Exploration of an Assembly-Based PCG Framework for 3D Asset Generation

Abstract

Contemporary procedural content generation frameworks typically target a single content category; cross-domain frameworks that also retain artistic control remain under-explored. This thesis explores an assembly-based PCG framework for modular 3D asset generation across two distinct categories of 3D assets: buildings and vegetation. The framework was developed and evaluated across two cycles utilising the Design Science Research Methodology (DSRM), with the first establishing a baseline implementation and the second extending the descriptor schema and evaluation methodology in response to findings from the first. Three evaluation criteria guided the assessment: authorial control over the output distribution, schema reuse across structurally distinct 3D assets, and the expressive range of the generated output, measured via expressive range analysis (ERA) The descriptor schema was applied to both domains without modification, with cascade activation supporting the deeper hierarchical structure of vegetation. Authored constraints were enforced reliably across all generations of the buildings domain. The expressive range proved one-dimensional for buildings, whose variation was purely combinatorial, and two-dimensional for vegetation once deformation operations introduced a parametric axis independent of selection. These results indicate that a single descriptor-based schema can support structurally distinct content domains, with expressive range determined by descriptor authoring rather than by inherent framework constraints.