Constraining Autonomous Character Behavior with Human Concepts

1
Constraining Autonomous Character Behavior with
Human Concepts

• Presentation by Jonathan Martin
• Based on Chapter 3.1 of AI Game Programming
  Wisdom 2, by Jeff Orkin

2
Two Ways to Create Lifelike Behavior

• Script it in

• This ensures we get the behavior we want
• Its also extremely time-consuming
• If the scripter makes a mistake, could have
  drastic consequences

3
The Alternative Create Autonomous Characters

• Whereas scripting involves programming an action
  for every possible situation

• Autonomous agents calculate their action based on

Desires
Sensory Input
Proximity to items of interest
4
A Hypothetical Situation

• We want to design a Tactical Espionage Action
  game
• The player will be sneaking around an enemy base,
  hopefully undetected
• Guards should have behavior that is both
  realistic and interesting

5
Creating our AI

• Step 1 Create enemies that can perform certain
  actions
• Step 2 Populate level with tagged objects
• Tags indicate the action that can be performed on
  an object

6
Pseudocode

• While (alive)
• Gather input from perceptors
• Select object on which to perform action
• Execute action

7
I want some Hot Coffee
I feel asleep
I need to get some work done
8
Adding Human Concepts to the AI

• Ownership
• Group Objects into sets
• A character can own one set of objects at a
  time
• Public objects can be in their own set, or none
  at all

• Dependancy
• Certain actions can only be performed before or
  after another
• Add a dependency tag to each object, which
  indicates actions that are preconditions for the
  current action

9
Adding Human Concepts to the AI

• Responsibility
• Tag each object with the class or classes of
  character that can interact use it
• We dont want soldiers performing experiments in
  a lab, or scientists picking up assault rifles

• Priority
• Divide behaviors into categories
• When choosing an action, choose the highest
  priority behavior possible
• Attack player might be the highest-priority
  behavior

10
Adding Human Concepts to the AI

• State of Consciousness
• If the game includes sleep/unconscious states,
  make sure actions are disabled while in those
  states

• Expected State
• On seeing an ally, check if he/she is in an
  unexpected state
• If an object is moved, set a changed flag, so
  that NPCs will notice the change

11
Adding Human Concepts to the AI

• Presence of Others
• Evaluate how an action will affect other
  characters
• This will be covered in the next section

12
A Modified Pseudocode

• While (alive)
• Gather input from perceptors
• Select object on which to perform action
• Am I awake?
• What is most important right now?
• Has anything changed?
• What is my purpose?
• Is this object mine to use?
• Is there anything I need to do before using
  this object?
• How will this action affect other characters?
• Execute action