Procedural Guard Placement and Behaviour Generation

Enemy observers, such as cameras and guards, are common elements that provide challenge to many stealth and combat games. Defining the exact placement and movement of such entities, however, is a non-trivial process, requiring a designer balance level-difficulty, coverage, and representation of realistic behaviours. In this work we explore systems for procedurally generating both camera and guard placement in a stealth game context. We first investigate a Monte-Carlo approach to optimize randomized enemy positions and motions based on Voronoi-regions. We then perform an automatic roadmap construction, generating more specific patrol behaviours through a grammar-based technique and a rhythm-based technique. We evaluate both approaches with a non-trivial implementation in Unity3D, and apply quantitative metrics to demonstrate how different parametrizations can be used to control level difficulty without sacrificing believability.