Open-R Techno Forum Japan 2003
This event was celebrated in Tokyo Japan on November 29, 2003.
My team was composed only by my wife, my two children and me.
We obtained the second best score out of 21 teams. In the PK competition, a pink ball was located on one of the corners of
a constructed soccer field. AIBO had to search the ball and
the goal locations and then score one goal. Time employed to score the
goal was measured and the team having the shortest time was considered
to win the contest. During my
participation, my robot employed 18.34 seconds to score the goal.
In order to reach the objectives, a program written in
Open-R
environment was developed. At first, the robot is searching
for both, the goal and the ball locations. As soon as finding both
places, the robot is trying to calculate the environment of the ball
position. That is, if the ball is located on the corner or if the
ball is located near one of the lateral white lines. This is reached by
measuring the amount of white pixels located around the ball.
In all cases, the answer behavior of the robot was scheduled to be
different. For example, if the ball is located just on the
corner, the robot is scheduled to move one of its front legs in order
to take out the ball from that place. If the ball is
located a little far from one of the lines, it was scheduled to take
the ball with his hands and then moving its body toward the goal.
During this step, of course, the robot was
scheduled to select the
most appropriate leg to hit the ball. The left or the right
leg is selected after calculating the ball and the goal positions.
The most important part is the way of hitting the ball with its legs.
That is, if the ball is so far from the goal, the ball had to be thrown
employing good power but if the ball is located near the goal, the ball
must be thrown using only a small amount of power. In order
to do this, an artificial neural network was constructed. Almost
the same technique employed for the RoboCup
Japan Open 2003 was used in order to know where the robot is
located respect to the soccer field.
The distances from the bottom of images to the white lines in the
picture was calculated and this information was used as the input of the
neural network. The goal color was defined as blue, so that, the amount
of blue color on images and the center of this color was calculated by
the robot and then the neural network was able to give the amount of
"power" to be employed by its legs in order to make the ball to reach
the goal. After several experiments, a list of 5 different "amount of
powers" was constructed and it was used for the training process.
Power 1 means almost no power and power 5 was defined to have a very big
power.
Power was changed by changing the velocity on the robot legs.
The neural network was constructed using 9 input units, 4 hidden
units and 5 output units as is shown in Fig. 1.
Figure 1. Neural network for calculating the
"amount of power".
I obtained the second best score out of 21
teams. During my participation, my robot Aibo was able to score one goal just 18.34 seconds
after starting the competition.
The time out was setup in 2 minutes.
Videos
of my participation and other teams participation can be seen here.
This site was last updated
07/06/14
|
