An Evolutionary Approach to Automatic Construction of the Structure in Hierarchical Reinforcement Learning

Stefan Elfwing^1,2,3, Eiji Uchibe^2,3, and Kenji Doya^2,3

¹KTH, Numerical Analysis and Computer Science department,
KTH, Nada,
100 44 Stockholm, Sweden

²ATR, Human Information Science Laboratories, Department 3

³CREST,
Japan Science and Technology Corporation
2-2-2 Hikaridai,
"Keihanna Science City" Kyoto 619-0288, Japan

Abstract. Hierarchical reinforcement learning (RL) methods have been developed to cope with large scale problems. However, in most hierarchical RL methods, an appropriate structure of hierarchy has to be hand-coded. This paper presents an evolutionary approach for automatic construction of hierarchical structures in RL.

LNCS 2723, p. 507 ff.

Full article in PDF

lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2003