WildFab: Multi-Axis 3D Printing from Models in the Wild

WildFab is a new computational pipeline for multi-axis 3D printing that aims to work on the kind of geometry teams actually get in production: scanned point clouds, implicit models, voxelized optimization outputs, and non-watertight meshes. Instead of forcing a clean-up pass first, it tries to derive spatial toolpaths and printer motion directly from the source model.

The core idea is a hybrid query representation that combines a neural unsigned distance field with a regularized generalized winding number field. In practice, that gives the system both differentiable distance/direction queries and a more reliable way to resolve near-surface ambiguity, distinguish solid from void, and localize surfaces when the input is messy.

For fabrication, WildFab computes toolpaths by iteratively projecting points between optimized guidance-field level sets and gradient-magnitude ridges. It also uses a coarse-to-fine collision checking scheme: a fast UDF-based rejection stage filters likely collisions, then reg-GWN verification resolves the remaining cases for both solid and shell components.

The practical payoff is support-free fabrication from models that would normally need manual repair, with the authors showing successful results on non-manifold parametric surfaces, topology-optimization outputs, implicit models, raw scans, and non-watertight meshes. For game teams, the interesting angle is less about consumer printing and more about how robust geometry processing can preserve authored intent when assets move from digital content into physical fabrication, props, or rapid...