Talk@IDSIA: Dr Krishnanand Kaipa Narasimha - Framework to identify robot peer
Dalle Molle Institute for Artificial Intelligence
Start date: 8 January 2010
End date: 9 January 2010
Sala Primavera, Galleria 2, Manno
Giovedì 21 Gennaio 2010, ore 12:00
Abstract
Humans use oneself as a model to interpret the behavior and experience of others. This talk summarizes how the above theory is instantiated in a social student robot that identifies a good teacher from a set of candidate external social agents using the three phases of self-discovery, teacher modeling, and teacher imitation. In particular, during the first phase of self-discovery, the student robot uses actuation-sensation relationships to infer its own structure. Next, the student uses a novel stereo vision based active learning approach to observe a teacher and creates its 3D visual model. Finally, during the third phase of teacher imitation, the success of finding a neural controller that allows its self model (evolved in the first phase) to imitate the action of the teacher's visual model (evolved in the second phase) determines whether or not the observed agent is a good teacher.