Multi-agent system

from Wikipedia, the free encyclopedia

A multi -agent system or MAS is a system of several similar or differently specialized acting units, software agents , that collectively solve a problem.

Multi-agent systems exist both in biology (natural multi-agent systems) and in technology. Ant colonies represent a family of examples of biological multi-agent systems. Some of the algorithms ( ant algorithms ) running in ant colonies represent heuristic solution processes for complex optimization tasks and, in addition to their fundamental interest in theoretical biology, are also a model for optimizing technical processes. One also speaks of distributed , in the field of technology of artificial intelligence or DAI ( distributed artificial intelligence ).

Outside of Europe, especially in the USA, the term agent-based modeling or simulation (ABM) has become established for MAS. The term is also used for a special kind of knowledge logic . In knowledge logic, the carriers of the knowledge modeled in each case (e.g. people, players, processors) are called "agents". It should be noted that systems of mobile agents ( mobile agent systems ) are often also abbreviated to MAS . Mobile agents are software agents for which the transfer of execution to other nodes in a network is of particular importance.

Multi-agent systems in technology

Multi-agent systems are one possible way of developing properties and the architecture of agent arrangements. Generally speaking, they are a research area of ​​Distributed Artificial Intelligence that deals with how autonomous, distributed and "intelligent" systems as a unit coordinate their specific knowledge, goals, skills and plans in order to act in a coordinated manner or to solve problems.

There are e.g. B. collaborative agent systems that are deliberately distributed in their architecture in order to perform tasks more flexibly and reliably than a single, local system. Based on the interaction of the actuators, it can solve a problem that a single unit would never be able to do. These units are each responsible for an activity, a superordinate control is not necessary and they find solutions together through self-organized coordination.

Brief explanation of the agent types:

  • Collaborative agents: Objectives are achieved through cooperation and negotiations with other agents, usually set up as multi-agent systems. Cooperativity and autonomy are in the foreground, but they are often also self-learning.
  • Interface agents: mostly communicate with a human being as the system user. Also on the Internet help to find offers and to negotiate specialized deals. Help assistants are not part of these agents.
  • Smart agents: have all the properties and are able to cope with a wide variety of tasks.

The main types of communication in multi-agent systems are:

  • Point-to-point
  • Broadcast
  • announcement
  • signal

The most common form of message used in dynamic networks is broadcasting , so that all agents receive each message and then decide whether and how to take action.

Cooperation and coordination

Cooperation between the agents is necessary due to the fact that the overall problem cannot be solved by a single agent. In the case of different tasks, the actions must be coordinated and planned in their time sequence in order to remain effective and efficient. If a task is accepted by an agent, the agent must immediately return this information so that several agents do not want to do the same thing and in extreme cases hinder each other. If there are waiting times, for example, this must also be taken into account. There are additional tasks that are not directly productive but of an organizational nature and are necessary as soon as several autonomous agents pursue their own goals in a common environment. If, for example, conflicting goals arise, these can be resolved through coordination mechanisms. In particular, automated negotiations between software agents are among the most suitable organizational coordination mechanisms to reduce transaction costs.

PAGE description

Agents can also be described using PAGE (acronym for percepts , actions , goals , environment = perceptual content , actions , goals , environment ).

BDI description

Another characterization uses the acronym BDI , which stands for beliefs , desires and intentions .

Applications

RoboCup , web crawler , production planning and control , software agent , multi-agent simulation

See also

literature

  • Gerhard Weiss (Ed.): Multiagent Systems. A Modern Approach to Distributed Artificial Intelligence . 2. print. MIT-Press, Cambridge MA 2000, ISBN 0-262-73131-2 .
  • Michael Wooldridge: Introduction to MultiAgent Systems . John Wiley and Sons, Chichester 2002, ISBN 0-471-49691-X .
  • Franziska Klügl: Multi-agent simulation , concepts, tools, applications . Addison-Wesley Verlag, Munich et al. 2001, ISBN 3-8273-1790-8 , (At the same time: Würzburg, Univ., Diss., 2001: Activity-based behavior modeling and its support in multi-agent simulations ).
  • Jacques Ferber: Multi-Agent Systems: An Introduction to Distributed Artificial Intelligence . Addison-Wesley, Harlow et al. 1999, ISBN 0-201-36048-9 .
  • Ricardo Büttner : Automated negotiations in multi-agent systems . Gabler-Verlag, Wiesbaden 2010, ISBN 978-3-8349-2131-4 . ( Table of contents )

Web links

Individual evidence

  1. ↑ WOOLDRIDGE AND JENNINGS (1995).
  2. ↑ FERBER (2001).
  3. ↑ Nwana (1996, 216).
  4. ↑ Stefano Albrecht and Peter Stone (2017). Multiagent Learning: Foundations and Recent Trends. Tutorial at IJCAI-17 conference. http://www.cs.utexas.edu/~larg/ijcai17_tutorial/multiagent_learning.pdf
  5. ↑ FERBER (2001, p. 343).
  6. ↑ FERBER (1999, p. 311).
  7. ↑ FERBER (2001, p. 431).
  8. ↑ BUETTNER (2010, p. 53 ff.).