La lecture en ligne est gratuite
Le téléchargement nécessite un accès à la bibliothèque YouScribe
Tout savoir sur nos offres
Télécharger Lire

Measurement and analysis of interactive behavior in tutoring action with children and robots [Elektronische Ressource] / Anna-Lisa Vollmer. Technische Fakultät. Research Institute for Cognition and Robotics

143 pages
Publié par :
Ajouté le : 01 janvier 2011
Lecture(s) : 0
Signaler un abus

Measurement and Analysis of
Interactive Behavior in Tutoring
Action with Children and Robots
Anna-Lisa VollmerDiplom Mathematikerin Anna-Lisa Vollmer
AG Angewandte Informatik
Research Institute for Cognition and Robotics (CoR-Lab)
Technische Fakult at Universit at Bielefeld
email: avollmer@techfak.uni-bielefeld.de
Abdruck der genehmigten Dissertation zur Erlangung
des akademischen Grades Doktor-Ingenieur (Dr.-Ing.).
Der Technischen Fakultat der Universit at Bielefeld
am 06.07.2011 vorgelegt von Anna-Lisa Vollmer,
am 31.08.2011 verteidigt und genehmigt.
Gutachter:
Dr.-Ing. Britta Wrede, Universit at Bielefeld
Dr.-Ing. Jannik Fritsch, Honda Research Institute Europe, O enbach/Main
Prof. Angelo Cangelosi, University of Plymouth
Prufungsaussc huss:
apl. Prof. Dr.-Ing. Stefan Kopp, Universit at Bielefeld
Dr.-Ing. Britta Wrede, Universit at Bielefeld
Dr.-Ing. Jannik Fritsch, Honda Research Institute Europe, O enbach/Main
Prof. Angelo Cangelosi, University of Plymouth
Dr.-Ing. Thies Pfei er, Universit at Bielefeld
Gedruckt auf alterungsbesant digem Papier nach ISO 9706.Measurement and Analysis of
Interactive Behavior in Tutoring
Action with Children and Robots
Der Technischen Fakult at der Universit at Bielefeld
zur Erlangung des Grades
Doktor der Ingenieurwissenschaften
vorgelegt von
Anna-Lisa Vollmer
Bielefeld, Juli 2011Abstract
Robotics research is increasingly addressing the issue of enabling robots to
learn in social interaction. In contrast to the traditional approach by which
robots are programmed by experts and prepared for and restricted to one
speci c purpose, they are now envisioned as general-purpose machines that
should be able to carry out di erent tasks and thus solve various problems
in everyday environments. Robots which are able to learn novel actions in
social interaction with a human tutor would have many advantages. Unex-
perienced users could \program" new skills for a robot simply by demon-
strating them.
Children are able to rapidly learn in social interaction. Modi cations in tu-
toring behavior toward children (\motionese") are assumed to assist their
learning processes. Similar to small children, robots do not have much ex-
perience of the world and thus could make use of this bene cial natural
tutoring behavior if it was employed, when tutoring them.
To achieve this goal, the thesis provides theoretical background on imitation
learning as a central eld of social learning, which has received much at-
tention in robotics and develops new interdisciplinary methods to measure
interactive behavior. Based on this background, tutoring behavior is exam-
ined in adult-child, adult-adult, and adult-robot interactions by applying
the developed methods. The ndings reveal that the learner’s feedback is a
constituent part of the natural tutoring interaction and shapes the tutor’s
demonstration behavior.
The work provides an insightful understanding of interactional patterns and
processes. From this it derives feedback strategies for human-robot tutoring
interactions, with which a robot could prompt hand movement modi ca-
tions during the tutor’s action demonstration by using its gaze, enabling
robots to elicit advantageous modi cations of the tutor’s behavior.
iiiAcknowledgements
I gratefully acknowledge the nancial support from Honda Research Insti-
tute Europe for the project \Acquiring and Utilizing Correlation Patterns
across Multiple Input Modalities for Developmental Learning".
ivContents
List of Figures ix
List of Tables xi
1 Introduction 1
1.1 Robots for Household Use . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Robot Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Main Goals and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Social Interaction: Imitation Learning 7
2.1 The Question of What to Imitate . . . . . . . . . . . . . . . . . . . . . . 8
2.2 To Imitate or to Emulate|De nitions and Evidence from Neuro-Science
and Behavioral Science . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Imitation Learning Approaches in Robotics . . . . . . . . . . . . . . . . 12
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 Tutoring Behavior 15
3.1 Behavior Modi cations in Infant-Directed Interaction . . . . . . . . . . . 15
3.1.1 Motherese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.2 Motionese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Operationalizing Motionese . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2.1 Methodology of Qualitative Data Analysis . . . . . . . . . . . . . 17
3.2.2 Annotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2.3 Visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.4 Quantitative Measures for Motionese . . . . . . . . . . . . . . . . 22
vCONTENTS
4 Analyzing Tutoring Behavior 29
4.1 The Motionese Corpus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2 Motionese Compared to Modi cations in Tutoring Robots . . . . . . . . 31
4.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 Motionese Toward Children of Di erent Age . . . . . . . . . . . . . . . . 42
4.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5 Analyzing Learner Behavior 47
5.1 Feedback: Children’s Contribution to Tutoring Interactions . . . . . . . 48
5.1.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.1.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.1.3 Group 1: Prelexical Infants (8 to 11 months) . . . . . . . . . . . 49
5.1.4 2: Early Lexical Infants (12 to 24 months) . . . . . . . . . 54
5.1.5 Group 3: Lexical Infants (25 to 30 months) . . . . . . . . . . . . 56
5.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6 The Interactional Account of Motionese 59
6.1 On the Loop of the Tutor’s Action Modi cations and the Learner’s Gaze 59
6.1.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.1.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.1.3 Starting Point: Variability of Hand Trajectories . . . . . . . . . . 62
6.1.4 Empirical Observations on the Interplay between the Tutor’s
Hand Motions and the Learner’s Gaze . . . . . . . . . . . . . . . 64
6.1.5 Systematization: From Empirical Observations to Formal De-
scription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7 A Human-Robot Interaction Study of Feedback in an Imitation Learn-
ing Scenario 79
7.1 Motivation to Investigate Online and Turn-based Feedback in a Demonstration-
Action Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.2 Design and Realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.2.1 Subjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.2.2 Setting and Experimental Conditions . . . . . . . . . . . . . . . . 82
7.2.3 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
viCONTENTS
7.2.4 Technical Realization . . . . . . . . . . . . . . . . . . . . . . . . 94
7.3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.3.1 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.4.1 Feedback Strategies . . . . . . . . . . . . . . . . . . . . . . . . . 103
8 Conclusion 105
References 109
A Conventions for Transcription 115
B Table Overviewing the Results of the Analysis of Tutoring Behavior121
C Questionnaire for Human-Robot Interaction Study 123
D Interview for Human-Robot Interaction Study 127
viiCONTENTS
viii

Un pour Un
Permettre à tous d'accéder à la lecture
Pour chaque accès à la bibliothèque, YouScribe donne un accès à une personne dans le besoin