Tracing Dependencies of Strategy Selections in Agent Design

Dung N. Lam and K. S. Barber, University of Texas at Austin

Given the diverse multi-agent system (MAS) implementations developed for various domains, there has been a lack of a comprehensive method for analyzing and evaluating the assortment of multi-agent architectures and technology resident in those architectures. This research proposes that the first step in providing a formal analysis method is to decompose an agent into its core competencies (CC), which define the major functionalities of an agent. For each CC, the designer chooses to implement a core competency strategy from a library of existing strategies. Inherent dependencies exist between CCs and the selected strategies must serve to satisfy those dependencies. The designer must recognize that selection of one strategy can constrain or eliminate the consideration of strategies for other needed CCs. With a fundamental understanding of how and why certain combinations of strategies produce specific agent-level and system-level behaviors, a method to analyze agents at an abstract level can be developed for MAS designers to investigate the system-level implications of selecting strategies that have dependencies spanning across multiple CCs and across agents resident in the system.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.