Presentation is loading. Please wait.

Presentation is loading. Please wait.

Object Persistence for Synthetic Characters Damian Isla Bungie Studios Microsoft Corp. Bruce Blumberg Synthetic Characters MIT Media Lab.

Similar presentations


Presentation on theme: "Object Persistence for Synthetic Characters Damian Isla Bungie Studios Microsoft Corp. Bruce Blumberg Synthetic Characters MIT Media Lab."— Presentation transcript:

1 Object Persistence for Synthetic Characters Damian Isla Bungie Studios Microsoft Corp. Bruce Blumberg Synthetic Characters MIT Media Lab

2 Expectations for Synthetic Creatures Expectations: Assumed aspect of world state that – for one reason or another – cannot be observed directly Assertion: The ability to form expectations and act on them is an essential component of common sense intelligence. Learning Gradual, long time-scales, large example sets e.g. learn to classify spoken utterances Expectations Immediate, short time-scale, small example sets e.g. Sheep walks behind a wall. Where did it go? When will I see it again?

3 Object Persistence Object persistence as Location Expectation When a target objects location is not observed for some time, how is the creatures idea of the location maintained / updated?

4 The Domain Duncan Duncan Concentrate on search tasks Concentrate on search tasks

5 Expectation Theory Observation + Predictor Expectation Observation + Predictor Expectation Expectation Verification Expectation Verification –Positive verification (confirmation) –Negative verification (expectation violation) –Unverifiable Verification Expectation refinement Verification Expectation refinement –Possibly also predictor refinement

6 Probabilistic Framework Usually a space of predictions Usually a space of predictions Negative verification: space of negated predictions Negative verification: space of negated predictions Distribution representation is key Distribution representation is key

7 Spatial Expectations Probabilistic Occupancy Map –Discrete spatial probability distribution –Uncertainty through discrete diffusion

8 POM Algorithm If target observed:Find closest node n* Otherwise:Divide map nodes into visible (V) and nonvisible (N) sets Either way:Diffuse Probability Positive Verification Unverifiable Negative Verification

9 Emergent Look-Around Also: Emergent Search Also: Emergent Search Simple rule: always direct gaze towards most likely location of the target Simple rule: always direct gaze towards most likely location of the target

10 Expectations and Emotions Many emotions imply expectations Many emotions imply expectations –Surprise, disappointment, satisfaction, confusion, dread, anticipation… Individual observations may have affective implications Individual observations may have affective implications Emotional autonomic variables: Emotional autonomic variables: Emotions may –Focus attention (salience) –Bias behavioral choices –Affect decision-making parameters –Affect animation (facial and parameterized) –Act as indicators of overall system state

11 Expectations and Emotions Surprise (unexpected observation ) Surprise (unexpected observation ) Confusion (negated expectation) Confusion (negated expectation) –Proportional to amount of culled probability Frustration (consistently negated expectations) Frustration (consistently negated expectations)

12 Architecture Synthetic vision Synthetic vision Rule-matching Rule-matching Parameterized animation engine Parameterized animation engine Burke et al., CreatureSmarts, GDC 2001 Burke et al., CreatureSmarts, GDC 2001

13 Results Emergent look-around Emergent look-around Emergent search Emergent search Salient Moving objects Salient Moving objects Distribution-based object- mapping Distribution-based object- mapping Emotional reactions Emotional reactions –Surprise –Confusion –Frustration Video

14 Issue: Scalability Adaptive resolution maps Adaptive resolution maps Logical maps Logical maps Hierarchical maps Hierarchical maps

15 Conclusions Simple mechanism, complex results Simple mechanism, complex results –Simple implementation –Intuitive Layered decision-making Layered decision-making –Pseudo-reasoning Useful theory Useful theory

16 Questions? Damian Isla naimad@media.mit.edu http://www.media.mit.edu/~naimad Bruce Blumberg bruce@media.mit.edu http://www.media.mit.edu/~bruce Synthetic Characters http://www.media.mit.edu/characters


Download ppt "Object Persistence for Synthetic Characters Damian Isla Bungie Studios Microsoft Corp. Bruce Blumberg Synthetic Characters MIT Media Lab."

Similar presentations


Ads by Google