Multi-agent Robot Learning by Means of Genetic Programming : Solving an Escape Problem

Created by W.Langdon from gp-bibliography.bib Revision:1.2031

@InProceedings{Yanai:2001:MAR,
  author =       "Kohsuke Yanai and Hitoshi Iba",
  title =        "Multi-agent Robot Learning by Means of Genetic
                 Programming : Solving an Escape Problem",
  booktitle =    "Evolvable Systems: From Biology to Hardware: 4th
                 International Conference, ICES 2001",
  year =         "2001",
  editor =       "Yong Liu and Kiyoshi Tanaka and Masaya Iwata and 
                 Tetsuya Higuchi and Moritoshi Yasunaga",
  volume =       "2210",
  series =       "LNCS",
  pages =        "192--203",
  address =      "Tokyo, Japan",
  month =        "3-5 " # oct,
  publisher =    "Springer-Verlag",
  keywords =     "genetic algorithms, genetic programming, bloat",
  ISBN =         "3-540-42671-X",
  ISSN =         "0302-9743",
  URL =          "http://www.springerlink.com/openurl.asp?genre=article&issn=0302-9743&volume=2210&spage=192",
  abstract =     "We present the emergence of the cooperative behaviour
                 for multiple robot agents by means of Genetic
                 Programming (GP). For this purpose, we use several
                 extended mechanisms of GP, i.e., (1) a co-evolutionary
                 breeding strategy, (2) a controlling strategy of
                 introns, which are non-executed code segments dependent
                 upon the situation, and (3) a subroutine discovery
                 technique. Our experimental domain is an escape
                 problem. We have chosen the actual experimental
                 settings so as to be close to a real world as much as
                 possible. The validness of our approach is discussed
                 with comparative experiments using other methods, i.e.,
                 Q-learning and Neural networks, which shows the
                 superiority of GP-based multi-agent learning.",
  notes =        "CODEN = LNCSD9

                 Subroutine discovery, ADF, placed in competitive shared
                 library. Escape problem turns out to be three Khepara
                 mini-robots {"}pushing{"} all 3 buttons before going to
                 exit. Buttons, exit etc all colour coded. GP Evolved in
                 simulation but works on real robots.

                 Second problem simplified so can try Q-learning on it.
                 \cite{Iba:1998:ISJ}.",
}

Genetic Programming entries for Kohsuke Yanai Hitoshi Iba