Evolving Keepaway Soccer Players through Task Decomposition. Shimon
Whiteson, Nate Kohl, Risto Miikkulainen,
and Peter Stone. In Proceedings of the Genetic and Evolutionary Computation
Conference 2003, pp. 356–368, July 2003.
GECCO 2003
[PDF]148.2kB [gzipped postscript]69.5kB
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.
@InProceedings{whiteson:gecco03,
author = "Shimon Whiteson and Nate Kohl and Risto Miikkulainen and Peter Stone",
title = "Evolving Keepaway Soccer Players through Task Decomposition",
booktitle = "Proceedings of the Genetic and Evolutionary Computation Conference 2003",
year = "2003",
month = "July",
pages = "356--368",
abstract = {
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.
},
wwwnote = {<a href="http://gal4.ge.uiuc.edu:8080/GECCO-2003/">GECCO 2003</a>},
bib2html_pubtype = {Refereed Conference},
bib2html_rescat = {Machine Learning, Robot Soccer}
}
Generated by bib2html.pl (written by Patrick Riley ) on Sat Aug 22, 2009 06:50:12