Maze Problem with the reinforcement learning method proposed by Murakoshi and Mizuno(2004)

Maze Problem with the reinforcement learning method proposed by Murakoshi and Mizuno(2004) (help[japanese])

First, make an agent learn a route to the goal.
Secondly, choice emergency 0 or 1 as a new wall.
This method appropriately controls three learning parameters.
(Compare with the conventional method.)

(If this program is not executed, please install Java VM on here or here.)