Digra 2103
-
Upload
magnus-johansson -
Category
Entertainment & Humor
-
view
194 -
download
0
description
Transcript of Digra 2103
Analyzing the believability of game character behavior using the Game Agent Matrix Henrik Warpefelt, Magnus Johansson,
Harko Verhagen {hw, magnus, verhagen}@dsv.su.se
Department of Computer and Systems Sciences, Stockholm
University
DeFragging Game Studies 2013
Agenda
• Aim
• Believable behavior
• Related research
• Method
• The Game Agent Matrix
• Results
• Conclusions
• Future work
8/24/13
Aim
“[w]e like to address the behavioral properties of Non Player Characters (NPCs) and the social awareness of and amongst NPCs aiming for humanlike behavior in NPCs. ” (Johansson, Warpefelt and Verhagen 2013)
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Analyze “how characters must act within the context of the fictive world in which they live in order to maintain the player’s sense of immersion”
AND
Believable behavior
“In order for NPCs to be perceived as believable they must act in a way that is appropriate given the context they inhabit” (Warpefelt et al, 2013)
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
Characters must be designed so that they exist in accordance with the story, and their behavior must be coherent with the personality they are said to have (Lankoski, 2007)
The character must at all times act in a way that is appropriate with the personality, feelings, situation and such of the character (Loyall, 1997; Mateas, 1999)
Related research
• Believable agents (Loyall, 1997)
• Interactive drama and believable agents (Mateas, 1999)
• Design patterns (Lankoski and Björk, 2007a;b)
• Johansson and Verhagen (2011)
• Warpefelt and Strååt (2013)
• Analyzing immersion (Ermi and Mäyrä, 2007)
• Narrative immersion (Adams, 2010)
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
The Gap
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
Lankoski (2004); Björk and Lankoski (2007)
Loyall (1997); Mateas (1999)
?
Method
• In-game observations in 11 AAA-games
• Observations from a black-box perspective
• 2-hour recordings from each game
• Analyzed using the Game Agent Matrix, focusing on exhibited behavior of non-player characters
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
8/24/13
Games in the study
Car crash in L.A Noire
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Hunter vs Crab
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
The Game Agent Matrix
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Acting agents
An acting agent does not interact with other entities in the world, but instead acts completely by internal volition. While
it is aware of the physical structure of the world, it is only so
to the degree that it can navigate static obstacles. It is
completely unaware of other entities acting in the world, i.e.
players or other characters. In essence, an acting character
views everything as rocks that may need to be navigated
around.
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
Reacting agents
A reacting agent has a greater degree of awareness of the world, but only to an extent that it can adapt to changes in the
social environment. If it does perform social interaction, this
is in response to simple stimuli, such as another entity moving
within a certain range. Reactions are often repetitive and
predictable, and the results from previous interactions are not
remembered – it will gladly answer the same question
multiple times in a row. A reacting agent is aware of the existence of other entities in the world, but is not capable of
modeling their internal state. It may have a model of its own
internal state though (model of self).
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Interacting agents
An interacting agent exhibits a much larger degree of social capability, and is able to alter its behavior to match changing
social situations, for example acting to maintain norms within
the culture or taking turns in conversation. Behavior is
generally characterized as being varied and able to in a
flexible way carry social interaction, rather than the
repetitions exhibited by the reacting agent. An interacting
agent is aware of other entities in the world and can model their internal state.
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
Results
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Act/Single agent
• Most NPCs manage to display these behaviors successfully
• Uses language, uses tools, “pantomimed” behaviors that with few exceptions were found successful.
• Pathfinding mostly not a problem
8/24/13
React/Single agent
8/24/13
Adaption “Able to adapt to changing social circumstances in the world at a given time”
– NPCs fail to adapt to improve its situation – Or, adapts in such a way that the situation worsens
Awareness “Aware of something in its immediate vicinity” – Lack of awareness – Hyperawareness
Interruptability “Able to stop doing what it is currently doing when another task takes priority”
Models of self “Knowledge of its own existence as an entity, physical or mental”
React/Multiple agents
8/24/13
Models of others “Awareness of the existence of other agents, where they are and what they are doing ”
Interact/ Multiple agents
Face to face “Turns towards the entity it is addressing” – Most often successful
Route following “Able to transport itself across open ground between two
points in the world”
Navigation “Able to dynamically adjust its rout through the world in order to
account for unexpected obstacles”
8/24/13
Less common values
Reinforcing immersion: – Advertising – Institutions – Roles – Group conflict
Problematic: – Cooperation – Etiquette – Norm maintenance – Sanctions
8/24/13
Conclusions
• Limited study
• 11 games, 2 hour in-game data recorded for each game
• Enough data to evaluate the Game Agent Matrix and inform further tune-ups
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Conclusions regarding method
• The Game Agent Matrix (GAM) useful for both detecting negative and
positive behaviors/traits in NPCs
• Some values appear simultaneously
• “Face to face” should be moved
• More work remains with the Game Agent Matrix
• “Memories of previous interactions”
• “Awareness” and “Acquires information” problematic
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
Conclusions regarding NPCs and Games AI
8/24/13 Magnus Johansson Department of Computer and Systems Sciences
Future work
• Refining the Game Agent Matrix further
• Additional data collections
• The social gap will be investigated further
• Evaluating the Game Agent Matrix through interviewing game developers
• Mapping the Game Agent Matrix to Design Patterns by (Lankoski and Björk, 2007 a b)
8/24/13 Analyzing the believability of game character behavior using the Game Agent Matrix
Questions?