Non-Player Character research

Project Abstract

This project aims to explore how we can design Non-Player Characters (NPCs) that create engaging and immersive player experiences. As part of this project we identified which behaviors contributed to making NPCs seem believable, and created a typology and model of which types of NPCs exist within games.

The products of the project are:

  • The Game Agent Model - A model describing in which situations an NPC has to exhibit certain behavior in order to be perceived as believable.

  • A Typology of Non-Player Characters - A description of the various forms of NPCs found in games.

  • The Non-Player Character Model - A model describing the types of NPCs listed in the typology achieve believability.

Research done within the scope of this project between 2010-2016 was overseen by Harko Verhagen. Junior researchers included myself (Henrik Warpefelt), Magnus Johansson, and Björn Strååt. The project description on this page focuses on my contributions to the project.

The Game Agent Matrix

The Game Agent Matrix (GAM) describes the behaviors that need to be exhibited by NPCs to be believable. These behaviors are highly situational, and must be interpreted in context. The GAM is based on the Carley & Newell Fractionation Matrix (Carley & Newell, 1994) and has been adapted to only show behaviors relevant to characters found in games.

The columns of the GAM represent the social complexity of the situation in which the agent is present. The Single Agent column represents an NPC acting without awareness of other NPCs. The Multiple Agents column contains behaviors that involve the NPC acknowledging the existence of other NPCs on a physical level, but it does not exhibit any behaviors that would imply modeling of other NPCs as reasoning entities. The Social Structural and Social Goals columns involve increasingly complex social behaviors that require the modeling of the intentions and wishes of other NPCs in the world. Finally, the Cultural Historical column represents the modeling of cultural norms and a historical understanding of behavior.

The Game Agent Matrix from (Warpefelt, 2016)

The rows of the GAM represent the complexity of behavior exhibited. The Act row is limited to behaviors that simply require the NPC to act without any understanding of the world beyond except in the purely spatial sense. In essence, NPCs in the Single Agent/Act cell consider the world to be composed of more or less mobile rocks. The React row involves behaviors where the NPC has to react to what other agents are doing, and the Interact row involves behaviors where the NPC must perceive the world and provide contextually appropriate responses to stimuli.

The color coding of the GAM denotes how well these behaviors worked at the time of publication (Warpefelt, Johansson & Verhagen, 2013). The color denotes the frequency and severity by which behaviors were found to have a negative impact on the believability of NPCs, and by extension the immersion of the player. The scale goes from green to red, with red being the most severe.

Defining Non-Player Characters

This project produced two key artifacts in understanding NPCs: A typology of NPCs, and a model describing in which situations these NPCs achieve believability.

The NPC typology provides a list of the various forms of NPCs found in games. They are sorted into three hierarchical classes: Metatypes, types, and subtypes. The metatypes provide an abstract description of the general class of NPC, the types provide a detailed descriptions of specific types of NPCs, and the subtypes provide special cases of the types listed in the typology.

This typology builds on a typology originally created by Bartle (2004). However, the majority of the types in the typology are not present in Bartle’s typology. Bartle also introduced a type of NPC he called Make the place look busy. The NPC typology assumes that this is a subtask for all NPCs, and thus not necessarily a distinct type in itself.

Two studies were conducted to establish (Warpefelt & Verhagen, 2015) and later refine (Warpefelt & Verhagen, 2016) the typology. These studies consisted of large online surveys, gathering 600 responses across both studies. Respondents were provided with a picture of an NPC and asked to the suggest a type and provide a motivation for their reasoning. In the second study (Warpefelt & Verhagen, 2016) respondents were also shown a video and asked if they wanted to revise their answer.

The Non-Player Character (NPC) Model takes the types of NPCs and contextualizes them using the complexity of the behavior they, at a minimum, must be able to exhibit to achieve their role. The rows are taken from the GAM (see above) and represent the same behavior complexity: acting, reacting, and interacting with other agents in the world.

The columns for the NPC Model represent the degree to which behavior can be dynamically created. The Embedded column represents behaviors exhibited in situations that can be hard-coded into the game, whereas the Emergent column represents behaviors exhibited in situations that arise dynamically as the player interacts with the game. An alternative naming could be Pre-defined vs Procedurally Generated, but Embedded and Emergent were chosen to conform to Jenkins’s typology of narrative (Jenkins, 2004). At the time of publishing (early 2017) there were no traces of truly believable emergent behavior, but plenty of believable embedded behavior.

The Non-Player Character Model from (Warpefelt & Verhagen, 2017)

References

  • Carley, K. and Newell, A., 1994. The nature of the social agent. Journal of mathematical sociology, 19(4), pp.221-262.

  • Jenkins, H., 2004. Game design as narrative architecture. In Wardrip-Fruin, N., Harrigan, P. (ed), First Person: New media as story, performance, and game. Cambridge, MA: MIT Press, pp.118-130.

  • Warpefelt, H., 2016. The Non-Player Character: Exploring the believability of NPC presentation and behavior (Doctoral dissertation, Department of Computer and Systems Sciences, Stockholm University).

  • Warpefelt, H., Johansson, M. and Verhagen, H., 2013. Analyzing the believability of game character behavior using the Game Agent Matrix. In DiGRA Conference (pp. 1-11).

  • Warpefelt, H. and Verhagen, H., 2015. Towards an updated typology of non-player character roles. In Proceedings of the international conference on game and entertainment technologies (pp. 1-9).

  • Warpefelt, H. and Verhagen, H., 2016. A typology of non-player characters. In First Joint International Conference of DiGRA and FDG.

  • Warpefelt, H. and Verhagen, H., 2017. A model of non-player character believability. Journal of Gaming & Virtual Worlds, 9(1), pp.39-53.