Transcript Powerpoint

Artificial Intelligence in
Game Design
Content Generation
Content Generation
• Creating world for game
– Terrain for battle games
– Levels for shooters
– Tracks for racing games …
• Most expensive part of development!
• Goal: Generate content automatically
– Cheaper, faster
– Infinite replayability
Terrain Maps
• Most commonly applied to terrain
maps in civilization games
Goals (possibly contradictory):
• “Realistic” appearing terrain
– Continents, islands, mountains, beaches,
deserts, etc.
• “Fair” to all players
– Equal amounts of resources, water, and
flat areas for building cities near initial
position for each player
Land Mass Generation
One method:
• Grow “land clumps” from each player location
– Initial “territory” each player given chance to expand into before
running into other players
Land Mass Generation
• Each iteration grows clump by 1 tile
– After given number of tiles, rest of map becomes “water”
• Chosen randomly from tiles adjacent to existing clump
– Must not be part of another player’s clump
• Evaluated using “fitness” criteria
• Nearness to initial player position
• Number of adjacent tiles already part
of player’s clump
• Distance from other player’s clumps
P1
P1
P1 P1
P1 P1
– If tile fails, choose another
– Goal: avoid “peninsulas”
P1 P1 P1
P1 P1
Island Map Generation
• Each player initially given separate island
– Can be built like land clumps
– Additional criteria:
New land tiles for player not allowed within some distance of tiles
for other players (keep islands separated)
– Additional islands can be generated in empty spaces after player
islands created
Terrain Generation
• Creating “realistic” looking map of land masses
• Generally height based
– Defining h(x, y) for points on the map
– Low at beaches
– High in mountains
– Flat areas for city
development
– …
Fractals
• Shapes which appear the same at any scale (self-similar)
• Based on some “variance” applied at each level of detail
– Modify area by some variance v
– Subdivide into sub areas
– Higher v  more “jagged” look
Fractals
• One dimensional example:
– Start with line:
– Vary height of center of line by random value between –v and v:
– Vary height of center of each segment by –v/2 and v/2:
Fractals
• One dimensional example:
– Vary height of center of each segment by –v/4 and v/4:
– And so on, until have as much detail as necessary
Fractals
• Create “random” terrain by doing this in two dimensions
Adding Detail
• Smoothing used to level out terrain in areas that should
be flat
– Beaches at interface of land/water tiles
– Flat areas designed to allow players to build cities, etc.
• Done by averaging heights in “flat” areas
Adding Detail
• Create mountains in non-flat areas
– Choose a location and increase its height
– Use averaging to affect height of nearby locations (foothills)
– Use wander behavior to create a mountain range over a certain
distance
Adding Details
• Create rivers as paths from mountains to ocean
– Choose random point on coast
– Seek nearest point at base of a mountain
– Add wander to avoid a straight line
Resource Distribution
• Must make sure resources are distributed fairly among
all players
– Gold, oil, uranium, etc.
– Land on which structures can be built
• Goals:
– Each player should have same amount of resources available at
any stage of game (depending on amount of territory captured)
– Players should be encouraged to explore increasingly further
from their initial position
Resource Distribution
• Generate rings around player initial position
• Randomly place r resources within each ring
– r increases with each ring
R
R
R
R
R
R
R
R
R
R
P1
R
R
R
R
R
P2
R
R
R
R
R
Generate and Test
• Generate level content in some random way
• Evaluate that content
– How playable/fun/challenging is it?
– Is the level of difficulty appropriate for that point in game?
• Modify that content
– Change features that are not appropriate
Track Generation
• Track generated as sequence of
randomly generated spline curves
• AI driver used to evaluate the difficulty
level track
– Overall time to drive the track
– Variation in speed needed to drive the
track
• Individual spline curvatures modified
for track sections which vary from the
rest of the level
Mario Level Generation
• Two-dimensional level generated
– Pits, coins, enemies, etc.
• Evaluated by human players
– Fun: time moving right, enemies stomped, etc
– Frustration: time standing still, falls into pits, etc.