Online engagement prediction in child-robot interaction

Keywords
Loading...
Thumbnail Image
Issue Date
2018-08-20
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Robots are becoming increasingly popular in society. Within education, research is mainly focused on tutoring robots. Currently robots are compared with other methods of tutoring, but these robots are not yet adaptive, whereas adaptivity might make child-robot interaction more natural, and improve learning. Robots that can adapt to the child's knowledge and emotional state, could have a great e ect on learning. One of these projects that aims to design a tutoring robot is L2TOR. Within L2TOR a Nao humanoid robot is used for tutoring children a second language. Some steps have been taken in making the robot more adaptive towards children based on the mistakes a child makes during a tutoring task. Currently the child's engagement with the robot is not taken into account. In this thesis a pipeline is suggested to account for the child's engagement with the robot. This pipeline is based on three features: gaze, smiling and posture. These features have been identi ed as important predictors for child-robot engagement, and can be used to adapt a robot's behavior. An experiment for validating this pipeline is described. Statistics for the individual features were collected, based on an L2TOR dataset annotated with task engagement, and it was found that smiling and a combination of the three features correlated signi cantly with engagement. Guidelines for future work are provided, since it is expected that online measuring of engagement can work given that enough (balanced) data is available.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen