Improving Artificial Companions in Computer Games

点击翻译按钮获取中文摘要

Many modern games provide artificial "companions" or followers for the player. Overtly, they function as story-driven proxies for human accompaniment, although in a more general sense they also serve as a form of natural interface into the game system, important to providing various forms of passive or active assistance to the player in achieving game goals or enacting game mechanics. Artificial companion design, however, remains a difficult problem. Positive examples exist, but tend to be lauded mainly for their aesthetic, story-oriented integration. More commonly, implementations of companions in games tend not to live up to player expectation, aggravating as much as actually helping players; misjudged support behaviours, poor player (and inter-companion) coordination, obtuse interactions, and a limited range of abilities are frequent complaints. Inappropriate, unhelpful responses greatly reduce the sense of immersion and accompaniment, and defeat the goal of using companions as part of the game interface.******This research aims to improve the technical capability and interactive qualities of artificial companions in modern computer games. We aim at the mechanical benefits, where algorithmic potential exists to develop advanced companion abilities, greatly extended and more flexible environment awareness, recognition of player intent and requirements, and to significantly improve coordination with other artificial, non-player characters. Our approach aims at the two core requirements of artificial character interaction. Companions first need to have an extensive set of capabilities, appropriate to their representations as intelligent agents. This includes the ability to act in ways that accord with common, complex modes of player behaviour (combat strategy, sneaking, exploration, etc), the ability perform relevant analyses that enable such behaviour, and interfaces for explanatory and interrogative modes. Use of such abilities is predicated on a good model and understanding of player intent and strategy. This requires the ability to recognize basic player properties such as skill level, frustration, etc., as well as more novel measures that capture the motivation for and use of specific strategic behaviours, such as exploration, sneaking, ambush, etc. The need for compact, efficient techniques for estimating and representing player knowledge leads to a range of interesting and complex knowledge representation and inference concerns.******The general problem of creating good virtual companions is of course unbounded. Games are a natural stepping stone in that direction, providing a tightly constrained and easily modifiable interaction context. Our work directly improves game designs, but has the advantage of also generalizing to the growing use of virtual environments, where natural interface requirements to complex virtual spaces are becoming increasingly important.**