Perception-Action Coupling via Imitation and Attention
Abstract
When we interact with objects, we give them meaning, i.e. we
show what they are potentially useful for. We believe that physical
entities are anchored to perceptual representations, and through
them to the actions that they `afford'. This paper brings an
imitation mechanism and an attention system together computationally,
with the aim to have a system that is capable of creating and
maintaining these anchors. The integrated system is implemented on two
different platforms: a simulated humanoid robot learning from another
how to drink a glass of beer, and a simulated mobile robot learning
from another how to follow walls.
The preprint is available as gzipped
postscript, or pdf.
Alternatively, you can request a copy by
e-mailing me.