VoroUDF: Meshing Unsigned Distance Fields with Voronoi Optimization

arXiv cs.GR details VoroUDF, a new reconstruction method for turning unsigned distance fields into triangle meshes with better geometric fidelity and cleaner topology. The pitch is especially relevant for teams working on procedural content, scan processing, or any pipeline that needs robust surface extraction from imperfect data.

The core idea is to replace fixed-grid meshing with a Voronoi-based optimization over movable seeds that can slide along the iso-surface. Those seeds are driven by a tangent-plane fitting term and a repulsion term, while surface samples are queried once from the UDF and its gradient. That design is meant to sidestep brittle inside/outside estimation, restrictive lookup tables, and noisy topology cleanup.

The method is designed to handle non-manifold geometry, sharp edges, and open boundaries, which are exactly the cases that tend to break conventional reconstruction workflows. For game teams, that matters if you are converting scans, authoring destructible assets, or building tools that need dependable mesh output without a lot of manual repair.

The paper says the result is a lightweight mesh that is better suited to downstream real-time and interactive applications. In practical terms, that could mean less cleanup for technical artists and tools programmers, and more usable geometry for runtime systems that need stable assets rather than just visually plausible surfaces.