Multi-Agent Systems

The characteristics of MASs are that (1) each agent has incomplete information or capabilities for solving the problem and, thus, has a limited viewpoint; (2) there is no system global control; (3) data are decentralized; and (4) computation is asynchronous. - Katia Sycar

The study of multiagent systems (MAS) focuses on systems in which many intelligent agents interact with each other. The agents are considered to be autonomous entities, such as software programs or robots. Their interactions can be either cooperative or selfish. That is, the agents can share a common goal (e.g. an ant colony), or they can pursue their own interests (as in the free market economy).

MAS researchers develop communications languages, interaction protocols, and agent architectures that facilitate the development of multiagent systems. For example, a MAS researcher can tell you how to program each ant in a colony in order to get them all to bring food to the nest in the most efficient manner, or how to set up rules so that a group of selfish agents will work together to accomplish a given task. MAS researchers draw on ideas from many disciplines outside of AI, including biology, sociology, economics, organization and management science, complex systems, and philosophy.

Good Places to Start

The Smartest Agents Will Learn to Be Team Players. By Christopher Locke. Red Herring (January 9, 2002). "An example of how the system would work is a crisis situation -- say, planning the evacuation of American citizens from a foreign city during a conflict. In this scenario, each agent in a multi-agent team would be assigned a specific task, like plotting an evacuation route, forecasting the weather, or laying out a flight plan. All of the agents would cooperate and gather data to determine the optimal choices."

MultiAgent Systems. By Katia Sycara (1998). AI Magazine 19(2). Discusses the need for multiple agent systems communicating peer-to-peer.

Multiagent Systems: An Emerging Subdiscipline of AI. Victor R. Lesser (1995). ACM Computing Surveys, 27 (3): 340 -342. "As more AI applications are being formulated in terms of spatially, functionally, or temporally distributed processing, multiagent systems (or what was previously called distributed AI) are emerging as an important subdiscipline of AI. ... In general multiagent systems are computational systems in which several semi-autonomous agents interact or work together to perform some set of tasks or satisfy some set of goals."

Dynamics of Multiagent Systems. From the Xerox Palo Alto Research Center. "Multiagent systems arise in human societies, biological ecosystems, the immune system and distributed computation. While very different in detail they all face the issue of producing complex global behavior through the local interactions of their constituent parts. This is particularly problematic since the individual parts have only a limited view of the system as a whole." Topics include Dynamics of Cooperation in Societies, Computational Societies and Economies, and Agent-Based Control of Smart Matter, and they offer lots of papers for those who want to explore these areas in depth.

MultiAgent Systems. Maintained by Jose Vidal, professor in Electronic and Computer Engineering, University of South Carolina. A wealth of links to a variety of resources.

Standardization of Multi-Agent System Frameworks. By Roberto A. Flores-Mendez. Well-written, basic overview.

To find out what Pandemonium & Demons have to do with MAS, see our Namesakes page.

Readings Online

Journal of Artificial Societies and Social Simulation, "an inter-disciplinary journal for the exploration and understanding of social proceses by means of computer simulation."

SIGMA: Multiagent Learning and Adaptation in an Information Filtering Market. By Innes A. Ferguson and Grigoris J. Karakoulas. The Interactive Information Group at the Institute for Information Technology, National Research Council, Ottawa ON. "By augmenting a particular collection of these IF agents with learning and adaptation techniques - including reinforcement learning, bidding price adjustment, and relevance feedback - we have created a robust network-based application which adapts to both changes in the characteristics of the information available on the network, as well as to changes in individual users' information delivery requirements."

Collaborative Systems: 1994 AAAI Presidential Address. By Barbara Grosz. 1996. AI Magazine 17(2), 67-85.

Tutorial on Intelligent Agents and Multiagent Systems. By Vasant Honavar, Department of Computer Science, Iowa State University.

Game Theory. Daphne Koller's article for the MIT Encyclopedia of Cognitive Science. "Game theory is a mathematical framework designed for analyzing the interaction between several agents whose decisions affect each other. In a game-theoretic analysis, an interactive situation is described as a game: an abstract description of the players (agents), the courses of actions available to them, and their preferences over the possible outcomes." In addition to her overview, you'll find references, readings, and links to other resources.

Software agents ask for help. By Kimberly Patch, Technology Research News (September 18/25, 2002). "If you're good at something, people naturally ask your advice about it. Researchers from the University of Porto in Portugal are tapping this learning strategy by programming tiny bits of software, called agents, to ask other agents for help as the group figures out how to control the timing of traffic lights."

On the Backs of Ants - New networks mimic the behavior of insects and bacteria. By Kimberly Patch. Technology Review (March 19, 2003). "Drawing heavily on the chemistry of biology, researchers from Humboldt University in Germany have devised a way for electronic agents to efficiently assemble a network without relying on a central plan."

Calculating Swarms - Ant teamwork suggests models for computing faster and organizing better. By Ivars Peterson. Science News, Vol. 158, No. 20 (November 11, 2000). "In effect, astonishing feats of teamwork emerge from a large number of unsupervised individuals following a few simple rules. This sort of self-organizing cooperative behavior among ants, bees, and other social insects has become the envy of engineers and computer scientists as they work to solve tough path-finding, scheduling, and control problems in industrial and other settings. In recent years, studies of ant behavior have suggested powerful computational methods for finding alternative traffic routes over congested telephone lines and novel algorithms for governing how robots operating independently would work together."

Seeing Around Corners. By Jonathan Rauch. The Atlantic (April 2002). "The new science of artificial societies suggests that real ones are both more predictable and more surprising than we thought. Growing long-vanished civilizations and modern-day genocides on computers will probably never enable us to foresee the future in detail -- but we might learn to anticipate the kinds of events that lie ahead, and where to look for interventions that might work."

Multiagent Systems: A Survey from a Machine Learning Perspective. By Peter Stone and Manuela Veloso, Computer Science Department, Carnegie Mellon University. "Distributed Artificial Intelligence (DAI) has existed as a subfield of AI for less than two decades. DAI is concerned with systems that consist of multiple independent entities that interact in a domain. Traditionally, DAI has been divided into two sub-disciplines: Distributed Problem Solving (DPS) focusses on the information management aspects of systems with several branches working together towards a common goal; Multiagent Systems (MAS) deals with behavior management in collections of several independent entities, or agents. This survey of MAS is intended to serve as an introduction to the field and as an organizational framework."

Related Web Sites

ACE. "Agent-based computational economics (ACE) is the computational study of economies modelled as evolving systems of autonomous interacting agents. ACE is thus a specialization to economics of the basic complex adaptive systems paradigm." Maintained by Leigh Tesfatsion, Department of Economics, Iowa State University. The site offers links to a variety of helpful resources.

Agent Lab Homepage. "The goal of the AFIT Lab is to bring together researchers with various expertise to solve interesting problems in the area of multi-agent systems development and agent security."

CASOS. Computational Analysis of Social and Organizational Systems. "Gourps, organizations, and societies are inherently computational and computational multi-agent systems are inherently organizational. Thus, within CASOS we attempt to understand and formally model two distinct but complimentary types of phenomena. The first is the natural or human group, organizational or society, which is universally informatted and continually acquires, manipulates, and produces information (and possibly other material goods) through the joint, and interlocked activities of people and automated information technologies. The second is the artificial computational systems which is generally comprised of multiple distributed agents who can mutually influence, constrain and suppurt each other as they try to manage and manipulate the knowledge, communication and interaction networks in which they are embedded. Computational analysis is used to develop a better understanding of the fundamental priciples of organizing, coordinating, and managing multiple information processing agents (whether they are human, WebBot, or robots) and the fundamental dynamic nature of groups, organizations and societies." - from their Mission Statement

CORO. Collective Robotics Research Group at the Center for Neuromorphic Systems Engineering at the California Institute of Technology. "The use of multiple mobile robots offers significant advantages over the use of single mobile robots: key features are the possibility of distributed sensing, distributed action, task dependent reconfigurability, and the enabling of robustness and system reliability through redundancy. There are several approaches for controlling systems of multiple mobile robots: most of them, rooted in the conventional artificial intelligence paradigm, base on complex robots that build and maintain internal models relevant to the task, and communicate only explicitly with other robots and external agents including humans. In the CORO group, we focus instead on the application of Swarm Intelligence principles, where local interactions among robots and between robots and environment play a crucial role for achieving the required task."

Control of Agent Based Systems. Defense Advanced Research Projects Agency (DARPA), U.S. Department of Defense. Overviews and detailed descriptions of military research projects involving agents.

Distributed Artificial Intelligence links from the Department of Sociology, University of Surrey.

The Flocking Robots Project at the Artificial Intelligence Laboratory, Department of Information Technology, University of Zurich. "Flocking adresses a variety of important topics in the field of multiagent simulation and collective robotics which include agent interaction, kin recognition, and finally the emergence of collective behavior." And their flocking applet is simply beautiful!

  • For related information, also see: Boids

International Network for Social Network Analysis (INSNA)

Learning in Multi-Agent Systems: Webliography. By M. V. Nagendra Prasad (University of Massachusetts, Amherst) and Thomas Haynes (University of Tulsa), Links to research sites, projects, conferences, journals and more.

Multi-Agent Systems at The Intell1igent Software Agents Group, Robotics Institute, Carnegie Mellon University. Be sure to scroll down their page to their collection of "Applications of Multi-Agent Research."

The Multi-Agent Systems Laboratory at the Department of Computer Science at the University of Massachusetts at Amherst. "[This lab] is concerned with the development and analysis of sophisticated AI problem-solving and control architectures for both single-agent and multiple-agent systems. The laboratory has pioneered work in the development of the blackboard architecture, approximate processing for use in control and real-time AI, and a wide variety of techniques for coordination of multiple agents."

Open Agent Architecture - A framework for integrating a community of heterogeneous software agents in a distributed environment. From SRI's Artificial Intelligence Center.

  • Videos of OAA are also available from SRI. Just scroll down to the "Planning and Simulation" section in their Video Archive.

Swarm Development Group "Swarm is a software package for multi-agent simulation of complex systems, originally developed at the Santa Fe Institute. Swarm is intended to be a useful tool for researchers in a variety of disciplines. The basic architecture of Swarm is the simulation of collections of concurrently interacting agents: with this architecture, we can implement a large variety of agent based models." Be sure to check out the demos, tutorial, and the other resources they offer.

Related Pages

More Readings

Durfee, Edmund H. 1992. What Your Computer Really Needs to Know You Learned in Kindergarten. In Proceedings of the 10th National Conference on Artificial Intelligence, 858-864. San Jose, CA: AAAI Press.

Ferber, Jacques. 1998. Multi-Agent Systems: Towards a Collective Intelligence. Reading, MA: Addison-Wesley.

Lesser, Victor., editor. 1995. Proceedings of the First International Conference on Multiagent Systems. Menlo Park, CA: AAAI Press.

Tokoro, Mario, editor. 1996. Proceedings of the Second International Conference on Multiagent Systems. Menlo Park, CA: AAAI Press. Topics cover coordination, distributed planning, implementing multi-agent systems, market-oriented approaches, multiagent applications, multiagent learning, multiagent search, mutual knowledge, negotiation, organizational aspects, real-world agents, situated agents, sociability, and teams of agents.

Fisika & Psichica