Topological pruning and ranking in generative schematic space planning: a conditional SDE–CSP pipeline for single-family housing
Probabilistic generative models increasingly support early-stage schematic space planning, yet their outputs do not guarantee compliance with architectural constraints. This paper proposes a hierarchical pruning-and-ranking pipeline that couples a pre-trained conditional generative model based on stochastic differential equations (SDEs) with feasibility checking grounded in the Constraint Satisfaction Problem (CSP) formalism. Operating at the bubble-diagram level, three stages remove geometric violations, enforce required and forbidden adjacencies and rank feasible solutions by adjacency, circulation and area scores. In a single-family housing scenario with 1,000 generated candidates, the pipeline achieved a 32.2% feasible yield, against 12.9% for a CSP-only search under the same sampling budget. Top-ranked layouts reached mean adjacency compliance of 0.999 while preserving the raw pool's topological breadth. In a formative review, fifteen expert practitioners and researchers rated functional adequacy and practical utility favourably. The resulting vector-based diagrams provide verifiable inputs for downstream design development and building-performance workflows.
Authors
- Gu Hyeongmo
- Seungyeon Choo (ORCID: https://orcid.org/0000-0002-8463-5307)
Publication Details
- Journal
- Figshare
- Published
- 2026-11-17
- DOI
- https://doi.org/10.6084/m9.figshare.35037349
- Primary Topic
- Architecture and Computational Design
- Type
- article
- Field-Weighted Citation Impact
- 0.00