Interaction with Group of Autonomous Characters

1
Interaction with Group of Autonomous Characters

• Craig Reynolds
• (Game Developers Conference 2000)
• Presentation by Son, Ji-Hye

2
Outline

• Introduction
• Related Work
• Behavioral Models
• Case study Pigeons in the Park
• Reacting to the User
• Realtime Performance
• Conclusions

3
Introduction

• Real world
• Busy, crowded
• Game world
• Static, empty
• Large numbers of autonomous characters
• Coordinate with or react to each other, the
  environment and users avatar
• Three key concepts
• Behavioral models
• Spatial data structure
• The techniques used to drive animation from
  behavior models

4
Related Work

• The historical trend
• Very simplistic autonomous characters
• Nimble reactive characters Reyn99
• Characters which can learnGran98,
  reasonLair00, and planFung99
• The games
• SimCity, SimAnt, SimTower, The Sims
• Lemmings, Populous
• Myth, Warcraft

5
Behavioral Models (1/2)

• The behavior model
• Drives autonomous characters
• position, velocity, orientation, and visual
  appearance
• Maps the characters environment into character
  actions.
• The characters environment
• Virtual world (external environment)
• Memory or metal processes (internal environment)
• Simple physical model
• A point mass with velocity and a local frame of
  reference
• A finite steering force
• The local coordinate system

6
Behavioral models (2/2)

• Multiple distinct behavioral states
• Transitions are triggered by changes in the
  characters environment.
• FSM (finite state machine)
• Ex) a fighter jet

7
Case study Pigeons in the park (1/3)

• Pigeons in the park
• User/player interaction
• Real-time
• A large group of simple autonomous character
• The virtual world
• The birds
• A radio controlled car
• An undulating terrain
• An invisible spherical obstacle
• The pigeons
• The Boids model Reyn87
• Steering behaviors Reyn99

8
Case study Pigeons in the park (2/3)

• A nearby car or the hand clap causes the birds to
  take flight.
• The urge to take off is contagious.
• Sensing the nearby birds who have recently taken
  off
• A bird land again.
• The annoyance value
• gets ratcheted up when a bird gets startled
• Then slowly decays over time

9
Case study Pigeons in the park (3/3)

• Animation for the birds
• Behavioral parameters
• Position, orientation, walking/flying flag, rate
  of change of the glide slope
• A set of pre-animated cycles
• Walking, Flying, Gliding, WingFold
• The demo runs on a PlayStation2 at an update rate
  of 60 fields per second
• 280 birds
• 55000 triangles

10
Reacting to the User (1/2)

• Respond to the user in two different ways
• Discrete reaction
• Continuous reaction
• A neighborhood around the car
• Depends on the cars position and velocity.
• Shape
• a half circle behind the car
• a half ellipse in front of the car.

11
Reacting to the User (2/2)

• The neighborhood
• Move away
• Take off
• In the move-away zone
• Move away from both the car and its current path.

12
Realtime Performance

• Three main sources of computational load
• Rendering
• Thinking
• Locality queries
• Rendering
• Rendering the 3D shaded, textured, scene, in
  perspective.
• PlayStation2 hardware, CDS software, simple
  shading models
• 15 to 20 of each display cycle (one-sixtieth of
  a second)

13
Realtime Performance Thinking

• Thinking
• The time taken for each of the autonomous
  characters to think
• The cost is multiplied by the number of
  characters.
• O(n)
• To reduce the cost
• Decouple the behavioral computation from the
  simulation rate
• a low rate 10Hz (one sixth of the animation
  rate)
• The thinking about obstacle avoidance is done
  at a higher rate (30Hz)

14
Realtime Performance Locality queries (1/2)

• Locality queries
• The cost of doing locality queries between
  characters
• An interacting particle system
• N comparisons for each particle
• Complexity of O(n2)
• To reduce the cost
• To store the characters in spatial data structure
• To keep the characters pre-sorted based on
  their location in space

15
Realtime Performance Locality queries (2/2)

• The bin-lattice spatial data structure
• Divide a space into a collection of smaller boxes
  (bin)
• A locality query
• performed by specifying a sphere and a function
• O(n)

16
Conclusions

• A realtime system with a large group of
  autonomous characters
• move themselves around
• React to each other, to obstacles, and the users
  avatar
• Realtime performance
• By decoupling the behavioral model from the
  graphical update rate
• By the use of a lattice-bin spatial subdivision
  technique