Presentation 1998/10/24
Multiple Model-based reinforcement learning for Non-linear control
Kenichi Katagiri, Kenji Doya, Mitsuo Kawato,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Reinforcement learning architecture can learn to accomplish a given task in an unknown environment. However reinforcement learning architecture can not easily deal with non-stationary, non-linear systems. In this study we propose a multiple models-based reinforcement learning (MMRL) architecture in which pairs of forward-models and reinforcement learning modules are switched or combined using the softmax function of the prediction errors. We performed a simulation of the task of swinging up a pendulum. The result indicates that MMRL can accomplish a highly non-linear control task.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) reinforcement learning / multiple models / motor control / swing up
Paper # NC98-46
Date of Issue

Conference Information
Committee NC
Conference Date 1998/10/24(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Multiple Model-based reinforcement learning for Non-linear control
Sub Title (in English)
Keyword(1) reinforcement learning
Keyword(2) multiple models
Keyword(3) motor control
Keyword(4) swing up
1st Author's Name Kenichi Katagiri
1st Author's Affiliation Graduate School of Information Science, Nara Institute of Science and Technology:ART Human Information Processing Res. Labs.()
2nd Author's Name Kenji Doya
2nd Author's Affiliation Kawato Dynamic Brain Project, JST:Graduate School of Information Science, Nara Institute of Science and Technology
3rd Author's Name Mitsuo Kawato
3rd Author's Affiliation ATR Human Information Processing Res. Labs.:Kawato Dynamic Brain Project, JST
Date 1998/10/24
Paper # NC98-46
Volume (vol) vol.98
Number (no) 365
Page pp.pp.-
#Pages 8
Date of Issue