|Artificial Intelligence Research Laboratory |
Department of Computer Science
Iowa State University
Coordination and Control of Multi-Agent Systems
Personnel Project Summary Funding Publications Additional Information Projects AI Lab
This research seeks to develop, implement, and evaluate algorithms for coordination and control of multi-agent organizations in the context of distributed knowledge networks consisting of stationary as well as mobile intelligent software agents designed to support the utilization of heterogeneous, distributed data and knowledge sources for automated data-driven knowledge acquisition and decision support.
Modular and open-ended design of distributed knowledge networks implies that the resulting system consists of multiple more or less autonomous intelligent agents with different capabilities. Each agent is responsible for a particular, usually fairly narrowly defined function. Effective use of such agents in distributed problem solving (e.g., in computer-aided scientific discovery in bioinformatics) intrusion detection in distributed computing systems), require mechanisms for control and coordination the behavior of individual agents in a way that leads to the desired global behaviors. In multi-agent systems, the notion of control suggests such functions as coordination among agents, synchronization among multiple agents, activation and deactivation of individual agents or groups of agents, selection among agents, creation of new agents when needed, elimination of agents that are no longer needed, adaptation of individual agents and agent populations to changes in the environments or task demands, learning (both at the individual as well as group levels) from experience, and (at a much slower timescale) evolution of agent populations toward more desirable behaviors. Both natural systems (e.g., cells, brains, immune systems, evolution, groups, social organizations, economies, societies) and artificial systems (computers, multi-computers, computer networks, programs, factories) offer rich sources of examples of a wide variety of coordination and control mechanisms that can be beneficially incorporated into the design of complex information processing systems in general, and multi-agent systems in particular: coordination that emerges from interaction among large number of agents that exhibit relatively simple behaviors inspired by organizations such as the ant colonies; hierarchical control where the flow of control follows the structure of the hierarchy (e.g., in the military); coordination that emerges from interaction (including communication and negotiation) among self-interested agents as exemplified in the contract net protocol and related negotiation mechanisms; control that emerges from competition for resources under the influence of environmental rewards as exemplified by evolutionary processes modeled by genetic algorithms. We have designed, implemented, and analyzed coordination strategies for multi-agent systems in the context of distributed routing by a collection of utility-driven agents in communication networks Recent work in our lab has resulted in a modular and extensible implementation of a general framework for coordination among self-interested rational agents in distributed knowledge networks using the contract net protocol for inter-agent negotiation. Within this framework, each agent can announce tasks, make bids, evaluate bids made by other agents to complete the tasks, and offer contracts. It has been shown that under certain conditions, each contract leads to a task allocation that is more beneficial to the entire society of agents that is involved in the negotiation process. We will build on this work to design, implement, and evaluate different multi-agent organizations and inter-agent coordination mechanisms for interaction among intelligent mobile agents in distributed knowledge networks. Some multi-agent organizations that will be examined include:
Anticipated products of this research include new software tools for coordination and control of multi-agent systems for a variety of applications including: monitoring and control of distributed power systems and computer systems, organizational decision support systems, and distributed knowledge networks for bioinformatics.
This research will be closely integrated with the education and training of graduate and undergraduate students in Computer Science and Bioinformatics and Computational Biology at Iowa State University.
© Vasant Honavar, 1999.