Using Medial Skeleton for Path Finding in Dynamic Stealth Games

Desirable difficulty in a game level is achieved by varying geometry of arena or placement of guards/obstacles inside it. Designing such levels requires a rigorous Design-Test cycle, for which the testing can be automated using Pathfinding algorithms. In this thesis, we propose an algorithm for roadmap generation for a player using a modified Medial skeleton of a 3D representation (2D spatial & 1D time) of a Stealth game level with elements that change over time. We define performance metrics and compare the paths obtained using our approach with those using Rapidly exploring Random Trees, on three different game levels. The paths obtained this way are more realistic (human-like), i.e. close to the center of walking area of the arena, keeping them at safe distances from obstacles and boundaries unlike other pathfinding algorithms that are “shortest-path" centric approaches.