Multi-agent opportunism

<p>The real world is a complex place, rife with uncertainty; and prone to rapid change. Agents operating in a real-world domain need to be capable of dealing with the unexpected events that will occur as they carry out their tasks. While unexpected events are often related to failures in an agent's plan, or inaccurate knowledge in an agent's memory, they can also be opportunities for the agent. For example, an unexpected event may present the opportunity to achieve a goal that was previously unattainable. Similarly, real-world multi-agent systems (MASs) can benefit from the ability to exploit opportunities. These benefits include the ability for the MAS itself to better adapt to its changing environment, the ability to ensure agents obtain critical information in a timely fashion, and improvements in the overall performance of the system.</p><p>In this dissertation we present a framework for multi-agent opportunism that is applicable to open systems of heterogeneous planning agents. The contributions of our research are both theoretical and practical. On the theoretical side, we provide an analysis of the critical issues that must be addressed in order to successfully exploit opportunities in a multi-agent system. This analysis can provide MAS designers and developers important guidance to incorporate multi-agent opportunism into their own systems. It also provides the fundamental underpinnings of our own specific approach to multi-agent opportunism.</p><p>On the practical side, we have developed, implemented, and evaluated a specific approach to multi-agent opportunism for a particular class of multi-agent system. Our evaluation demonstrates that multi-agent opportunism can indeed be effective in systems of heterogeneous agents even when the amount of knowledge the agents share is severely limited. Our evaluation also demonstrates that agents that are capable of exploiting opportunities for their own goals are also able, using the same mechanisms, to recognize and respond to potential opportunities for the goals of other agents. Further and perhaps more interesting, we show that under some circumstances, multi-agent opportunism can be effective even when the agents are not themselves capable of single-agent opportunism.</p>

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Dissertation

Year

2005

Author dc:creator

• Lawton, James H

Contributors dc:contributor

• Roy M Turner

• Computer Science

Repository record dc:identifier

https://scholars.unh.edu/dissertation/288

OAI identifier oai:identifier

oai:scholars.unh.edu:dissertation-1287