Natural environments composed of many manipulable objects can be
described in terms of probabilistic relational models. Autonomous
learning, exploration an planning in such environments is generally
hard, but can be tackled when exploiting the inherent relational
structure. I will first cover some basic research of our lab in the area
of planning by inference before I address in more detail our recent
advances in relational exploration, learning and planning, with emphasis
on robotics applications. The question of how neurons could do such kind
of ``inference in relational representations'' is rather puzzling to me
- but I conjecture that animals and humans in some way or another have
to do such kinds of computations.
The workshop is sponsored by the
European Neural Networks Society
The workshop is sponsored by the
European Neural Networks Society