Semantic navigation in video games

Leonard van Driel

Video games in the action-adventure genre often provide artificially controlled characters (AI) to provide more realistic and diverse game play. In order to exhibit intelligent or challenging behavior, these characters strongly depend on their navigational performance, often referred to as pathfinding. While recent games solve increasingly complex and specific navigation problems, the documented approaches to navigation remain primitive and isolated within the gaming software realm.

This project develops a semantic approach to navigation that allows AI to extract and incorporate a wide variety of navigation information into the planning of character behavior, including raw geometry, strategic objects, and abilities for locomotion. This approach leads to a rich and clear semantic structure with respect to designers, programmers, the game engine, and system resources.

Following this approach, a navigation system was designed and implemented as a subsystem inside Unreal Engine 3. This system requires no unnecessary designer intervention, has a rich interface for AI programmers, and can be extended with other types of semantic information. It proves to outperform the original pathfinding functionality by delivering more natural paths, fewer world annotation, and dynamic re-planning.