Robot Learns Fundamental Mathematical and Physical Concepts by Experimentation and Observation

April 20, 2009

ST. AUGUSTIN, Germany, April 20 /PRNewswire/ — Researchers in the
European research project XPERO have developed a machine learning method,
which enables a small humanoid to learn rather fundamental mathematical
concepts such as position and orientation in a coordinate system. The
algorithm takes the robot’s sensor data recorded while it moves through the
surrounding world and creates a model, which allows the robot to predict how
the objects in its vicinity will change their position relative to the robot
when it moves. “What is a trivial thing for a human is a rather difficult
problem for a robot,” explain Jure Zabkar and Ivan Bratko, from Univ. of
Ljubljana, the inventors of the algorithm. Our robot has less knowledge than
a baby. Seeing an object does not mean anything to it. It only perceives
color blobs or edges. It has neither a sense of objects and nor of a position
of an object in a coordinate system and nor how that changes as it moves. The
robot is neither told to learn a coordinate system nor how to learn it nor
what the use of a coordinate system is. We have developed mechanisms, which
allow the robot to extract regularities in its sensor data and to translate
them into models or theories which in turn allow the robot to better explain
and predict what is going on around it. Learning a coordinate system is just
a demonstration of this capability. With the same algorithm we have learned
physical concepts such as “movability” of an object or “degree of freedom”
(number of axes in/around which an object can be moved).

What seems a rather basic research problem, however, has also a
significant technological relevance, claims Erwin Prassler from
Bonn-Rhein-Sieg Univ. in Sankt Augustin, Germany, the coordinator of the
project. The XPERO project lays the first cornerstones for a technology,
which has the potential to become a key technology for the next generation of
so-called service robots, which clean our houses, mow our lawns, or polish
our shoes. Existing products are rather dumb, pre-programmed devices. They
can only perform a single pre-programmed task. They cannot perform any new
tasks or cope with unforeseen operational conditions. Future service robots
will have to be able to learn entirely new concepts and models based on their
existing knowledge and their sensor observations and with this new knowledge
also perform new tasks.

XPERO’s learning robot will be demonstrated during the FET’09 conference
(Future and Emerging Technologies) in Prague, Czech Republic from April
21-23, 2009

    For more information contact:
    Prof. Dr. Erwin Prassler
    Bonn-Rhein-Sieg University of Applied Sciences
    Grantham-Allee 20
    53757 Sankt Augustin
    Email: erwin.prassler@h-brs.de
    Phone: +49-2241-865-257
    Mobile: +49-179-129-1079

    URL: http://www.xpero.org, http://www.ailab.si/xpero/

SOURCE Bonn-Rhein-Sieg Univ. of Applied Sciences

Source: newswire

comments powered by Disqus