Validating Navmesh using Geometry: Voxel-Based Analysis with Prioritized Exploration

Abstract

Navigation mesh (Navmesh) inconsistencies affect the player experience by directly impacting the navigation systems used by non-playable characters (NPCs) in game environments. While navmeshes are generated from world geometry using well-established algorithms, environments change throughout development as terrain is adjusted and assets are moved or replaced, resulting in mismatches between the navmesh and the actual environment. Existing automated approaches attempt to detect navigation issues using exploration agents and reinforcement learning techniques. However, since these methods rely on the navigation data itself or evaluate navigation behavior indirectly, they do not explicitly verify whether the navigation representation reflects the walkable space defined by underlying geometry. This paper presents a framework for validating navigation meshes through an independent, geometry-driven analysis of navmesh correctness. The approach reconstructs walkable space directly from environment geometry using a voxel-based representation, followed by constraint-aware traversal and connectivity evaluation. Validation is formulated as a prioritized search problem over the voxel space, where reinforcement learning guides sampling toward regions more likely to exhibit inconsistencies. At each sampled location, reachability derived from the voxel representation is compared against reachability obtained from the navmesh via engine-level queries. Experiments across multiple large-scale open-world game environments show that the approach consistently lowers exploration effort while maintaining similar defect detection coverage. The framework runs offline within the game engine and can be integrated into automated quality assurance pipelines. Since the method relies on geometry, it can be adapted across game engines with minimal changes, making it suitable for production deployment.