A generative design tool for designing the physical elements of communities

This research examines the use of generative design (GD) for neighbourhood design, specifically to optimize land use distribution and road networks. Neighbourhood design is inherently complex, involving trade-offs among accessibility, land use diversity, and stakeholder engagement. Traditional planning processes rely on manual workflows and disconnected analytical tools, limiting efficiency and stakeholder engagement. This study develops a GD tool that generates neighbourhood designs based on stakeholder preferences and constraints. The GD tool uses a genetic algorithm to optimize land use distribution and road-network layout. User-defined inputs include potential land uses, street networks, and weights derived from stakeholder consultation. Designs are encoded as numerical strings, evaluated using a weighted objective function that reflects stakeholder goals, and improved through iterative selection and mutation. Outputs are geodatabase features for 2D or 3D visualization. The GD tool supports iterative refinement through stakeholder feedback. A case study of Ookwemin Minising (formerly Villiers Island) in Toronto demonstrates responsiveness to varied objectives within reasonable computation times, supporting its potential for stakeholder-aligned, multi-objective solutions. While the weights used in the experiments in this paper are selected to demonstrate proof of concept, it is possible to derive weights directly from stakeholders using a stated preference survey. This research advances urban planning practice by offering an open-source alternative to commercial solutions such as Grasshopper or Dynamo (Revit), aligned with evidence-based design. This study provides a proof of concept for the GD tool, demonstrating its ability to optimize neighbourhood designs effectively and efficiently.