hide
Free keywords:
-
Abstract:
Autonomous, cognitive agents, e.g. mobile robots, have become more and more important through the last years. In order to obtain a high level of autonomy, an agent has to be able
to interact with objects even though it might not have seen this kind of object before. Here,
categorizing objects by its functionality seems to be the clue.
In this thesis, we introduce briefly the idea of functional object category detection in
the context of human-object interaction. We present a system which extracts visual shape
descriptions (affordance cues) of object parts that are characteristic for a certain task based
on observation of a prototypical interaction. The system is implemented and integrated into
a cognitive agent framework which allows cooperation with e.g. manipulation systems. We
show that the system has the applicability to extract affordance cues for different grasping
techniques on different objects but also highlight restrictions of the system w.r.t. the scenery
where the interaction is observed.
In cooperation with other reseach groups, we were able use the detected affordance
cues to detect objects in cluttered scenes and as input for manipulation.
