- author = "Sean Luke and Charles Hohn and Jonathan Farris and Gary Jackson and James Hendler",
- title = "Co-evolving Soccer Softbot Team Coordination with Genetic Programming",
- booktitle = "Proceedings of the First International Workshop on RoboCup, at the International Joint Conference on Artificial Intelligence",
- year = "1997",
- address = "Nagoya, Japan",
- keywords = "genetic algorithms, genetic programming, lil-gp",
-
URL = "
http://www.cs.gmu.edu/~sean/papers/robocupc.pdf",
-
URL = "
http://www.cs.gmu.edu/~sean/papers/robocupc.ps.gz",
- size = "14 pages",
- abstract = "Genetic Programming is a promising new method for automatically generating functions and algorithms through natural selection. In contrast to other learning methods, Genetic Programming's automatic programming makes it a natural approach for developing algorithmic robot behaviors. In this paper we present an overview of how we apply Genetic Programming to behavior-based team coordination in the RoboCup Soccer Server domain. The result is not just a hand-coded soccer algorithm, but a team of softbots which have learned on their own how to play a reasonable game of soccer.",
-
notes = "IJCAI-97
https://www.robocup.org/a_brief_history_of_robocup
Given the acknowledged challenges of applying Genetic Programming to robot soccer, we were happy to just show up at Nagoya with an entry in the RoboCup simulation track. However, Maryland's Genetic Programming entry in in fact beat its first two competitors (5-2 against U British Columbia, Canada and 17-0 over Toyohashi University of Science and Technology, Japan) before losing to University of Tokyo (last year's champion, 6-1) and subsequently Tokyo Institute of Technology (16-4) in the single-elimination round. For its research achievement in demonstrating the feasibility of evolutionary computation in a very difficult domain, Maryland's entry also won the RoboCup Scientific Challenge Award. http://ci.etl.go.jp/~noda/soccer/RoboCup97/result.html Part of Email from John Koza Fri, 29 Aug 1997 21:37:50 PDT to genetic-programming@cs.stanford.edu {"}The Maryland entry competed against various hand-written robot controllers (all of which are very good examples of clever human programming) and its success demonstrated, I think, that GP is precisely the right way to create programmers when the task really gets difficult. {"}
Too short to give full technical details: STGP, 50ish problem dependant functions. team composed of 2-3 squads of identical players. Each squad 2 trees (used for possetion and non-possetion of ball. 6 or 12 trees per GP indivdual. Co-evolution. lil-gp. Stepped evolution (like seeding?) build squad from good players, team from good squads.",
