Artificial Intelligence in
Game Design
Path Following
•
Examples:
– Move from one location to another along known route
– Patrol area cyclically
– Implementation of high-level path planning
• Go to door of bookstore
• Go down hall
• Go into PC room
Path Following
•
Note that will never
follow path perfectly
– Like arrive, can’t ever
reach an exact location
– More realistic to move in smooth curves
– More realistic if don’t
follow exact path every
time (patrolling) Desired
path
Path Following
•
Break desired path down into set of points along path
•
Seek each point in sequence
– When “close enough” to point, seek next point
Path Following
•
Waypoints at important
path points
– Generated automatically or chosen manually by level designer
– Most accurate to meet
goal of path
– Path planning algorithms used to create path from waypoints
Wall Avoidance
•
Goal: keep character some avoidDistance from walls
– Usually based on “bounding sphere” of object
• Most graphical objects represented this way
Wall Avoidance
•
Project path of character to some lookahead away
– lookahead > minimum distance required for stop at maxAccel when traveling at maxSpeed
– Projection can use same algorithms that graphics use for ray casting
•
If wall within lookahead, steer away
Wall Avoidance
•
Determine collision point
•
Project to point perpendicular to wall safeDistance away
•
Seek
that location
avoidDistance
Collision point
Wall Avoidance
•
May still collide with wall
•
Algorithm will keep steering until parallel to wall
New collision point
Wall Avoidance
•
Corners can be problem if not detected by ray trace
– Very common problem in robotics
• One solution: “whiskers”
Project three lines from character
No collision detected
Collision Avoidance
•
Goal: avoid collisions with other characters who are
also in motion
– Is collision going to occur?
– If so, how should we steer?
•
Location of closest approach
Collision Avoidance
•
If distance less than some
safeRadius at that point
Flee
that point
•
If already in collision with
object, just
Flee
its current
location
safeRadius
Collision detected
Flee where
obstacle will be at that point
safeRadius
In collision
Collision Avoidance
•
Much more difficult if multiple obstacles
–
Determine time to collision of all
–
Flee
from location of first collision
–
No perfect solution to problem
No collision First collision
Collision Avoidance
•
Expensive to compute if many obstacles
– C characters checking for collisions with each other C2 checks
Jumping
•
NPC jumping over obstacles key to many games
– Run towards obstacle to achieve needed speed
– Jump over obstacle
– Land on other side
•
Must be convincing
– Speed, launch point must look realistic
– Character should not jump unless can succeed
(unless player interrupts)
Going too slow for realistic jump
Jumping
•
Jump points
– Target for “run up” by NPC
– Usually have corresponding “landing pad”
• Large enough so covers all landing points from all of jump point
– Created by level designer
• Ideally room designed so jump point looks “natural”
Jump point
Jumping
• “Starting radius” for run up based on maxSpeed, maxAccel – Outside radius Seek jump point
– Inside radius Arrive point outside radius Seek jump point
Jump point
Combining Steering Behaviors
•
What if multiple steering rules fire?
–
What should resulting steering be?
–
Blending results of rules
–
Prioritizing some rules over others
$ $ $
$ $
Flee player
Arrive gold
Blended Steering Behaviors
•
Weighted sum of steering from different rules
–
Weights = relative importance of action
– Can depend on “strength of stimulus (distance, etc.)
•
Example:
– Flee player with weight 0.5/distance from player
– Arrive gold with weight 1/distance from gold
– Steering = sum of these
$ $ $
Blended Steering Behaviors
• Can result in “intelligent” path choices
$ $ $
Blended Steering Behaviors
•
Flocking
–
Seek
“center of gravity” of flock
–
Align
with average direction
–
Avoid
direct collisions
Blended Steering Behaviors
•
Blending conflicting behaviors can give results that meet
no goals
pursue
Wall avoid
result
seek flee
Prioritized Steering Behaviors
•
Choose most important goal and execute its steering
completely
•
Often proximity based
– Within safeRadius of wall execute WallAvoid behavior
– Within attackRadius of player execute Seek behavior
Pursue
not considered Wall avoid
Prioritized Steering Behaviors
•
Can group similar behaviors
– Give each unique priority
• Often based on “level of damage”
to character if not handled
– If total steering from high-priority group < threshold
try next highest priority group
• Stimulus for high-priority behavior not strong enough to matter
Avoid obstacles group
Avoid walls to side Avoid wall on turns
Avoid collisions group
Avoid cars to side Avoid car in front
Follow path group
Prioritized Steering Behaviors
Avoid obstacles group
Avoid walls to side Avoid wall on turns
Avoid collisions group
Avoid cars to side Avoid car in front
Follow path group
Seek next point on path around track
Next path point
Steer away from wall has priority
Ignore possible collisions with cars Ignore next point on path
Prioritized Steering Behaviors
Avoid obstacles group
Avoid walls to side Avoid wall on turns
Avoid collisions group
Avoid cars to side Avoid car in front
Follow path group
Seek next point on path around track
Avoid collision has priority
Avoid obstacles steering below threshold Ignore next point on path
Next path point
Collision detected
Minimal steering from walls
Prioritized Steering Behaviors
Avoid obstacles group
Avoid walls to side Avoid wall on turns
Avoid collisions group
Avoid cars to side Avoid car in front
Follow path group
Seek next point on path around track
Seek next path point has priority
Avoid obstacles steering below threshold Avoid collisions steering below threshold
Next path point Minimal steering
from walls
Steering
Prioritized Steering Behaviors
•
Can still get stuck in cycles
– Steer for one high-priority behavior
• Far enough away so no longer priority
– Other behavior becomes priority
• Steers back toward first problem
flee
flee Flee enemies
Seek target
Higher Level Control of Steering
•
Conflicts best resolved by higher-level reasoning
– What is current state?
– What overall goal are we attempting to accomplish?
Same room as player
Seek player
Different room as player
Arrive door Player exits room
Enter room where player is