Transcript presentation source

Towards engaging
full-body interaction
P.Petta1, A.Staller1, R.Trappl1,2,
S.Mantler3, T.Psik3, Z.Szalavári3, M.Gervautz3
HCI’99:
Beliefs/Emotions in
Adaptive Interface Design
1Austrian
Research Institute for Artificial Intelligence
2Dept.Med.Cybernetics & Artificial Intelligence, Univ.of Vienna
3Inst.of Computer Graphics, Vienna Univ. of Technology
Summary
 principled background:
– appraisal theory: emotions as (grounded) process
 threefold use of emotions:
– for uptake of stimuli
– for selection of action
– for expressive behaviour
 implementation
– appraisal integrated into the
trionic agent control architecture model
– “radically simple” scenario
Principled background: appraisal theory
 the Emotional as
concern satisfaction process
 Concerns
curiosity, familiarity, well-functioning,
exercising, control, attachment, …
 Appraisals
mediate between occurrence of
concern-relevant events and
changes of internal action tendencies
 Action tendencies
states of readiness to achieve or
maintain a given kind of relationship
with the environment:
stimulus-driven, not goal-oriented
stimulus event
appraisal:
concern relevance,
coping potential
action tendency
change
expressive action
Threefold use of emotions: stimulus uptake
 stimulus uptake at different levels:
- in the context of the current behaviour
- in the context of the current activity (schema)
- in broader context (planning,…)
Threefold use of emotions: action selection
 action tendencies
approach (“desire”), avoidance (“fear”), being-with (“confidence”),
attending (“interest”), rejecting (“disgust”), non-attending (“indifference”),
agonistic (“anger”), interrupting (“surprise”), dominating (“arrogance”),
submitting (“humility”), deactivation (“sorrow”), bound activation (“effort”),
excitement, free activation (“joy”),
inactivity (“contentment”), inhibition (“anxiety”)
as separate classification system for behavioral
repertory of an agent
 behavior classes
can be derived in a principled way from
an analysis of a given lifeworld
(including artificial ones)
3-fold use of emotions: expressive behaviour
 emotion categorization by action tendency change
vs.
 emotion categorization by nature of emotional object:
- Highly dependent upon which objects are
distinguished and considered important
- Well-known example: OCC model
3-fold use of emotions: expressive behaviour
 “...emotions defined primarily by their object …have
no characteristic facial expression and ... their
presence cannot be recognized by means of
expressive behavior alone” (N.Frijda)
 action expression problem as particular problem in
architectures based solely on the second principle
(e.g., the OCC Model)
 Action tendencies are tied to
characteristic expressive behavior
 Behavior classes share specific expressive characteristics
 “Emotions” as changes in action tendency
Implementation:
lessons from control architecture design
 Advances in particular with respect to layered designs
 Change in appreciation of the role of the “middle” tier
of the established trionic model,
the scheduler (“middle-out” model of control):
 scheduler receives important information from
lower level behaviors (“cognizant failure”) &
spawns long-term planning activities
(cf. e.g. E.Gat’s ATLANTIS)
 scheduler as natural location for implementation of
(a substantial part of) expressive action generation support
Implementation: trionic agent architecture
Plan library
(static)
Communications:
 vision data,
animation status
 parameterized
animation
selection,
emotion system
state
Scheduler:
• event appraisal,
• plan activation,
• “RAP” selection
• “RAP” execution
monitoring
“Reactive action packages”:
• start/context/ending conditions
• idle/cleanup procedures
• “cognizant failures”
Emotion
subsystem:
action
tendency
dynamics
Implementation: scenario (ALIVE, MIT)
“magic video mirror”
with back-projection
video image mixed
with computer
generated
overlay
•unencumbered interaction between human
visitor and virtual character based on
position, postures, and gestures
•main focus on “virtual presence”
Implementation: constraints
 interface & scenario issues
 avoid specialized scenarios,
suggest straightforward, simple interaction
 balance sensory&behavioural capabilities of
the synthetic actor and
capabilities elicited of human users
Implementation: expressive behavior
 Animation is assembled in
real time from basic motion
captured sequences that
are blended
 Dynamic surface texture
synthesis in real time,
based on current action
tendency values
Implementation: system architecture
VISION
Identification of
user location,
posture&gesture
appraisal of
user location,
posture&gesture
AGENT
CONTROL
action tendency update
action selection
segmentation
video image
acquisition
motion sequencing
texture
parameters
motion blending
texture generation
ANIMATION
real-time rendering
and compositing