kwc.org Photos Spare Cycles MythBusters

Talk: AI Characters and Directors for Interactive Computer Games

Brian Magerko, John E. Laird, Mazin Assanie, Alex Kerfoot, Devvan Stokes

Information on a storytelling environment built in Unreal Tournament.

Haunt II

Built in Unreal Tournament

Design goals

  1. Generalizable design
  2. Flexible efficient system
  3. Fully-structured story: allow author complete control over specifying plot content and structure
  4. Interactive player experience: direct connection between player's behavior and experience player gets
  5. Believe behavior

Architecture:

SOAR-based * long term memory (skills, doctrine, tactics) * short term memory (situation, goals) * perceive->decide->act loop

Story director

  • inputs: sensing of AI's and human player, abstract story specification
  • output: commands to AI character and environment
  • predictive model of user behavior. lookahead to see if player is going to muck up the direction of the story, and if so, try to guide player to stay within the bounds of the story.

Director execution cycle

  • knowledge maintenance
  • plot maintenance
  • execute direction

Knowledge maintenance: Hypothesize entity knowledge and keep track of world state

Plot representation: plot points are atomic events that happen in world that can change world state. Specified in first-order logic as preconditions and actions. Plot points are partially ordered. * Timing constraints: Can also add in timing constraints to control pacing. If time constraints are violated, signals that director agent should try to make preconditions true (e.g. tell player to move, create sounds to attract player). Use user prediction to predict whether or not user is going to violate timing constraints by modeling user actions in fake world state. All actions are modeled with some fixed cost.

Soar as Actors

Goal-based behavior

Individualized personality (physiology, emotion modeling (Bob Mariner))

Directability (Cantina domain)

Knowledge organization

Bottom-up behaviors (get warm) and top-down goals

Directable Agent

Achieves or improvises directions received from director. Choose goals whose transition is most coherent (e.g. measure of node distance), can have partially instantiated goals. Improvises set of goals.

Future work

  • evaluation (with and w/o directions)
  • authoring content
  • integration of emotional model (SOAR method actors)
  • exploration of user modeling component of director agent

Post a comment


tags.

related entries.

what is this?

This page contains a single entry from kwc blog posted on July 27, 2004 2:19 PM.

The previous post was Talk: Real Robots for the Real World.

The next post is Talk: A Multi-Resolution Pyramid for Outdoor Robot Terrain Perception.

Current entries can be found on the main page.