Incorporating Human Relationships Into Path Planning

While pathfinding is ubiquitous in games, it is most often abstracted away from everything else that is going on in the game and viewed as a simple point-to-point shortest path problem. The notable exception to this is work on tactical pathfinding. Tactical pathfinding incorporates tactical elements, most often for battle simulations, into the pathfinding engine in order to improve high-level behavior. This paper looks at how the relationship between characters in a game should influence the paths that are taken through the game. We suggest a simple model of character relationships and show how this can be used to influence the paths that are taken through the world. Finally, we discuss design-related issues of how this could be incorporated into games. Introduction and Overview The problem of finding point-to-point paths in a world is a common task in many games, and there are a wide variety of planning techniques and pathfinding representations (Board and Ducker 2002; Nieuwenhuisen et al. 2004; Demyen and Buro 2006; Kallmann 2010; Sturtevant 2011) which have been developed for solving this problem within the constraints of modern video games. These approaches form a technical solution to the problem of finding paths, and as such, some might argue that this isn’t really Artificial Intelligence (AI), or that it isn’t an particularly interesting aspect of AI in video games. We argue, however, that if agents cannot move around the world in a competent manner, then they will not appear intelligent. It doesn’t matter what techniques are applied to give this illusion of intelligence, only that the illusion remains relatively intact throughout a playable experience. Because the reasoning process behind pathfinding can be expensive, research into representations and other technical problem details is an important part of maintaining this competency. But, the sentiment behind the critique also has some merit. The best path between two points is not necessarily the shortest one as measured by path length. The best path should take into account everything that is going on within the world being simulated. Work in the area of tactical pathfinding (Millington and Funge 2009) is the primary research that has considered richer world and cost repCopyright c © 2013, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. resentations, although a majority of this work has been in the context of first-person shooters (Straatman, van der Sterren, and Beij 2005; Kondeti et al. 2005) or military simulations (Darken, McCue, and Guerrero 2010). Example applications of this work in games has been to help agents avoid moving through corridors of fire, or to coordinate movement for tactical purposes. In this paper we suggest that there are many situations where a pathfinding engine should be taking into account the relationships between characters when planning paths. That is, the paths planned by characters in games should take into account their relationships and emotions, reflecting the state of mind of the agents. Properly integrating this into the pathfinding process has the potential to enhance the ability for game designers to create compelling characters that engage in unique interactions and performances within a game. We propose a simple model of character intent when moving and look at how this model can be incorporated into a pathfinding engine, and then discuss the practical application of this work. Note that this work focuses primarily on the path being planned for movement. The selection of body and facial animations when following the path are a separate area of work beyond the scope of this paper. Background and Related Work There has been a significant amount of research on human relationships, but very little of this work is focused directly on the modeling needed for this paper. We point out several particular fields of research and some related work in crowd simulation, but there are a large number of publications outside of AI which have some level of relevance to this work. Studies of Human Space and Behavior The field of proxemics looks at how humans move and position themselves in relationship to each other. This is influenced by the emotional state of an individual as well as cultural norms (Hall and Hall 1969). The sense of personal space depends on who is being interacted with and the cultural norms of those who are interacting. This work divides space around an individual into four categories, which we describe below. These were further sub-divided into close and far phases, although we will not make that distinction here. Intimate Distance Personal Distance Social Distance Public Distance Figure 1: The relationship distances, as defined by Edward Hall. • Intimate distance. This is described as “the distance of love-making and wrestling, comforting and protecting.” In other words, this is the distance of the most tender and most harsh words between individuals. • Personal distance. This is the distance that most noncontact species maintain; it is close enough to touch, but not close enough that vision is distorted. • Social distance. This distance is close enough to allow for social interactions, with most physical details still clear, but far enough that in a louder environment voices might need to be raised to communicate clearly. • Public distance. This is the limit at which humans can clearly communicate, but vocal nuances begin to be lost, and non-verbal communication becomes more dominant. Visually, as shown in Figure 1, these distance form concentric circles or ellipses around an individual, from the furthest distances where an individual is safest, to the intimate distances where an individual is most vulnerable. We will use this formation as one axis of a simplified model for planning preferences. Related to work on proxemics is work on describing human movement. This originated from work by Rudolf Laban (Laban and Ullmann 1966), and has been widely applied to dance and other artistic endeavors. This work provides insights not only to high-level motion planning, but also to the individual animations used for characters. Although we will not use this work directly here, it has been used previously for analyzing motion capture data (Bouchard and Badler 2007).