Puigbò, Jordi-Ysard and Vouloutsi, Vasiliki and Moulin-Frier, Clément and Verschure, Paul FMJ.
We propose a model integrating a reactive and an adaptive control loops for social learning in the context of humanrobot interaction. We equip a humanoid iCub robot with a set of homeostatic social drives, or needs. Homeostatic drives define a comfort zone (CZ) within a range of sensory data, and elicit reactive actions when outside of the CZ. These needs are related to physical and speech interactions and are modeled as a set of decaying dynamical systems. For example, the need for speech interaction increases with time as long as the robot does not perceive any speech sound from the human. Out of its comfort zone, the reactive actions are synthesized speech sentences and body movements supposed to attract the attention of the human. On top of these reactive homeostatic regulators, an adaptive controller learns from experience how to anticipate actions to avoid the crossing of the comfort zone boundaries. This anticipation is learned from a neurocomputational model of the cerebellum. We show how these mechanisms allow the robot to learn the appropriate timing of its own actions with respect to those of the human in order to maintain the social interaction over the time.
@misc{puigbo2015icdlsocialworkshop,
title = {Reactive and adaptive control loops for social learning in human-robot
interaction},
author = {Puigb\`o, Jordi-Ysard and Vouloutsi, Vasiliki and Moulin-Frier, Cl\'{e}ment and Verschure, Paul FMJ.},
booktitle = {Workshop "Mechanisms of learning in social contexts", IEEE International Conference on Development and Learning, ICDL/Epirob, Providence (RI), USA},
year = {2015},
url = {http://clement-moulin-frier.github.io/pdf/puigbo2015icdlsocialworkshop.pdf}
}
