patterns. In this work, we conducted a user study to investigate how effectively people can regulate (in real time) a group of autonomous agents in a shared resource environment. Specifically, we developed a software test-bed of a water supply system that simulates an apartment building. We then asked users to supply real-time (hourly) prices of water under several conditions. Our results show that adding real-time interventions to societies of both static and learning agents reduced social welfare. However, adding constraints to regulatory power can help to overcome these deficiencies. We noticed that regulators created better interventions when they were given limited regulatory power. Therefore, further studies needed to help us understand how to help regulators efficiently regulate adaptive agents in real-time dynamic systems.

Date of Award	2014
Original language	American English
Supervisor	Jacob Crandall (Supervisor)