Autonomous Virtual Humans Tyler Streeter April 15, 2004.

Autonomous Virtual Autonomous Virtual HumansHumans

Tyler StreeterTyler Streeter

April 15, 2004April 15, 2004

2004 Autonomous Virtual Humans 2004 Autonomous Virtual Humans Tyler StreeterTyler Streeter

Project OverviewProject Overview

Goal: To create virtual humans in a Goal: To create virtual humans in a 3D environment and give them 3D environment and give them complete motor controlcomplete motor control


Current TechnologyCurrent Technology

FarCry, UbiSoft Entertainment



The Sims 2, Electronic Arts Inc.



Very limited intelligenceVery limited intelligence No low-level motor controlNo low-level motor control Typically, “finite state machines” are usedTypically, “finite state machines” are used

– Example:Example: If in If in AggressiveAggressive State, attack State, attack If in If in ScaredScared State, run away State, run away If in If in IdleIdle State, stand still State, stand still

– Use a pre-scripted animation for each state Use a pre-scripted animation for each state


My ApproachMy Approach

Create physically-simulated humans Create physically-simulated humans (demo)(demo)

Give humans simulated brains Give humans simulated brains (demo)(demo)

Train/teach them to perform certain Train/teach them to perform certain taskstasks


Artificial Neural NetworksArtificial Neural Networks

Use computer software or hardware to mimic Use computer software or hardware to mimic biological nervous systemsbiological nervous systems

Useful for things like speech & handwriting Useful for things like speech & handwriting recognitionrecognition

Can also be used to control robots or simulated Can also be used to control robots or simulated creatures…creatures…


Training a Neural Network…Training a Neural Network…

Use a Genetic AlgorithmUse a Genetic Algorithm– Start with a “population” of random Start with a “population” of random

neural networksneural networks– Evaluate each one on some task (e.g. Evaluate each one on some task (e.g.

standing or jumping)standing or jumping)– Throw away the bad neural networksThrow away the bad neural networks– ““Mate” the good networks to produce Mate” the good networks to produce

offspringoffspring– Randomly mutate the new offspringRandomly mutate the new offspring


Demo VideosDemo Videos

Standing VideoStanding Video Jumping VideoJumping Video


Future WorkFuture Work

New sensory inputsNew sensory inputs– Better sense of touchBetter sense of touch– Sense of sightSense of sight– Sense of hearingSense of hearing

Robotics ApplicationsRobotics Applications


Future WorkFuture Work

Try more complex behaviorsTry more complex behaviors– Staying balanced when pushedStaying balanced when pushed– Walking across uneven terrainWalking across uneven terrain– Carrying objectsCarrying objects– Jumping over obstaclesJumping over obstacles– Operating virtual machineryOperating virtual machinery– Competitions between virtual humansCompetitions between virtual humans


Questions?Questions?

Please come to my demo booth Please come to my demo booth upstairs to see more demonstrations upstairs to see more demonstrations and ask questions.and ask questions.

Autonomous Virtual Humans Tyler Streeter April 15, 2004.

Documents

Transcript of Autonomous Virtual Humans Tyler Streeter April 15, 2004.