6162009

1
SOAR and Video Games

• By Rex Oleson II

2
Video Game Industry

• Video games industry reached 9.9 billion in
  sales for 2004
• Grand Theft Auto San Andreas sold more then 5.1
  Million units
• Halo 2 sold more then 4.2 Million units

3
Games to Application Crossover

• Video Game to Simulation
• SOCOM Navy Seals
• Developed along side of a military training app
• Call Of Duty
• Considered to be a realistic simulation of
  military conflict

4
Traditional Game Logic

• Patterned Tile Movement
• Zelda
• Flocking
• Potential Functions
• Basic Probablity

5
AI techniques become involved

• Fuzzy Logic
• Threat Assessment
• Neural Networks
• Finite State Machines

6
Intro to SOAR and Gaming

• John Laird and Mike van Lent develop and
  interface between QuakeII/Decent3 and SOAR
• Both of Which are First Person Shooter Style
  Games(FPS)
• John Laird has been presenting papers, and
  hosting open talks almost every year at the GDC
  since then

7
Industry

• Create system for developing intelligence for
  game character
• Make the games more fun
• Refine a reusable knowledge base
• Develop a common interface for games to access
  the knowledge base

8
Research Fields

• Environment for testing concepts
• Development of new research ideas
• Visual environment to see the implications of the
  AI ideas

9
(No Transcript)
10
SOAR SGIO

• Came out of interfacing SOAR with video games
• SGIO SOAR General Input Output
• SOAR does not make calls to the environment, only
  to the output

11
SGIO Classes

• Soar
• Object representing the connection to SOAR
• Agent
• Represents an individual agent in SOAR
• Working Memory
• Handles the bookkeeping of the agents memory

12
SGIO connection types

• There are 2 connection types to use
• API SOAR
• Compiled directly into the application
• SIO SOAR
• Communicates remotely via sockets

13
SGIO Framework
14
SOAR Game Cycle Interaction

• Unreal Bot notices entities that have changed
• Native DLL sends updated info to SGIO
• When Observing is done, commit all changed WMEs
  to SOAR
• SOAR decides the command/s to issue

15
SOAR Game Cycle Interaction

• After Decision, the command/s are retrieved by
  the SGIO
• Top Command on the queue is passed to the native
  code
• Native DLL reports the current command to UNREAL
• Bot performs action based on command

16

• //For API Soar (i.e. integrated kernel)
• sgioSoar soar new sgioAPISoar()
• //For SIO Soar (i.e. TSI debug windows)
• sgioSoar soar new sgioSIOSoar("127.0.0.1",6
  969,true) //IP,port,lockstep
• sgioAgent agent soar-CreateAgent("my-agent")
  //agent name
• agent-LoadProductions("my-agent.soar") //file
  name
• sgioWorkingMemory mem new sgioWorkingMemory
  (agent)

17

• //Args for IDs parent, ID name
• sgioSoarId radarId mem-CreateIdWME(mem-GetI
  Link(),"radar") //parent,name
• sgioSoarId tankId mem-CreateIdWME(radarId,"t
  ank") //parent,name

18

• //Args for Elements parent, attribute name,
  attribute value
• sgioIntElement distance mem-CreateIntWME(tan
  kId,"distance",5)
• sgioStringElement position
  mem-CreateStringWME(tankId,"position","left")

19

• Modifying Existing WMEs on Input-Link
• mem-Update(distance,4) //element, new attribute
  value
• Removing WMEs on Input-Link
• mem-DestroyWME(tank) //element to remove
  children automatically removed
• Sending Changes to Soar
• mem-Commit()

20

• Running Agents
• //To run all the agents on a particular
  connection for 15 decision cycles
• soar-RunTilOutput()
• //To run a single agent for 15 decision cycles
• agent-RunTilOutput()
• Checking Output-Link for Commands
• bool waiting agent-Commands()

21

• Reading Commands from the Output-Link
• stdauto_ptrcmd
  agent-GetCommand()
• stdstring name cmd-GetCommandName() //name
  "move"
• stdstring value cmd-GetParameterValue("direct
  ion") //value "right"

22

• Marking Command as Processed
• //If everything goes well
• cmd-AddStatusComplete()
• //If there is an error (i.e. missing attributes)
• cmd-AddStatusError()
• cmd-AddErrorCode(5) //integer

23
Divergence of Research and Game Development

• Most game developers are interested in cheats,
  not true intelligence
• Researchers have migrated to problems where
  empirical comparisons are possible
• Game developers need real time algorithms
• Game-Playing Public is pushing for much more
  realistic and advanced AIs

24
Future Work

• SGIO is being abandoned for a new technique
• Details where not given

25
References

• U.S. video game industry sales dip in 2004,
  http//yahoo.reuters.com/financeQuoteCompanyNewsAr
  ticle.jhtml?duidmtfh70002_2005-01-18_20-45-13_n18
  697880_newsml, Tue Jan 18, 2005 0345 PM ET
• AI for Game Developers, David M. Bourg Glenn
  Seemann, OReilly, July 2004
• Intelligent Agents in Computer Games, Joe
  Hartford, John Laird, et al., http//ai.eecs.umich
  .edu/people/laird/papers/AAAIDemo.pdf, Aug. 1999
• It Knows What Youre Going To Do Adding
  Anticipation to a Quakebot, John E. Laird,
  Agents, 2001, pps 385-392
• GAME AI THE STATE OF THE INDUSTRY, PART TWO,
  David C. Pottinger and John E. Laird,
  http//www.gamasutra.com/features/20001108/laird_0
  1.htm
• Stokes-SoarToUnreal-S22
• SGIO tutorial
• SGIO quick reference