let’s shake hands! on the coordination of gestures of humanoids
DESCRIPTION
Let’s shake hands! On the coordination of gestures of humanoids. Zsofi Ruttkay Herwin van Welbergen Balázs Varga. Our goals. Coordinating gesture to external signals Coordinating gesture to other modalities Comparison of synch phenomena of rhythmic motion and speech-accompanying gesture - PowerPoint PPT PresentationTRANSCRIPT
Let’s shake hands!
On the coordination of gestures of humanoids
Zsofi RuttkayHerwin van Welbergen
Balázs Varga
Our goals Coordinating gesture to external signals Coordinating gesture to other
modalities Comparison of synch phenomena of
rhythmic motion and speech-accompanying gesture
Define a synchronization language Create an adaptive real-time animation
engine
Content Examples of coordination The multimodal coordination problem
Existing solutions Our solution
Coordinating gesture To what? How to adapt gesture for coordination? How to specify coordination?
Conclusions Open issues Questions
Coordination example:Gesture-speech coordination
Gestures and speech come from a single process of utterance formation (McNeill) => Gesture timing is not a slave of speech timing
Time alignment is achieved while we speak The stroke of gesture precedes or ends at the
phonological peak syllable of the speech Often we need to adjust the timing of gestures or
speech to make the alignment fit Gestures can be speeded up or slowed down Gestures can be ‘paused’ using hold phases Speech can be stopped to finish complex gestures Connection words (uh…) can be used to slow down speech Etc
Coordination example:The virtual dancer: moving to the music
‘Beat moments’ in the animation should be aligned to beats in the music
Annotate all beat moments in the animation Predict the beats in the music Locally speed up or slow down the animation to fit to the music
There is a maximum stretch or skew in the dance motion
Coordination example:The virtual trainer: tutoring exercises on music
Exercises are executed using several modalities Body movement Speech Music/metronome Sound (clap, foot tap)
Challenges Coordination Monitoring user => real time (re)planning
Exaggeration to point out details Speed up / slow down Feedback/correction …
Coordination example:Handshake: coordination between 2 humans
Handshake is used for greeting, agreeing and accepting
Complex coordination between two persons
Guided by Social protocols Haptic feedback Visual feedback
Generalizing: the multimodal coordination problem
‘Behaviors’ on different modalities (speech, gesture, dance motion, music)
Synchronization between behaviors at key time moments
The timing of flexible behaviors can be adapted to achieve synchronization
Coordination: related work Classic approach in speech/gesture coordination:
Speech leads, gesture follows MURML (Kopp et al.)
No leading modality Planning in sequential chunks containing one piece of
speech and one aligned gesture Co-articulation at the border of chunks
BML (Kopp, Krenn, Marsella, Marshall, Pelachaud, Pirker, Thórisson, Vilhjalmsson)
No leading modality Synchronized alignment points in behavior phases For now, aimed mainly at speech/gesture synchronization In development
Coordination: our previous work Virtual Dancer
Synchronization between music (beats) and dance animation
Linear time stretching/skewing Virtual Presenter
Synchronization between speech, gesture, posture and sheet display
Leading modality can change over time GESTYLE markup language with par/seq
and wait constructs to define synchronization
Our multimodal synchronization model No leading modality, just align key moments Every phase of a behavior has a preferred
length Stretching/skewing/skipping if necessary
Coordination of hand gestures to external signals
What do we want to coordinate with?
How do synchronization constraints effect movement? How to stretch/skew?
How can we define synchronization? using BML scripts?
Ontology of coordination signals
Flexibility
Origin
Fixed Flexible
World - pointing at a moving object- clapping to rhythm of music
-
Humanoid’s own modality
- gesture aligned to speech which is taken as leading signal
- gaze and hand coordination
Other humanoid’s modality
- back-channeling as listener to a speaker e.g. by head nods
- hand shake- two hands
involved in taking over an object
Clapping experiment
Clapping and counting How is the synchrony between clap and count? How do the movement characteristics of
clapping change with tempo? Time distribution Amplitude Left/right hand symmetry
Clapping experiment: setup
Mocap analysis of two subjects
Instructions: Clap and count
from 21 to 31 Clap and count
to the metronome
Clapping experiment: results The phonological synchrony rule was valid for
counting while clapping: Clap before phonological peak of the count
The clapping was speeded up by decreasing the path distance of the hand
A pre-stroke hold can be used to slow down For our right-handed subjects, the right hand
was moving ahead in phase compared to the left
The standard deviation of the relative phase between the left and the right hand increased with the clapping frequency
Hand shake experiment Which movement phases can be
identified? How are they coordinated? What gaze patterns can be seen? What movement characteristics can be
identified in the different phases? Timing, duration Form
How is above effected by Refusal or avoidance to shake hands Social relations between participants
Hand shake experiment: setup Motion capture of two subjects (P1, P2)
shaking hands Annotation of gaze patterns Variations
Basic Triggered P2 initiates P2 tries to avoid shaking hands P2 rejects
Modeling coordination in BML
BML is an XML language defining multimodal synchronization
BML events can be used to synchronize with other BML scripts/world events
The BMLT observer is introduced for synchronization with (repeated) outside world events
Coordination with events BML is designed to work in event driven
systems <event> is used to fire an event
message <wait> is used to wait for an event
If the event does not occur after a set time, wait can fire a no-event message
After the event occurs, or the timeout is exceeded, the script continues
Coordination with events:handshake
extend withdrawpumpconnect
extend withdrawpumpconnect
wait
timeout
subject 1
subject 2
Coordination with the observer An observer observes a specific part of
the world and provides timing information on that Example: beats in music
Why observers instead of events? Explicit outside world trigger Multiple (repeated) trigger Timing of observer triggers can be predicted
for easy planning Synchronization within behavior phases <wait> does not suffice
Conclusions
Gesture synchronization mechanisms are also found in rhythmic motion
Adaptation of timing in gesture affects several movement characteristics Linear speedup/slowdown does not suffice
Gesture coordination can be modeled using BMLT
Open issues
What modalities do we have to stretch/skew/skip?
Can we generalize our findings from clapping/handshake?
Do the semantics of a motion change if we change its timing? E.g. emotions, individual features