Invited Speaker 2


Assoc. Prof. Yukie Nagai

Osaka University, Japan

Yukie Nagai is a Specially Appointed Associate Professor at Osaka University, Japan. She is also a Visiting Researcher at Bielefeld University, Germany. Her research interests are in the area of human-robot interaction for robot learning.

For more information visit her web page:


Invited Talk: How Interaction Shapes the Way Robots Learn

Abstract: Interaction with human tutors plays an important role in robot learning. Tutors do not simply demonstrate tasks or provide feedback to robots but also emphasize what aspects of the tasks (e.g., the goal or the means) robots should learn and how robots should improve their actions. Similar phenomena can be observed in caregiver-infant interaction. Caregivers adjust the complexity of their behaviors depending on infants’ abilities and emphasize important aspects of their actions by exaggerating the actions.

We have been investigating how such scaffolding facilitates robot learning by analyzing and modeling caregiver-infant interaction as well as human-robot interaction. How and why do caregivers/human tutors reduce the complexity of their actions and emphasize important aspects of the actions? What signals from infants/robots elicit such caregivers’/tutors’ scaffolding? How do caregivers and infants/tutors and robots mutually shape interaction in order to facilitate infants’/robots’ learning?

This talk will present three studies to address the above issues: (a) analysis of scaffolding behaviors, (b) modeling of an infant-like robot to elicit tutors’ scaffolding, and (c) analysis and modeling of mutual shaping of interaction. Our experiments demonstrate that (a) exaggerated actions by caregivers properly guide the bottom-up visual attention of infants so that infants would better learn the goal of the actions, (b) saliency-based bottom-up attention of a robot expresses its immature ability to anticipate the goal of a demonstrated action and thus elicits exaggerations of a tutor’s actions, and (c) caregivers gradually increases the complexity of their actions as infants develop, which leads to further development in infants. Taken together, these results emphasize the importance of designing human-robot interaction to facilitate robot learning.


Sample Videos

You can find videos with robot experiments conducted by Prof. Nagai on her website here: