A Semantic Navigation Model for Video Games

Abstract. Navigational performance of artificial intelligence (AI) char-acters in computer games is gaining an increasingly important role in the perception of their behavior. While recent games successfully solve some complex navigation problems, there is little known or documented on the underlying approaches, often resembling a primitive conglomerate of ad-hoc algorithms for specific situations. In this paper we develop a generic navigation model which we call semantics-based, because it enables AI to incorporate a wide variety of navigation information into the planning of a character’s behavior, including raw geometry, strategic objects, and locomotion abilities. The role of semantics in this domain is highlighted and the presentation of the main components of this semantic navigation model is illustrated with a variety of examples. The semantic navigation model has been validated and implemented within a navigation system for Unreal Engine 3 that requires little level designer intervention, has a rich interface for AI programmers, and can be extended with other types of semantic information. It is concluded that using this navigation model delivers more natural paths, requires fewer world annotation, and supports dynamic re-planning