10 tips. De-risk your deals.
We've interviewed hundreds of property developers from all over the world.
We put together this email masterclass to distill the top 10 tips of wisdom we learned from them.
Get the 10 tips.

Urban AI

September 23, 2019

Giraffe has been lucky enough to work with UDIA NSW, UNSW and COX Architecture on the Urban AI project. This has been a project that connects computational urban design with a valuation model.


As Sydney grows further to the west, a pressing policy question is the form it takes. Suburban carpet could spread to the border of the Blue Mountains, or a different kind of built form might evolve. The Western Sydney Airport and committed Stage 1 Metro are catalytic pieces of infrastructure that inform how a new kind of city might be imagined - one that is nodal, transit based and integrated with the landscape.

An enormous amount of thinking and working as been done in designing and imagining this future city. Our contribution with the UDIA has been to build a computational design model that generates synthetic cities based on a combination of constraints layers and parameters (such as FSR, height of building and setbacks.)

The algorithm combines these inputs into a data driven, 3D synthetic city model. Because our work has been developing an algorithm - rather than a design - we can rapidly test different options, like the forms below.

Different urban typologies

Combining multiple threads

Responsive built form

It was important to us that the generative design algorithm creates complex, diverse and realistic built form. Only then do evaluations of the synthetic city make sense. Built form emerges from the intersection of multiple drivers - commercial, engineering, environmental and political. The only way to get realistic built form is to to consider the city from these multiple perspectives.

To address this we built into the algorithm different floor plates for commercial and residential uses, different approaches to dividing cadastral boundaries, and the handling ofsetbacks and building separation. We also spent time differentiating between dwelling typologies - like detached houses, missing middle medium density and apartments. Open space and social infrastructure requirements (like schools) are also automatically added by the algorithm based on the population increase.

One pleasing outcome of the project was that because we are generating the synthetic city computationally we didn't have to rely on 'handmade' FSR and height assignment. We set FSR's and heights as functions of distance from key infrastructure to drive finely graduated and controlled built form - as in the diagrams below of the missing middle and the transect.

The algorithm allows the creation of synthetic neighbourhoods that follow the transect pattern, or neighbourhoods that reject it.

More options, better conversations, better outcomes

The city generated by the computer is not perfect, but we have aimed to get it is close to what might be built. The model generates the numbers which planners need- dwellings, jobs and population - but it also shows a form the city might take. We hope this will drive better discussion and communication between planners and citizens. The modelling is also extremely fast - fast enough that you can change the urban fabric on the fly. This allows better comparisons to be made.

Keeping the human involved

The algorithm allows human over-rides. This means a designer interacts with the model, pushing and pulling the generated city around in a kind of collaboration with the computer until an optimum is discovered. Human intuition is vitally important in design - the computer should always work for the person and not the other way around. Cities are built for people, and not people for cities.

Leveraging the computer

The built form generated by the UrbanAI algorithm is not just geometry. It is a rich, machine-readable representation of the city which we use to drive a valuation model. This digital representation of the city could be used by any other computational model to derive rapid results - infrastructure requirements, transport models or flood models.

We have focused on understanding the economic impact a new city might have. We trained a machine learning algorithm on price data, and then used the trained model to understand the aggregate value higher density development around potential metro stations might have.

This project was a proud collaborative effort between

- Rob