Transcript Game AI
Artificial Intelligence in
Video Games
Jason Fuller
1
What is Game AI?
Imitate intelligence in the actions of non-player
characters (NPCs).
Make the game “feel” real.
Obey laws of the game
Show decision making
and planning
2
Goals of Game AI
Be fun!
Be challenging but not
overwhelming (unless the
hardest difficulty is selected)
Do not hog all the resources!
(CPU time)
Make sure the AI does not
cheat! (At least do not get
caught)
AI often get bonuses when
difficulty increases
3
Types of Games
1. Action games
Shooters (FPS and Third-Person)
Racing, Sports
2. RPG games (Role Playing Game)
Often include many
action game aspects of AI
3. RTS games (Real Time Strategy)
4
Game AI Types
Action and RPG AI tend to work better with Finite State
Machine based AI
RTS AI used Finite State Machines in the early years of AI
development.
RTS AI work best with Artificial Neural Networks and Fuzzy
Logic
Both contain path finding components
5
AI Path Finding
Dijkstra’s Algorithm
A* Algorithm
Most commonly used
Finds the shortest path
The world or map of the game is represented by a grid of points
6
A* Algorithm
Allows for high optimization
Either by changing the search algorithm to better suit the game or by
changing the data structures.
Very similar to how people move between locations in a city.
7
Finite State Machines (FSM)
Simplest and most basic AI model.
Consists of:
States
State Transitions
Most common for Action games!
Not many different actions for NPCs
8
Finite State Machines
Among the States and State Transitions there are 4
components:
States which define behavior
State transitions which are the movement
from one state to another
Conditions which must be met for state transition
Events/Actions which are internally or externally
generated which may lead to a state transition
9
FSM Example
10
FSM Disadvantages
Very predictable
Too many states get tough to organize
Since there are such crisp rules between states, NPC does not
feel natural
11
FSM within a State
States have a FSM within them
12
Modern FSM Example
13
History of Finite State Machines
In 1952, the game Nim used AI to play against an opponent.
1960’s & 1970’s
Spacewar!
Pong
Space Invaders
1980’s
Simcity
14
History Continued
1990’s
Dragon Quest IV
Warcraft
Half-Life
15
Artificial Neural Networks (ANN)
No agreed definition, most common one is “a network of
simple processing elements, which can exhibit complex global
behavior, determined by the connections between the
processing elements and element parameters.”
Mathematical model inspired by biological neural networks.
An adaptive structure that can learn.
16
ANN Structure
Very similar to the structure of our brain.
Input layer, processing (hidden) layer, output layer
Learns by example
17
ANN Structure
The hidden layer is not just a straight line of nodes
Each node in the hidden layer will contain just a small part of
the overall calculation
The nodes have connections
between each other with
certain weights
18
ANN Structure
The weight of the connections between the nodes determine the outcomes
of the calculations
If a node is triggered by 2 different nodes it can then determine which one is
more important
19
ANN Learning
20
21
22
Black & White
Came out in 2001
First to effectively use
Artificial Neural Networks
23
Halo Wars
Came out in 2009
Featured a “Custom” difficulty level that used Artificial Neural
Networks
24
Fuzzy Logic
Introduced in 1965 for use in Artificial Intelligence research
Present problems to computers in a way similar to how humans
solve problems and that everything is a matter of degree (or
preference or context).
25
Example Problem
A store owner needs to decide how much produce to order.
Elements to take in to consideration:
Time of year?
What is the weather like?
Is there a Holiday coming up?
26
Video
Fuzzy Logic: An Introduction
27
Games
Now that the major types of AI have been covered, I will go into
more detail about what games they are used in.
28
Racing Game AI
Large-scale cheating!
AI already know the track and optimal path
AI already has complete behavior determined before the start
of the race
29
Racing AI graphs
General Path
Optimal Path
30
Racing Game AI
In its basic form, it is the most basic of game AI but in some of
the racing simulators, the AI are more complicated
If the player is using the optimal path, the AI will actively try to
push them off of it.
The AI will also use tricks such as spinning out opponents by
making their back tires lose grip.
31
FPS Game AI
Implemented with a layered structure
Bottom layers control the path finding tasks and animation
selection
Higher layers control the tactical reasoning which is where the
Finite State Machine would be.
32
A* Graph of FPS or RPG World
General path
Playable Zone
Unplayable Zone
33
FPS Continued
F.E.A.R. series has revolutionary AI
AIs have knowledge of map elements and will flank the player
AI will break through walls and windows to get to the player
AI will rush when they heavily outnumber the player
34
RPG Game AI
Many encounters with AI are unscripted
GTA IV and Far Cry 2 made great leaps in “friendly” AI
35
Elder Scrolls IV: Oblivion
Released in 2006
During testing, a story important NPC kept being found dead.
A mechanic of the AI was the cause.
36
Bioshock Infinite
The player companion, Elizabeth (who is an AI), is almost
entirely unscripted.
37
RTS Game AI
Started out using Finite State Machines to control AI
Too many options to cover
AI was “dumb”
AI would build up in a strict way
Once the player found a strategy that worked against the AI, it
would always work.
RTS AI switched to a combination of Fuzzy Logic and Artificial
Neural Networks
38
RTS Game AI
By changing to Fuzzy Logic and Artificial Neural Networks (ANN):
Fuzzy Logic led to smarter responses to attacks
ANN led to smarter development of base and better long term decisions
39
A* Graph of RTS Map
40
RTS Continued
Maxis is again changing the simulation landscape with the new
Simcity
Every “Sim” is a full AI
Have there own agenda
Have specific wants
and needs
41
Video
SimCity: Economics AI
42
Future of Game AI
Game AI have made great leaps forward since they were first
developed.
An AI that can learn how you play a game would be a great
opponent
43
References
Champandard, Alex. "Top 10 Most Influential AI Games." Aigamedev.com. N.p., 12 Sept.
2007. Web. 25 Feb. 2013. <http://aigamedev.com/open/review/top-ai-games/>.
Grant, Eugene, and Rex Lardner. "The Talk of the Town." TheNewYorker.com. The New
Yorker, 02 Aug. 1952. Web. 25 Feb. 2013.
<http://www.newyorker.com/archive/1952/08/02/1952_08_02_018_TNY_CARDS_00023
6053>.
Grzyb, Janusz. "Artificial Intelligence in Games." - CodeProject. Software Developer's
Journal, n.d. Web. 25 Feb. 2013. <http://www.codeproject.com/Articles/14840/ArtificialIntelligence-in-Games>.
"Neural Networks: A Requirement for Intelligent Systems." N.p., 2007. Web. 25 Feb. 2013.
<http://www.learnartificialneuralnetworks.com/#training>.
"Short Term Decision Making with Fuzzy Logic And Long Term Decision Making with
Neural Networks In Real-Time Strategy Games." Hevi.info. N.p., n.d. Web. 25 Feb. 2013.
<http://www.hevi.info/tag/artificial-intelligence-in-real-time-strategy-games/>.
44
Videos
Fuzzy Logic
http://www.youtube.com/watch?feature=player_detailpage&v=P8wY6mi1vV8#
t=117s
SimCity
http://www.youtube.com/watch?feature=player_detailpage&list=UUnje_8ilXP7
KB2vdssyAWug&v=MxTcm1YFKcU#t=37s
45
QUESTIONS?
46