Generative design is the automated algorithmic combination of goals and constraints to reveal solutions.
We are a practical startup. We build tools users can use right now and can implement into their workflow today. Over the last 2+ years of working on TestFit, we have been incrementally working towards building-level generative design. Along the way we took some detours for our customers and built things like manual mode, the unit editor, and multiple sites. But the long-term goal has been building-level generative design.
With TestFit, our thinking was to create one quality scheme, and to do it in a few milliseconds. This way users could rapidly prototype different schemes. One thing that our users requested was the ability to compare different schemes easily. With our schemes tool, users can now do that!
Invariably, when presented with more than a few options, it becomes hard to sort out what is good and what is bad–and we have a hard limitation: the human mind can really only compare four things at once. This is why data-driven design tools typically have a parallel coordinate graph to filter out solutions that are not ideal, and to present a narrow range of options. We have built our own parallel coordinate graph, comparing design, zoning, and financial information.
Automatically Generating Schemes
The blueprint for this started with the first version of TestFit. With the 1.0 update, we launched presets. Presets enable users to create their own lists of parametric values, and to simply click between them (think 5 over 2 or type III wrap). TestFit also has a feature called options, which are simply different building mass layouts for the site. The cross-product of “presets” and “options” revealed that there were already several schemes available for users. Here is what automatically generated schemes looks like:
We think, for starters, that 64 schemes should be more than enough. We currently exist in a world where fewer than three schemes are ever test-fitted per site. With a 20-fold increase, we should be able to find more opportunities to solve real estate deals, or at least the data to kill them.