Abstract
This research has lead us to show that it is possible, for reactive systems, to learn how to solve complex tasks. The task proposed in the “blocks world”, considering the initial set of actions the system knows, is not currently resolvable by any other direct learning method. The success of our proposal is due to the use of a learning mechanism robust to ambiguous information, that can improve the abilities of the system, learning new behaviors to solve general tasks.
Chapter PDF
Similar content being viewed by others
Keywords
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.
References
D. Chapman. Penguins can make cake. AI Magazine, 10:45–50, winter 1989.
L.P. Kaelbling. Learning in Embedded Systems. MIT Press, 1993.
M. Martin. Learning to solve complex tasks by reinforcement: A new algorithm. Technical report, Universitat Politècnica de Catalunya, 1995.
R.S. Sutton. Learning to predict by the methods of temporal differences. Machine Learning, 3(1):9–44, 1988.
C. Watkins. Learning from delayed rewards. PhD thesis, Cambridge Univ., 1989.
S.D. Whitehead and D.H. Ballard. Learning to percive and act by trial and error. Machine Learning, 7:45–83, 1991.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1995 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Martin, M., Cortés, U. (1995). Learning to solve complex tasks for reactive systems (Extended abstract). In: Lavrac, N., Wrobel, S. (eds) Machine Learning: ECML-95. ECML 1995. Lecture Notes in Computer Science, vol 912. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-59286-5_76
Download citation
DOI: https://doi.org/10.1007/3-540-59286-5_76
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-59286-0
Online ISBN: 978-3-540-49232-0
eBook Packages: Springer Book Archive