Maze Problem with the conventional reinforcement learning method
(help[japanese])
First, make an agent learn a route to the goal.
Secondly, choice emergency 0 or 1 as a new wall.
In this conventional method, the agent relearns very solowly.
(Compare with the proposed method.)
(If this program is not executed, please install Java VM on here or here.)
Copyright(c) Ryoji Ino and Kazushi Murakoshi. All rights reserved.
Based on Hajime Kimura.
Last Modified: 27 December, 2007