You are on page 1of 44

Realistic Psychosocial Behaviour in Non Player Characters

Reading Course by Christine Bailey

Overview

NPC psychology Typical NPC control architectures Social Science Relevant social computing research

NPC Psychology

Cognition

Thinking

Emotion

The sum of past events Many many different models possible Categorical vs. dimensional

Typical NPC Control

Actors vs. Agents The usual suspects actor control General agent architectures Specialized agent architectures

Actors or Agents?

Actors

Not autonomous Actions are scripted/hard-coded Actors are directed Autonomous Not hardcoded Not rigidly directed

Agents

Some NPCs can be both via switching

NPC Control in Games - Actors

Usually:

FSMs (i.e. for state of mind) Scripting for behaviour

All states and transitions must be defined a priori

Agent Architectures

Reactive/Reflex Triggering

Internal state self-awareness & memory Preferred states Goal importance: First-come, first-served Also, no plan preference Utility-value on goals, plans

Goal-based

Utility-based

Anytime

Agent Architectures

All of these have cognition Only utility-based has emotion (nonarbitrary plan-selection)

Emotion as decision-making

According to Damasios studies of people with brain damage:

If cognition and emotion are separated, decision-making cannot occur in a nonarbitrary fashion

Cognition=plan-generation Emotion=plan-selection (non-arbitrary)

Specialized Agent Architectures

Soar H-CogAff EXCALIBUR being-in-the-world

Soar

Utility-based Some components:

Learning, knowledge, goals

Preference list So, this is an architecture that considers emotion

H-CogAff

Anytime-based Several layers:

Reactive (quick responses) Deliberative (goal-based) Meta-management (learning)

Also: Personality module

EXCALIBUR

Anytime-based Concerns:

Incomplete knowledge planning Resource needs Temporal scheduling

being-in-the-world

Hybrid architecture Real-time coping

Perception & Reaction Knowledge base Plans -> actions Plan generation

Reasoning

Specialized Agent Architectures

One of the architectures has emotion (Soar) One even has personality (HCogAff) But none of these architectures really consider social systems

Social Science

NPCs should consider social stimuli as well as physical stimuli when making decisions

An overview of the social stuff

Attitudes:

Faking it with reputation systems Remembered interaction histories Change of attitude The flow of opinion

Group influences:

An overview of the social stuff

Social roles, world view, etc.:

A high level architecture for believable social agents

An attempt to integrate mood, personality, & social networks:

Realistic Reaction System

Gimme some attitude

Faking it with reputation systems Remembered interaction histories Change of attitude

Reputation Systems

i.e. Fable, Thief: Deadly Shadows, Ultima Online, NPC opinions of player No opinions about other NPCs No individual opinions (sometimes groups) Immediate and global

Reputation Systems

Note that this is only faking a social system

Remembered Interaction Histories

Tomlinson & Blumberg (2002) Every new interaction with a stimulus changes the way we feel about that stimulus

So says Damasios Somatic Marker Hypothesis

Remembered Interaction Histories

Key elements

Current emotional state Current object of interest Emotional memory (per object)

Remembered Interaction Histories

Each new interaction, current emotional state changes:


NPCs current emotional state Remembered emotional memory

On interaction end, remembered emotional history is updated:

Factor in the emotional state during interaction

Attitude Change

Prendinger & Ishizuka (2002) Attitude: a disposition to like/dislike something (or someone, or some place, )

Attitude Change

Attitude affected by:


Previous attitude Weight of emotional events Recency of emotional events Current events

Events that conflict with previous attitude?

Momentary vs. Essential like/dislike Momentary event is integrated with previous attitude Essential event replaces previous attitude

The influence of the group

Opinion flow Peoples opinions are not static they may change depending on new events, peer opinions, etc.

Opinion flow as flock

Cole (2006) Combines:

Social Network Analysis Boids algorithm

Social Network Analysis (SNA)

Nodes Ties Hubs

Nodes w/ high number of ties

Attack of the Boids

The Boids algorithm


Alignment (steer towards avg. heading) Cohesion (move to avg. position) Separation (avoid crowding) Field of vision

Opinion flow as a flock

Alignment steer opinion towards avg. group opinion Cohesion steer opinion towards avg. opinion of people of respect Separation steer opinion away from avg. opinion of people of contempt Field of vision tie connectivity score

Cool, dynamic changes of opinion

In real life, people groups change their opinions

i.e. groupthink, conformity,

Most simulated environments: everyone has static opinions/beliefs

Social roles, world view, etc.

A high level architecture for believable social agents Social roles, social roles, social roles!

A high level architecture for believable social agents


Guye-Vuilleme & Thalmann (2001) Social roles:


A coherent set of standard behaviours (plans) Multiple exist in single individual Are a way of adapting to a situation Can inherit from other social roles (i.e. woman, mother) Are associated with other social roles (i.e. mother, father)

A high level architecture for believable social agents

Role switching through activation energy Role activation energy associated with:

Initial role weight Locations Times Locality to associated roles (i.e. mother/father)

Role energy can also be inhibited

A high level architecture for believable social agents

Values

pursuit of an ideal abstract state (because I am honest.) A classifying criteria (because hes a crook.) Action-selection based on being right, not being successful

World view via own interests, integrated values, norm-based behaviour

A high level architecture for believable social agents

Cognitive module

Active role? Belief generation selects behaviour by weighing beliefs/values Inference engine, world view

Social reasoner

Type identifiers

Integrating psychology and social networks

The Realistic Reaction System

The Realistic Reaction System (RRS)

Gruenwoldt, Katchabaw, & Danton (2005) Integration of:

Personality Mood Social relationships

The RRS

Elements:

AI Controller

Personality/Mood Filter Relationship Manager

Relationship Network

The RRS

Social ties have:


Type History Regularity Strength Polarity

Summary

NPC Psychology

Cognition & Emotions Actors, Agents,

NPC control

Social Science Relevant social computing research

Attitudes, group influences, social roles, world view, etc., integrating mood, personality, & social networks

The End

Questions?

You might also like