Transcript DesaiMentorshipPoste..

An Implementation of Artificial Physics Using AIBO
Robots and the Pyro Programming Environment
Ankur Desai
Introduction
One major problem that researchers face
when experimenting with new forms of artificial
intelligence is the vast proliferation of research
platforms. While many computer scientists
prefer to use very complex and powerful
programming languages when designing new
applications, those who specialize in robotics
tend to favor simpler and more user-friendly
languages. In addition, in order to control a
specific type of robot, it is often necessary to
program in specific languages. In research
today, adapting a program to a certain robot is a
very time consuming task that frequently
inhibits the overall pace of research and
development.
The purpose of this project was to
determine the effectiveness of the Sony AIBO
robot as a platform for future testing of the
artificial physics algorithm, and to create an
artificial physics Python module that could be
used to control the AIBO robots.
Fig. 1.
Sony AIBO
Procedures
The first step in this project was to create a
working Python program that mirrored the
preexisting artificial physics C library.
This
process was facilitated by the use of SWIG, an
open-source tool that automatically generates
portions of code. Once this wrapper was created,
a compiler was used to build a dynamic library
that linked to the wrapper code and all necessary
object files from the C library. This process was
done using the GCC compiler. The end product
of this process was a file that could be imported
into Pyro as a robot “brain.”
The next goal of the project was to
determine whether the AIBO could be effective as
a platform for artificial physics. The AIBO keeps
track of its location using odometry data.
Therefore, it was necessary to test this data from
the AIBO.
The AIBO was first instructed to move
forward 10 meters. The dependent variable was
the actual distance that the AIBO traveled. This
test was repeated ten times using each gait. The
AIBO was then instructed to make a turn of 360°,
and the actual number of degrees that the robot
turned was measured. Again, the test was
repeated ten times.
(Image from
http://www.cifrovik.ru/cifr
ovik/services/catalog/ima
ges/14077-photomedium.jpg)
Discussion
Results
In the second, third, fifth, and tenth trials of
the walking test, the measured values missed
the expected values by a wide margin, with
greater than 35% error. In the other trials, the
error was less than 7.0%. These errors in
odometry data were due to slippage. Having
four legs, it is very easy for the robot to place
too much weight on one leg, causing it to slip.
During each of the trials in which the odometry
data shows an error of greater than 10%, the
robot slipped during the course of motion.
When this slippage occurred, the Pyro
simulator was unable to compensate, and the
odometry data became unreliable. In straight
line trials, the robot slipped 60% of the time
while walking and 40% of the time while
crawling. In turning trials, the robot slipped 30%
of the time while walking and 60% of the time
while crawling.
Conclusion
The results of this experiment showed that
the AIBO is in fact not a suitable platform for
artificial physics. The other portions of this
project were more successful, as the Python
module functioned correctly on multiple
platforms.
Literature Cited
Fig 2.
Grid formation and resource protection,
possible applications of artificial physics.
(Spears, 1999)
Blank, D., Meeden, L., & Kumar, D. (2003). Python robotics: An environment for exploring
robotics beyond LEGOs. SIGSCE ’03, 35, 317-3121.
Ikemoto, Y., Hasegawa, Y., Fukuda, T., & Matsuda, K. (2005). Gradual spatial pattern formation
of homogeneous robot group. Information Sciences, 171, 431-445.
Lee, M. (2003). Evolution of behaviors in autonomous robot using artificial neural network and
genetic algorithm. Information Sciences, 155, 43-60.
Oliveira, E., Fischer, K., & Stepankova, O. (1999). Multi-agent systems: Which research for
which applications. Robotics and Autonomous Systems, 27, 91-106.
Röfer, T., & Jüngel, M. (2003). Fast and robust edge-based localization in the Sony four-legged
robot league. In Polani, D., Browning, B., Bonarini, A., & Yoshida, K. (Eds.), RoboCup
2003: Robot soccer world cup VII (pp. 262-273). Berlin: Springer.
Spears, W. M., & Gordon, D. F. (1999). Using artificial physics to control agents. 1999
International Conference on Information Intelligence and Systems, 1999, 281-288.
Tira-Thompson, E. J., Halelamien, N. S., Wales, J. J., & Touretzky, D. S. (2004). Tekkotsu:
Cognitive robotics on the Sony AIBO. Proceedings of the Sixth International
Conference on Cognitive Modeling, 6, 390-391.