Evolving Keepaway Soccer Players through Task Decomposition

  • Shimon Whiteson
  • Nate Kohl
  • Risto Miikkulainen
  • Peter Stone
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 2723)


In some complex control tasks, learning a direct mapping from an agent’s sensors to its actuators is very difficult. For such tasks, decomposing the problem into more manageable components can make learning feasible. In this paper, we provide a task decomposition, in the form of a decision tree, for one such task. We investigate two different methods of learning the resulting subtasks. The first approach, layered learning, trains each component sequentially in its own training environment, aggressively constraining the search. The second approach, coevolution, learns all the subtasks simultaneously from the same experiences and puts few restrictions on the learning algorithm. We empirically compare these two training methodologies using neuro-evolution, a machine learning algorithm that evolves neural networks. Our experiments, conducted in the domain of simulated robotic soccer keepaway, indicate that neuro-evolution can learn effective behaviors and that the less constrained coevolutionary approach outperforms the sequential approach. These results provide new evidence of coevolution’s utility and suggest that solution spaces should not be over-constrained when supplementing the learning of complex tasks with human knowledge.


Multiagent System Hide Node Machine Learning Algorithm Training Environment Layered Approach 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    T. Balch. Teambots domain: Soccerbots, 2000. Scholar
  2. 2.
    F. Gomez and R. Miikkulainen. Incremental evolution of complex general behavior. Adaptive Behavior, 5:317–342, 1997.CrossRefGoogle Scholar
  3. 3.
    F. Gomez and R. Miikkulainen. Solving non-Markovian control tasks with neuroevolution. Denver, CO, 1999.Google Scholar
  4. 4.
    F. Gomez and R. Miikkulainen. Learning robust nonlinear control with neuroevolution. Technical Report AI01-292, The University of Texas at Austin Department of Computer Sciences, 2001.Google Scholar
  5. 5.
    T. Haynes and S. Sen. Evolving behavioral strategies in predators and prey. In G. Weiß and S. Sen, editors, Adaptation and Learning in Multiagent Systems, pages 113–126. Springer Verlag, Berlin, 1996.Google Scholar
  6. 6.
    W. H. Hsu and S. M. Gustafson. Genetic programming and multi-agent layered learning by reinforcements. In Genetic and Evolutionary Computation Conference, New York, NY, July 2002.Google Scholar
  7. 7.
    I. Noda, H. Matsubara, K. Hiraki, and I. Frank. Soccer server: A tool for research on multiagent systems. Applied Artificial Intelligence, 12:233–250, 1998.CrossRefGoogle Scholar
  8. 8.
    A. D. Pietro, L. While, and L. Barone. Learning in RoboCup keepaway using evolutionary algorithms. In GECCO 2002: Proceedings of the Genetic and Evolutionary Computation Conference, pages 1065–1072, New York, 9–13 July 2002. Morgan Kaufmann Publishers.Google Scholar
  9. 9.
    M. A. Potter and K. A. D. Jong. Cooperative coevolution: An architecture for evolving coadapted subcomponents. Evolutionary Computation, 8:1–29, 2000.CrossRefGoogle Scholar
  10. 10.
    C. D. Rosin and R. K. Belew. Methods for competitive co-evolution: Finding opponents worth beating. In Proceedings of the Sixth International Conference on Genetic Algorithms, pages 373–380, San Mateo, CA, July 1995. Morgan Kaufman.Google Scholar
  11. 11.
    J. D. Schaffer, D. Whitley, and L. J. Eshelman. Combinations of genetic algorithms and neural networks: A survey of the state of the art. In D. Whitley and J. Schaffer, editors, International Workshop on Combinations of Genetic Algorithms and Neural Networks (COGANN-92), pages 1–37. IEEE Computer Society Press, 1992.Google Scholar
  12. 12.
    P. Stone. Layered Learning in Multiagent Systems: A Winning Approach to Robotic Soccer. MIT Press, 2000.Google Scholar
  13. 13.
    P. Stone, (ed.), M. Asada, T. Balch, M. Fujita, G. Kraetzschmar, H. Lund, P. Scerri, S. Tadokoro, and G. Wyeth. Overview of RoboCup-2000. In RoboCup-2000: Robot Soccer World Cup IV. Springer Verlag, Berlin, 2001.Google Scholar
  14. 14.
    P. Stone and R. S. Sutton. Scaling reinforcement learning toward RoboCup soccer. In Proceedings of the Eighteenth International Conference on Machine Learning, pages 537–544. Morgan Kaufmann, San Francisco, CA, 2001.Google Scholar
  15. 15.
    P. Stone and R. S. Sutton. Keepaway soccer: a machine learning testbed. In RoboCup-2001: Robot Soccer World Cup V. Springer Verlag, Berlin, 2002.Google Scholar
  16. 16.
    P. Stone and M. Veloso. Layered learning. In Machine Learning: ECML 2000, pages 369–381. Springer Verlag, Barcelona, Catalonia, Spain, May/June 2000. Proceedings of the Eleventh European Conference on Machine Learning (ECML-2000).CrossRefGoogle Scholar
  17. 17.
    S. Whiteson and P. Stone. Concurrent layered learning. In Second International Joint Conference on Autonomous Agents and Multiagent Systems, July 2003. To appear.Google Scholar
  18. 18.
    C. H. Yong and R. Miikkulainen. Cooperative coevolution of multi-agent systems. Technical Report AI01-287, The University of Texas at Austin Department of Computer Sciences, 2001.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2003

Authors and Affiliations

  • Shimon Whiteson
    • 1
  • Nate Kohl
    • 1
  • Risto Miikkulainen
    • 1
  • Peter Stone
    • 1
  1. 1.Department of Computer SciencesThe University of Texas at AustinAustin

Personalised recommendations