EnglishPrecedentSuivant

Ce document n'est pas traduit, désolé...



Artificial Creatures


Copyright

Author(s)

Institute(s)

Project : ALIVE

  • URL : http://alive.www.media.mit.edu/projects/alive/

    Video(s) and extracted images: 320*240

    Film
    1
    Video QuickTime -> Film/Video (4.9 Mo)
    Jpeg Images -> (12 Ko)

    Film
    2
    Video QuickTime -> Film/Video (4.2 Mo)
    Jpeg Images -> (13 Ko)



    Description





    More Information...


    • Bibliography :

      • "Learning in artificial creatures",Imagina proceedings, 1993, pp248-255
      • "A Bottom-up mechanism for Behavior Selection in an Artificial Creature". Proceedings of the first Intenational Conference on Simulation of Adaptive Behavior, edited by J.A. Meyer and S. Wilson, MIT PRESS. 1991.
      • "How To Do the Right Thin". Connection Science Joumal 1 (3). 1989
      • "Learning to Coordinate Behaviors". Proceedings of the AAAI-90 Conference. 1990.
      • "Learning Behavior Networks from Experience". Proceedings of the first European Artificial Life Conference, Paris, December 1991, MIT Press. 1992.


    • Abstract :

      For the past couple of years I have been developing models of action selection and Iearning in artificial creatures (cfr. publications cited below). In particular I developed a distributed algorithm for motivational competition and selection of behaviors in an artificial creature. This algorithm has particular desirable characteristics, such as that it combines aspects of traditional planners and reactive systems, that is highly distributed, that its action selection characteristics can be tuned, and so on. In more recent work I integrated two learning algorithms in the action selection algorithm. The first algorithm makes a creature adapt its behavior selection policy so as to maximize positive feedback and minimize negative feedback which is generated by some external source. The second algorithm allows a creature to learn about the results of actions and the likelihood of those results being produced. The learning directly alters the creature's action selection behavior in that it becomes more and more effective and efficient at satisfying its motivations (or goals). I propose to give a demo of a creature which me and some students built to illustrate and test these models. In particular, we have been modeling an artificial ~dog~ which lives in a simulated simple: Its behavioral repertoire includes a dozen behaviors. We are making this model more sophisticated every day and hope to demonstrate a more complicated dog (with a finer granularity of behaviors) at the time of the conference. The software is very user-friendly. With a little bit of explanation a user can interact with the artificial dog: chance the dog's motivational levels to see how this affects its behavior, try to teach it a trick by generating positive feedback for certain behaviors and negative for others, etc.




    • Some internal links :

      (oo) Same Author
      (ooo)ALIVE
      (oo) Same Institute
      (ooo)ALIVE
    • Some more Comments :

      Information from a fax sent by Pattie Maes

    EnglishPrecedentSuivant
  • Copyright © 1994-2015 mediaport.net/w3architect.com | Hébergé par p2pweb
    Autres Sites : afromix.org | Actualité Afrique et Caraïbe | Flux d'actualité thématiques | Actualité Européenne