“Low Level” Intelligence for “Low Level” Character Animation
Post on 30-Dec-2015
39 Views
Preview:
DESCRIPTION
Transcript
““Low Level” Intelligence for Low Level” Intelligence for “Low Level” Character “Low Level” Character
AnimationAnimation
DamiDamiáán Islan IslaBungie StudiosBungie Studios
Microsoft Corp.Microsoft Corp.
Bruce BlumbergBruce BlumbergSynthetic Synthetic
CharactersCharacters
MIT Media LabMIT Media Lab
““Low level” Animation …?Low level” Animation …?
Animation not having to do with gross Animation not having to do with gross body movement or “behavior”body movement or “behavior”– Eye gazeEye gaze– Facial expressionFacial expression– Ambient / idling animationAmbient / idling animation– Animation styleAnimation style– Speech?Speech?
Interesting because an “internal life” is Interesting because an “internal life” is impliedimplied
Cognitive ModelingCognitive Modeling
CM: Giving characters an internal lifeCM: Giving characters an internal life
Too much autonomy?Too much autonomy?
ProsProsUnpredictabilityUnpredictabilityResponsivenessResponsiveness
Leverage animationLeverage animation
ConsConsUnpredictabilityUnpredictabilityReproducibilityReproducibilityControllabilityControllability
““Low level” Cognition …?Low level” Cognition …?
A class of abilities that are relevant to, but A class of abilities that are relevant to, but independent of, high-level independent of, high-level actionaction
PerceptionPerception Knowledge modelingKnowledge modeling AttentionAttention MemoryMemory Emotional reactionEmotional reaction Motion qualityMotion quality ……
Perception
Memory AttentionKnowledge
Base
ActionSelection
Emotions
Gross BodyAnimation
FacialAnimation
GazeControl
UserControl
Example 1: AlphaWolfExample 1: AlphaWolf
Emotional memories: player Emotional memories: player has total control, but wolves has total control, but wolves react to instructions based on react to instructions based on past experiencepast experience
B. Tomlinson, “Synthetic Social B. Tomlinson, “Synthetic Social Relationships for Relationships for Computational Entities”, PhD. Computational Entities”, PhD. Thesis, MIT Media Lab 2002Thesis, MIT Media Lab 2002
Wolves maintain Wolves maintain their own their own cognition, cognition, memory and memory and emotion modelsemotion models
Example 2: Object PersistenceExample 2: Object Persistence
Piaget: The persistence of a mental image after the Piaget: The persistence of a mental image after the sensory stimulus has been removedsensory stimulus has been removed
Object Persistence = Object Persistence = location expectation location expectation formationformation
Focus on search tasks (where do I expect the sheep Focus on search tasks (where do I expect the sheep to be?)to be?)
Spatial ExpectationsSpatial Expectations
Probabilistic Occupancy MapProbabilistic Occupancy Map– Discrete spatial probability distributionDiscrete spatial probability distribution– Uncertainty through discrete diffusionUncertainty through discrete diffusion
POM AlgorithmPOM Algorithm
If target observed:If target observed: Find closest node n*Find closest node n*
Otherwise:Otherwise: Divide map Divide map nodes into visible (V) and nonvisible (N) nodes into visible (V) and nonvisible (N) setssets
Either way:Either way: Diffuse ProbabilityDiffuse Probability
*0, ( )n n p n
*1( )p n
( )culled
n V
p p n
0, ( )n V p n
1, ( ) ( )
1culled
n N p n p np
Emergent Look-AroundEmergent Look-Around
Also: Emergent SearchAlso: Emergent Search
Simple rule: always direct gaze towards most likely location Simple rule: always direct gaze towards most likely location of the targetof the target
Expectations and EmotionsExpectations and Emotions
Observations can have emotional impactObservations can have emotional impact– Wanted to see something but didn’t Wanted to see something but didn’t confusion confusion– Saw something where you didn’t expect it to be Saw something where you didn’t expect it to be surprise surprise– Having trouble finding the target Having trouble finding the target frustration frustration
… … plus variationsplus variations– Target desired + confusion Target desired + confusion disappointment disappointment– Target feared + surprise Target feared + surprise panic panic– Target desired + surprise Target desired + surprise delight delight
Emotions mayEmotions may– Focus attention (salience)Focus attention (salience)– Bias behavioral choices / Affect decision-making parametersBias behavioral choices / Affect decision-making parameters– Affect animation (facial and parameterized)Affect animation (facial and parameterized)– Act as a debugging channel!Act as a debugging channel!
Expectations and EmotionsExpectations and Emotions
Emotional Autonomic variableEmotional Autonomic variable
Surprise (unexpected observationSurprise (unexpected observation))
Confusion (negated expectation)Confusion (negated expectation)– Proportional to amount of culled Proportional to amount of culled
probabilityprobability
Frustration (consistently negated Frustration (consistently negated expectations)expectations)
*
*
)
)
(
(
highesttinst
p p ns
p n
( )tinst culled
n V
c p p n
t tinstf kc
Time
Confusion FrustrationSurprise
Duncan instructed to approach sheep
Discovers sheep is notin last-observed location
Sheep found inunexpected location.
Va
ria
ble
Va
lue
1t t tinstx x x
Results: Duncan the Highland Results: Duncan the Highland TerrierTerrierDuncan:Duncan: Virtual sheep-herdingVirtual sheep-herding Layered behavior systemLayered behavior system Synthetic visionSynthetic vision
Results:Results: Emergent look-aroundEmergent look-around Emergent searchEmergent search Salient Moving objectsSalient Moving objects Distribution-based object-Distribution-based object-
mappingmapping Emotional reactionsEmotional reactions
– SurpriseSurprise– ConfusionConfusion– FrustrationFrustration
VideoVideo
ConclusionsConclusions
““Low Level” ConclusionsLow Level” Conclusions– A model of Object PersistenceA model of Object Persistence– Simple mechanism, complex resultsSimple mechanism, complex results
Simple implementationSimple implementation IntuitiveIntuitive
““High Level” ConclusionHigh Level” Conclusion– Intelligence >> Action-selectionIntelligence >> Action-selection
You control the wolves, but what they feel mattersYou control the wolves, but what they feel matters You control Duncan, but what he knows mattersYou control Duncan, but what he knows matters
Questions?Questions?
DamiDamiáán Islan Islanaimad@media.mit.edunaimad@media.mit.edu
http://www.media.mit.edu/http://www.media.mit.edu/~naimad~naimad
Bruce BlumbergBruce Blumbergbruce@media.mit.edubruce@media.mit.edu
http://www.media.mit.edu/http://www.media.mit.edu/~bruce~bruce
Synthetic CharactersSynthetic Charactershttp://www.media.mit.edu/charactershttp://www.media.mit.edu/characters
top related