Learning to behave in natural environment

Colloquium
Date: 
30 May 2012
Begin time: 
18:00
End time: 
19:30
Room: 
Q2-101

Abstract

Robots and other technical artifacts are clumsy and inflexible when interacting with an ever changing and rather unstructured environment, mostly because they are programmed by humans with little insight into what the world is like for a robot. Drawing on principles of how the brain organizes processing and learning, I will argue for a systems approach to detect, localize and internally preserve behaviorally relevant situations for learning while interacting. One of the arguments is that it is not only the acquisition of knowledge on the environment we should focus on but equally important is the acquisition of knowledge on how to acquire knowledge on the environment, which means how to optimally structure and parameterize the processing architecture in an interaction situation. First steps towards an implementation of this systems concept will be discussed in the context of learning visual representations for driver assist systems.