Engineering Mechanics

Proceedings Vol. 9 (2003)

ENGINEERING MECHANICS 2003

Copyright © 2003 Institute of Theoretical and Applied Mechanics, Academy of Sciences of the Czech Republic, v.v.i., Prague

 

Věchet S., Miček P., Březina T.
pages 368 - +6p., full text

Locally Weighted Learning (LWR) is a class of approximations, based on a local model. In this paper we demonstrate using LWR together with Q-learning for control tasks. Q-learning is the most effective and popular algorithm which belongs to the Reinforcement Learning algorithms group. This algorithm works with rewards and penalties. The most common representation of Q-function is the table. The table must be replaced by suitable approximator if use of continuous states is required. LWR is one of possible approximators. To get the first impression on application of LWR together with modified Q-learning for the control task a simple model of inverted pendulum was created and proposed method was applied on this model.

Text and facts may be copied and used freely, but credit should be given to these Proceedings.

All papers were reviewed by members of the scientific committee.