In South Korea, high-density residential complex design is inherently complex, demanding both rule-compliant morphology generation and multi-criteria performance optimization simultaneously. Automated design research has addressed this through two dis...
In South Korea, high-density residential complex design is inherently complex, demanding both rule-compliant morphology generation and multi-criteria performance optimization simultaneously. Automated design research has addressed this through two distinct paradigms: Exploratory Generative Design (EGD), which prioritizes morphological diversity, and Performance Optimization (PO), which targets environmental efficiency. EGD paradigms predominantly employ bottom-up control, managing individual buildings or units as discrete variables. This bottom-up structure produces compounding gaps. First, a dimensionality gap: as spatial entity count grows, the parameter space expands proportionally, rendering site-scale optimization computationally intractable. Second, a workflow gap: because exploration and evaluation are computationally slow in the same loop, existing frameworks decouple them into sequential stages, preventing the simultaneous discovery of morphologically diverse yet optimal solutions. In terms of PO, there is a coverage gap: the limited existing study simultaneously addresses all five performance dimensions, including FAR, daylight, south-facing orientation, view quality, and circulation. This study proposes an integrated top-down Voronoi framework that resolves all three gaps within a single unified loop. Rather than managing discrete buildings, the framework perturbs five top-down Voronoi grid parameters: seed angle (θ), scale factor (φ), irregularity factor (α), and two cell-category-based height factors (Hl, Hs), to control all downstream design variables simultaneously, and holds search space dimensionality constant regardless of site scale. A hierarchical repair protocol (rotation → type conversion → height reduction) corrects regulatory violations internally before fitness evaluation, substantially reducing invalid phenotypes and minimizing reliance on penalty functions. A weighted single-objective RBFOpt engine then evaluates five performance objectives concurrently: FAR compliance, daylight access, preference orientation, view quality, and circulation integration (RRA). Validated on two pilot sites under different local ordinances, the framework demonstrated constant search- space dimensionality regardless of site scale, produced approximately 60% regulatory-compliant layouts across 1,000 iterations, and confirmed that the Voronoi-based parametric system successfully controls both morphological diversity and density at the global level, especially the irregularity parameter (α) and building height. The Voronoi grid also demonstrated stronger adaptive control over design outcomes than equivalent orthogonal grid approaches, alongside faster convergence. The framework thus offers designers a scalable tool that reconciles density targets, regulatory compliance, and environmental quality in a single automated run. Also, it generates a diverse pool of valid layouts within a short time, replacing weeks of manual iteration.