|
PURPOSIVE VISUAL PERCEPTION, CO-OPERATIVE BEHAVIOUR, AND CREATIVITY:
some design principles for Autonomous Agents Playing Soccer |
Giovanni Adorni
Depatment of Computer Engineering
University of Parma
Italy
e-mail: bambi@CE.UniPR.IT
A large number of researchers considered visual perception as a recovery
problem,
that is, as the problem of reconstructing a precise segmentation of the real
world and its properties
from image cue such as contours, shading, colour, texture, stereo, motion, etc.
This approach has influenced many theoretical results in the field and has led
to new mathematical techniques.
Why do human beings/animals have visual perception capability and why do we
want to understand it?
The answer is, of course, that we need vision in order to accomplish every day
visual perception tasks.
Why do robots need visual perception capability?
In the world of robots, visual perception is needed to make them capable of
performing various tasks
while interacting and (possibly) co-operating each other and with their
environment. However, recovering the operating environment and its attributes
is not a necessary condition for
for the accomplishment of visual perception task. Many such tasks can
be achieved visually without reconstruction but through the recognition of
patterns, objects,
situations. What to recognise is directly related to the purpose of the
robot action(s).
The management of interaction and co-operation among different robots
(physical agents in the following),
acting in the same environment to accomplish purposive actions
through visual information is one of the major challenges facing
Artificial Intelligence and Robotics.
A physical agent executing a purposive action in an environment where other
physical agents act, has to co-ordinate its actions with the actions of the
other agents.
Co-ordination is necessary to detect and solve potential conflicts between
agents actions
through the exchange and analysis of information concerning plans and goals of
the involved physical agents. Explicit communication of this information is
not always possible and so,
physical agents must use some other means by which to gather the
necessary information regarding other agents plans. In these cases, plan
recognition
is usually used. However, both explicit communication and plan recognition may
be very costly and may be used with difficulty to solve conflicts between
robots and humans (which can be
in the same environment). Therefore, it would be productive to find
other adaptive fast mechanisms that can be used by robots to detect and solve
certain classes of conflict with other physical agents.
In this talk, I will discuss the previous issues through the paradigm of the soccer game, where a team of multiple fast-moving autonomous physical agents play/compete in a dynamic, non deterministic, and adversarial environment.
e-mail: hmp@ipvvis.unipv.it