Initially, a discrete growth model was developed to create layouts based on modules with a minimum plot area of 216 square meters, defined by CMDA rules. The model connected these modules to form road networks and communal areas. This approach was used to optimize for a maximum number of units while considering constraints such as vehicular circulation, area optimization, and open space relationships. However, the model's efficiency was limited, as it tended to produce predictable patterns rather than diverse or "exciting" outcomes that fully correlated with the specified relationships. 

Computational Framework

To address these limitations, the project pivoted to a Generative Adversarial Network (GAN) model for the later stages. Using datasets derived from the urban fabric of Miami, the GAN was trained to generate basic building volumes based on contextual features and road systems. This approach allowed for a more complex and nuanced generation of designs. The final outputs were then rigorously evaluated based on key performance indicators: maximizing pedestrian circulation, minimizing vehicular circulation, and maximizing the number of residential units.

This two-phase methodology demonstrates a comprehensive approach to complex urban design challenges, showcasing an evolution in computational strategy to achieve more effective and innovative results.

Computational Framework

Outputs