Computational design of urban residential layout

The goal of this research is to develop an algorithm to generate a road and land-use plan for a regular or irregular residential development site. It starts from scratch without external intervention. The resulting algorithm has been named COULD (COmputational Urban Layout Design). Inspired by developmental biology, COULD commences with an ‘embryo’ (or ‘embryos’) and grows it into a plan rather than trying to begin with complete solutions, although it could also do the latter. The universal modules are the basic building blocks, which play similar roles to biological cells. They are ‘genetically identical’ with full developmental potential but will change their physical form to adapt to their own local context. The ability to adapt is maintained throughout the developmental process.

The modules can be seen as self-propagating agents that sort out their dependencies through local interactions. The genetic package that they all carry determines the rules of interaction. The separation between genotype and phenotype plays an important role in decoupling them. The generative genotype can therefore generate much from little, displaying the characteristics of an emergent system.

The developmental process is a guided search with random selection of branching points followed by operations to add links or make connections. Deletion operators guide development by removing branches that do not improve the outcome.

A hypothetical application, with the objective of maximizing combined everyone-to-everyone connectivity and dwelling density, has evolved highly interconnected street plans. However, no step is specific to the example; the operators will grow a road and land-use network under various specifications and constraints guided by an objective function. Making the process applicable to an actual development might require more constraints, operators and certainly an enlarged objective function. Cost and other goals can be included so long as each goal is functionally related in some way to every change in the plan made by the search procedure.

In numerical terms, the computer generated plans are better than the manually constructed grids under the current parameter settings. The analysis of the manual grids also revealed the diminishing returns in street connectivity, which has some practical implications in balancing high connectivity and the cost of providing it. The plans COULD generated feature high internal connectivity with low external permeability. The former should promote walking and cycling, enhance sense of community and even out the distribution of internal traffic. The latter should increase safety and discourage through traffic.