In hard- and software we currently observe technological breakthroughs
towards cognitive agents, which will soon incorporate a mixture of
miniaturized sensors, cameras, multi-DOF robots, and large data
storage, together with sophisticated artificial cognitive
functions. Such technologies might culminate in the widespread
application of humanoid robots for entertainment and house-care, in
health-care assistant systems, or advanced human-computer interfaces
for multi-modal navigation in high-dimensional data spaces. Making
such technologies easily accessible for every day use is essential for
their acceptance by users and customers. At all levels for such
systems learning will be an essential ingredient to meet the
challenges in engineering, system development, and system integration
and neural network methods are of crucial importance in this arena.
Cognitive robots are meant to behave in the real world and to interact
smoothly with their users and the environment. While off-line
learning is well established to implement basis modules of such
systems and many learning methods work well in toy domains, in
concrete scenarios on-line adaptivity is necessary in many respects:
in order to cope with the unevitable uncertainties of the real world,
the limited predictability of the interaction structure, to acquire
new and enhance preprogrammed behavior. Online-learning is also the
main methodological ingredient in the developmental approach to
intelligent robotics, which aims at incremental progressing from
simple to more and more complex behavior.
Therefore in this seminar we will treat a number of current approaches
to implement online-learning on real robots including domains as
cognitive vision (eg. visual object learning, acquisition of visual
memory, adaptive scene analysis),
localization and map building in mobile robots,
online trajectory learning and acquisition,
adaptive control of multi-DOF robots,
learning in behavioral architectures, and
learning by demonstration and imitation.
Talks may be given in English or German. Basic knowledge in
Robotics and Neural Networkso is useful.
Students who apply until 01.10. get the opportunity to select a paper within
the scope of the seminar on their own. All others get a paper for a talk in the
first session on 18.10.
Further information can be found at