Presentation 1998/3/19
Modular Reinforcement Learning
Satoshi Yamada,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We propose a modular reinforcement learning system which contains control modules and a selection module. The selection module selects an appropriate control module dependent on states. Both the control modules and the selection module are trained by Q-learning. The modular reinforcement learning was applied to the navigation and target-collection control of Khepera robot. Inputs of each control module are part of whole inputs, and inputs of the selection module are maximum or minimum of Q-values calculated in the control modules. The modular reinforcement learning system learned the navigation and target-collection control faster than reinforcement learning with a single network because it reduced the searching space for reinforcement learning.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) modular reinforcement learning / Khepera robot / selection module / control module
Paper #
Date of Issue

Conference Information
Committee NC
Conference Date 1998/3/19(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) Modular Reinforcement Learning
Sub Title (in English)
Keyword(1) modular reinforcement learning
Keyword(2) Khepera robot
Keyword(3) selection module
Keyword(4) control module
1st Author's Name Satoshi Yamada
1st Author's Affiliation Advanced Technology R&D Center, Mitsubishi Electric Corporation()
Date 1998/3/19
Paper #
Volume (vol) vol.97
Number (no) 623
Page pp.pp.-
#Pages 8
Date of Issue