Upload
dinhtuong
View
214
Download
0
Embed Size (px)
Citation preview
Li, Peltason and Wrede, March 2007 1
Why and how to model multi-modal interaction for a mobile robot companion
Shuyin Li, Julia Peltason and Britta Wrede
Bielefeld UniversityGermany
Li, Peltason and Wrede, March 2007 2
Outline
Introduction to Human-Robot Interaction (HRI)Observations in a user studyA multi-modal interaction frameworkSummary
Li, Peltason and Wrede, March 2007 3
Robots should be aware of environmental changes
Introduction: HRI with a personal robot
Robot characteristics
Requirements for the system
situated situation-awareness
Li, Peltason and Wrede, March 2007 4
Robots should be aware of environmental changes
Introduction: HRI with a personal robot
Robot characteristics
Requirements for the system
Users expect human-like behaviorsanthropomorphic social
behaviors
situated situation-awareness
Li, Peltason and Wrede, March 2007 5
Robots should be aware of environmental changes
Introduction: HRI with a personal robot
Robot characteristics
Requirements for the system
Both users and robots have visual access to their
interaction partener's bodyembodied multi-modal
interaction
Users expect human-like behaviorsanthropomorphic social
behaviors
situated situation-awareness
Li, Peltason and Wrede, March 2007 6
Robots should be aware of environmental changes
Introduction: HRI with a personal robot
Robot characteristics
Requirements for the system
Both users and robots have visual access to their
interaction partener's bodyembodied multi-modal
interaction
Users expect human-like behaviorsanthropomorphic social
behaviors
situated situation-awareness
Li, Peltason and Wrede, March 2007 7
Robots should be aware of environmental changes
Introduction: HRI with a personal robot
Robot characteristics
Requirements for the system
Both users and robots have visual access to their
interaction partener's bodyembodied multi-modal
interaction
Users expect human-like behaviorsanthropomorphic social
behaviors
situated situation-awareness
Li, Peltason and Wrede, March 2007 8
Outline
Introduction to Human-Robot Interaction (HRI)Observations in a user studyObservations in a user study
BIRON@HomeBIRON@HomeQuiet speakersQuiet speakersMeta-commentatorsMeta-commentators
A multi-modal interaction frameworkSummary
Li, Peltason and Wrede, March 2007 9
The user study: BIRON@Home
Li, Peltason and Wrede, March 2007 10
The user study: BIRON@Home
Experimental setup with BIRON
Non-task behaviors of BIRON:1. situation awareness2. social behavior:
14 subjects, each interaction about 7 min.
Only output-modality of BIRON: speech
Li, Peltason and Wrede, March 2007 11
The user study: situation awareness
Face recognition
sound source detection
Face recognition
Human leg detection
Li, Peltason and Wrede, March 2007 12
Observation I: quiet speakers
Li, Peltason and Wrede, March 2007 13
Observation I: quiet speakers
The problem
No means to communicate pre-interaction attentionPossible reasoninapproprateness of speech modality
(“legs detected, face detected, face lost again, face detected, ...”)
Solutionuse non-verbal modalities
(because they are suitable to represent static information which is only occasionally updated)
Li, Peltason and Wrede, March 2007 14
The user study: social behavior
...
User: Follow me.
BIRON: OK, I'm following you.
User: This is a cup.
BIRON: It's nice.
You are really doing very well!
User: (laugh)
BIRON: Come here.
Performance remark
Li, Peltason and Wrede, March 2007 15
Observation II: meta-commentators
Li, Peltason and Wrede, March 2007 16
Observation II: meta-commentators
Problemusers reply to social comments using out-of-vocabulary words
Possible reasonreciprocity and obtrusiveness of the speech modality
Solutionmaking remarks using non-verbal modalities
(because they are unobtrusive and do not impose strong obligation to reply)
Li, Peltason and Wrede, March 2007 17
Outline
Introduction to Human-Robot Interaction (HRI)Observations in a user studyA multi-modal interaction frameworkA multi-modal interaction framework
Currently popular approachesCurrently popular approachesOur approachOur approach
Summary
Li, Peltason and Wrede, March 2007 18
Interaction framework: existing works
Currently popular approaches
address differences between multi-modal information by grouping it into categories
and handle different categories separately
Li, Peltason and Wrede, March 2007 19
Interaction framework: existing works
Cassell: generic architecture for embodied conversational agents
Li, Peltason and Wrede, March 2007 20
Interaction framework: existing works
Traum: dialog model for multi-modal, multi-party dialog in virtual world
(based on information-state theory)
Li, Peltason and Wrede, March 2007 21
Interaction framework: multi-modal grounding
Our approach
address a common feature of multi-modal information: evocative functions
Li, Peltason and Wrede, March 2007 22
Interaction framework: multi-modal grounding
Evocative functions of conversational behaviors (CBs)
Definition: CBs evoke a reaction from the interaction partner
Validity: for both propositional and interactional info
Li, Peltason and Wrede, March 2007 23
Interaction framework: multi-modal grounding
Grounding
Definition: the process of establishing shared understanding during a conversation.
Basic idea: for each account (Presentation) issued in a conversation, there needs to be a feedback (Acceptance) from the interaction partner
Application area: traditionally adopted to model evocative functions of propositional information→ to be extended!
Li, Peltason and Wrede, March 2007 24
Interaction framework: multi-modal grounding
Our approach: extending grounding
Modeling both propositional and interactional contribution with Interaction Unit Organizing IUs based on the principle of grounding
Li, Peltason and Wrede, March 2007 25
Interaction framework: multi-modal grounding
Interaction Unit (IU)
Motivation Layer
Behavior Layer
Li, Peltason and Wrede, March 2007 26
Interaction framework: multi-modal grounding
Interaction Unit (IU)
(motivation conception)
verbal generator
non-verbal generator
Li, Peltason and Wrede, March 2007 27
Interaction framework: multi-modal grounding
gesturespeech
speech
speech
facial expression
...
Grounding
Motivation
Motivation
Motivation
Motivation
Motivation
gaze
...
Li, Peltason and Wrede, March 2007 28
Interaction framework: multi-modal grounding
Grounding models:[Clark92][Traum94][Cahn&Brennan99]
Our approach: [Li2006]
S. Li, B. Wrede, and G. Sagerer. A computational model of multi-modal grounding. In Proc. ACL SIGdial workshop on discourse and dialog, in conjunction with COLING/ACL 2006, pages 153-160. ACL Press, 2006.
Discourse
IU 1Ex1
Ex2
Ex3
IU_2
IU 3
IU 4
R(Ex4,Ex2) = support
R(Ex3,Ex2) = default
Li, Peltason and Wrede, March 2007 29
Interaction framework: multi-modal grounding
Pre-interaction attention: solving the quiet-speaker-problem
User: (shows legs)uninstantiated shows legs
unintentional motivationU1
Li, Peltason and Wrede, March 2007 30
Interaction framework: multi-modal grounding
User: (shows legs)
BIRON: (opens eyes)
uninstantiated shows legs
unintentional motivation
uninstantiated opens eyes
provides acceptance to user IU
U1
U2
Pre-interaction attention: solving the quiet-speaker-problem
Li, Peltason and Wrede, March 2007 31
Interaction framework: multi-modal grounding
User: (shows face)
BIRON: (raises head)
uninstantiated shows face and legs
unintentional motivation
uninstantiated raises head
provides acceptance to user IU
U3
U4
Pre-interaction attention: solving the quiet-speaker-problem
Li, Peltason and Wrede, March 2007 32
Interaction framework: multi-modal grounding
User: This is a cup.deictic gesture
shows BIRON a cupU5
Making social comments: solving the meta-commentator-problem
“This is a cup”
Li, Peltason and Wrede, March 2007 33
Interaction framework: multi-modal grounding
User: This is a cup.
BIRON: I beg your pardon?
“This is a cup” deictic gesture
shows BIRON a cup
initiates conversational repair
U5
U6
Making social comments: solving the meta-commentator-problem
“I beg your pardon” uninstantiated
Li, Peltason and Wrede, March 2007 34
Interaction framework: multi-modal grounding
User: This is a cup.
BIRON: I beg your pardon?
BIRON: (looking embarrassed)
“This is a cup” deictic gesture
shows BIRON a cup
uninstantiated
initiates conversational repair
U5
U6
uninstantiated looking embarrassed
shows social awarenessU7
Making social comments: solving the meta-commentator-problem
“I beg your pardon”
Li, Peltason and Wrede, March 2007 35
Interaction framework: multi-modal grounding
Implemented on systems BIRON and BARTHOC
BIRON BARTHOC
Li, Peltason and Wrede, March 2007 36
Summary
Two cases studies revealing the importance of multi-modality in situatedness and social behaviors of a robot
A multi-modal interaction framework that addresses the evocative function of conversational behaviors