11/7/08

1

Professor Charles Rich Computer Science Department rich@wpi.edu

IMGD 400X (B 08) 1

AIIDE’08

Artificial Intelligence for Interactive Media and Games

IMGD 400X (B 08) 2

Artificial Intelligence and Interactive Digital Entertainment Conference

2008

October 22-24, 2008 Stanford University Stanford, California, USA

11/7/08

2

Mission Statement

IMGD 400X (B 08) 3

AIIDE is the definitive point of interaction between entertainment software developers interested in AI and academic and industrial AI researchers. Sponsored by the Association for the Advancement of Artificial Intelligence (AAAI), the conference is targeted at both the research and commercial communities, promoting AI research and practice in the context of interactive digital entertainment systems with an emphasis on commercial computer and video games.

By the Numbers

  3 days   about 130 attendees

•  about 80% academic, 20% industry [guessing]   26 papers presented

•  23 academic, 3 industry   8 technical sessions   1 panel (mixed)   5 invited talks

•  1 academic, 4 industry

  7 poster/demos (all academic) IMGD 400X (B 08) 4

11/7/08

3

Technical Sessions

1.  Applications of Reinforcement Learning

2.  Singular Techniques

3.  Planning for Actions, Story and Design

4.  Social Intelligence

5.  Stochastic and Evolutionary Approaches

6.  Hierarchical Models for Behavior and Plot

7.  Robustness and New Capabilities from Reasoning

8.  Unconventional Pathfinding Applications & Approaches

IMGD 400X (B 08) 5

1. Applications of Reinforcement Learning

Intelligent Trading Agents for Massively Multi-player Game Economies

•  J. Reeder, U. Central Florida •  G. Sukthankar, U. Central Florida •  M. Georgiopoulos, U. Central Florida •  G. Anagnostopoulos, Florida Inst. of Technology

IMGD 400X (B 08) 6

11/7/08

4

Intelligent Trading Agents...

  Goal: create intelligent trading agents for virtual markets

  Example: Eve Online •  220,000 active players •  460,000 player characters •  trading billions of units per

month

IMGD 400X (B 08) 7

Intelligent Trading Agents...

  Roles for autonomous trading agents in MMORPG marketplaces

•  provide liquidity for human players in less active markets (cf. Wall Street)

•  mechanism for game designers to manage markets (e.g., deflate prices)

IMGD 400X (B 08) 8

11/7/08

5

Intelligent Trading Agents...

  focus on the problem of creating agents with good “financial tactics” in buying and selling

  show that a reinforcement learning approach based on the market microstructure can give a trading agent a competitive advantage in amassing wealth over standard fixed policies.

  also need to protect agents from player exploitation (the introduction of easily duped trading agents in the virtual market would create an easy avenue for smart players to cheaply acquire rare items

  imbuing agents with the ability to learn trading policies from recent historical data will make them potentially more resistant to predatory trading practices.

IMGD 400X (B 08) 9

Intelligent Trading Agents...

IMGD 400X (B 08) 10

agent trains on recent market data to learn optimal trading strategy

11/7/08

6

Intelligent Training Agents...

IMGD 400X (B 08) 11

2. Singular Techniques

Lightweight Procedural Animation with Believable Physical Interactions

•  Ian Horswill, Northwestern University

IMGD 400X (B 08) 12

11/7/08

7

Lightweight Procedural Animation...

IMGD 400X (B 08) 13

!"#$!"#$

• %#&'(')!*+' • ,+'-!#.!/'+$'011)

– 2'+3045'(6!(.#7(!#+.

– 8#7/60!/9)1#31

/ $

• :/0.!1+5'30

• ;.!0.404!*+'!#.!0'(3!#<0!.(''(!#<0

– =5.1!5.40'!>?@

– :5$9!!!+!'5.!+.!>&+A!B.+!!!01!04C

– D(1!

– @;"*'#0.46)

85//+'!1 13'#/!#.$ '5..#.$

• E"+!(//6#3(!#+.1

8#756(!#+. +* @#.1"+'!9F1– 85//+'!1!13'#/!#.$G!'5..#.$!(1!(!10'<0'G!+'!4#'03!!(5!9+'#.$!+*!&09(<#+'1

– 8#756(!#+.!+*!@#.1"+'!9 1!H1(*0!9+70!&(10I!/90.+70.+.

,0&3+7#3– ,0&3+7#3

Lightweight Procedural Animation...

  http://twigblog.wordpress.com

IMGD 400X (B 08) 14

11/7/08

8

Lightweight Procedural Animation

IMGD 400X (B 08) 15

!"#$%&!'$#!()*+#,"+!-./%0!1+-/2$3!+45+"/1+%6'!.%!7'8(!9/2*$+3

!"#$%&!*.3-!'2"/56

'2"/56

:"$01+%6

!"#$%&!0.6. 2$1+"$!;<=!>

!"#$%&!'$#!(?.7!@$'6$"-'A(!2$1+"$

9/2*$+3&!0.6. !"#$%!B<CD!E"F!B!BG!>

)*70& 0.6. !"#$% B D EB B BG >:"$01+%6 )*70&!0.6. !"#$%!B<D!EB!B!BG!>

9/2*$+3&!'$#!(H7/+6A(!!"#$%

9/2*$+3&!'$#!()*+#,33!*+$"!#.7A(!!"#$%

!"#$%&!'$#!(I,1!%.6!'.1+!3$@!$%/1$3A(!9/2*$+3

)*70&!'$#!(I,1!J/6*!KKKI<L%M.1+ J/6*!1+(!!"#$%

!"#$%&!3..N$6 )*70

!"#$%&!'$#!(I!N%.J!1#!"/0*6'A(!)*70

!"#$%& '$# (IO! J.73- %+P+" '/0% .:: .% 6*/'A( 9/2*$+3!"#$%&!'$#! IO!!J.73-!%+P+"!'/0%!.::!.%!6*/'A !9/2*$+3

)*70&!*.3-!!"#$%

9/2*$+3&!'$#!(I6,'!"7%!@#!K3@+"6.!Q.%R$3+'!%.J<(!!"#$%!>

!"#$%&!:/0*6!)*70

B D5$7'+!B<D

!"#$%&!'$#!(S/-0+6'!.:!6*+!J."3-!7%/6+A(!)*70

)*70&!0.6. .::'6$0+!>

!"#$%&!-".5!'2"/56# 5 5

!"#$%&!(T.#3+%6 0"++%AL%I6,' 1$-+!.76!.:!5/4+3'AU!!"#$%

6/63+'&!:$-+6.@3$2N C

3. Planning for Actions, Story and Design

Simulation-Based Story Generation with a Theory of Mind

•  H. Chang & V. Soo, National Tsing Hua University, Taiwan

IMGD 400X (B 08) 16

11/7/08

9

Simulation-Based Story Generation...

  Goal: “Emergent narrative”, i.e., stories emerge from autonomous interactions among NPC’s and/or player

  Challenge: How to guide narrative to be interesting

IMGD 400X (B 08) 17

IMGD 400X (B 08) 18

11/7/08

10

IMGD 400X (B 08) 19

IMGD 400X (B 08) 20

11/7/08

11

Simulation-Based Story Generation...

  “our main concern is whether the system can generate new stories”

IMGD 400X (B 08) 21

4. Social Intelligence

Otello: A Next-Generation Reputation System for Humans and NPCs

•  M. Sellers, Online Alchemy, Inc.

IMGD 400X (B 08) 22

11/7/08

12

Next-Generation Reputation System...

23

Next-Generation Reputation System...

  as more games merge into online social spaces, reputation systems are becoming increasingly important to players looking for gaming partners

  also new gameplay ideas •  compete on who is more respected, more loved,

or more feared •  reward players who fill social roles, such as being

reputation “hub”, etc.

IMGD 400X (B 08) 24

11/7/08

13

5. Stochastic and Evolutionary Approaches

  Constructing Complex NPC Behavior via Multi-Objective Neuroevolution

•  J. Schrum & R. Mukkulainen, U. Texas Austin

IMGD 400X (B 08) 25

Constructing Complex NPC Behavior...

  Goal: Discover NPC behavior automatically

  Benefits: •  save production time/effort •  learn counterintuitive behaviors •  find weaknesses in static scripts •  tailor behavior to human plays

IMGD 400X (B 08) 26

11/7/08

14

27

15 NPCs attack player

Player

Bat

Infinite Plane

6. Hierarchical Models for Behavior and Plot

Hierarchical Petri Nets for Story Plots Featuring Virtual Humans

•  D. Balas, C. Brown, A. Abonyi & J. Gemrot, Charles U. in Prague

IMGD 400X (B 08) 28

11/7/08

15

Hierarchical Petri Nets...

IMGD 400X (B 08) 29

Hierarchical Petri Nets

  formal and graphical (intelligible, like FSM’s)

  story plots can be branching

  episodes can happen in parallel

  can be hierarchical

IMGD 400X (B 08) 30

11/7/08

16

IMGD 400X (B 08) 31

7. Robustness and New Capabilities from Reasoning

Recombinable Game Mechanics for Automated Design Support

•  M. Nelson, Georgia Inst. of Technology •  M. Mateas, U. California Santa Cruz

IMGD 400X (B 08) 32

11/7/08

17

Recombinable Game Mechanics...

  GameMaker and Alice provide support for novices to implement games

  How about a similar approach for designing game mechanics?

IMGD 400X (B 08) 33

Recombinable Game Mechanics...

  Factor game design into four areas •  abstract mechanics: state and state evolution

•  concrete representation: audiovisual realization of game state

•  thematic content: real-world references a game makes

•  control mappings: how player interacts with game

  Library of combinable components at all levels

IMGD 400X (B 08) 34

11/7/08

18

IMGD 400X (B 08) 35

8. Unconventional Pathfinding Applications and Approaches

A Cover-Based Approach to Multi-Agent Moving Target Pursuit

•  A. Isaza, J. Lu, V. Bulitko & R. Greiner, U. Alberta

IMGD 400X (B 08) 36

11/7/08

19

A Cover-Based Approach...

  Goal: an efficient multi-agent system (algorithm) that is capable of capturing a single moving target

  Challenge: to coordinate multiple pursuers

  Solution: an elegant uniform algorithm based on “cover sets”

IMGD 400X (B 08) 37

A Cover-Based Approach...

IMGD 400X (B 08) 38

11/7/08

20

Panels & Invited Talks

  Realistic Human Characters •  Chris Darken, Naval Postgraduate School

•  Richard Evans, EA/Maxis

•  Borut Pfeifer, EA Los Angeles

•  Michael Mateas, UC Santa Cruz

  Experiments in Musical Intelligence •  David Cope, UC Santa Cruz

  The AI of Spore •  Eric Grundstrom, EA/Maxis

IMGD 400X (B 08) 39

Invited Talks

  Halo 3 Objective Trees: A Declarative Approach to Multiagent Coordination •  Damian Isla, Bungie Studios

  The Past, Present and Future of Game AI •  Steve Rabin, Nintendo of America

  Performing Intent •  Doug Church, EA Los Angeles

•  Borut Pfeifer, EA Los Angeles

IMGD 400X (B 08) 40

11/7/08

21

Questions? Comments?

IMGD 400X (B 08) 41