Analyzing Stealth Games with Distractions

Abstract

The ability to distract opponents is a key mechanic in many stealth games.  Existing search-based approaches to stealth analysis, however, focus entirely on solving the non-detection problem, for which they rely on static, ahead-of-time models of guard movements that do not depend on player interaction.  In this work we extend and optimize an approach based on heuristic search of stealth games to model variation in guard paths as dynamically triggered by player actions.  Our design is expressive, accommodating different distraction designs, including remote activation and time delays.  Using a Unity3D implementation, we show our enhanced search can solve distraction puzzles found in real games, as well as more complex, multiple-distraction level designs.  Our work shows how heuristic search can be applied to dynamically determined contexts, and significantly extends the ability to model and solve stealth games.