A first step toward the future of neighborhood design
Introducing our generative design tool, which can help generate millions of planning scenarios — and identify options that best reflect local priorities.
This post was co-written with Designer Brian Ho.
When designing a new neighborhood, planners, architects, and developers must weigh a wide range of competing objectives that impact quality of life. For example, increased density can generate more jobs and more housing options, but it can also mean more traffic congestion or taller buildings that cast shadows onto public spaces.
A holistic understanding of these tradeoffs is critical, since the built environment is very difficult to change, and communities must live with these choices for years to come. But planning work can be remarkably fragmented, lengthy, and costly, making it hard for professionals — let alone communities — to fully evaluate and understand all their options.
Today, the various experts on a planning team often run separate analyses to produce a neighborhood design: an architect uses one type of software to simulate sunlight, an engineer uses another to plan streets, a real estate developer models economics in a spreadsheet, and so on. The time and cost needed to coordinate all these competing elements often means a project can only afford to develop a handful of designs for the project team, with limited insight into how these options will impact the community.
To help address these challenges, we’ve been developing a generative design tool that can do two things existing planning tools can’t. First, using machine learning and computational design, it can help planners generate not just one or two but millions of comprehensive planning scenarios. Second, it can help evaluate all kinds of impacts these different scenarios could have on key quality-of-life measures, producing a set of options that best reflects a community’s priorities.
As we continue developing this tool, we’d like to share some initial details about how it can help planning today — as well as some exciting ways it could help cities in the future.
How it works: A case study
To make one thing clear from the start: the generative design tool doesn’t automate the urban planning process or eliminate the need for human-driven design. Instead, it provides a set of features that can empower planning teams to do their job even better.
The generative design tool starts with a set of foundational information that can include a geographic area, physical or regulatory qualities of the place, and (if available) existing development plans. Guided by a designer’s input, it can then draw from environmental (non-personal) data that’s commonly used by engineers, architects, and developers to help plan a neighborhood: things like street layouts, block orientations, real estate economics, weather patterns, building heights, and more.
Using all that information, the tool can “generate” a series of possible scenarios — along with their expected performance — for planners to consider using or refining as the design process evolves. Often it arrives at designs that traditional planning methods might not have found.
Let’s use a case study to show what we mean.
To demonstrate the tool’s capabilities, we conducted a preliminary study of possible configurations for a two-by-two block urban area targeted for future development. In this case, the area’s existing planning framework calls for a baseline development design like the one below, which aims to emphasize three community priorities: open space, daylight, and density (in this case via a proxy of gross floor area, which is basically total building space).
- Open space — 45.3%
- Daylight access — 49%
- Total gross floor area (GFA) — 1,513,144 square feet
That’s a great start, but the generative design tool can help explore more options and maximize the design’s impact. In an initial run, our planning team used the tool to generate thousands of design permutations, surfacing roughly 400 plans that outperformed the baseline plan on all three priority factors. Let’s take a look at a few:
Generative design #00530
- Open space — 5.2% increase
- Daylight access — 13.6% increase
- Total GFA — 24,243 additional square feet
This run made marginal changes to the baseline plan, removing some buildings but adding height to others. As a result, this design creates small increases in open space and density, with a large increase in daylight access.
Generative design #00469
- Open space — 3.31% increase
- Daylight access — 20.61% increase
- Total GFA — 196,710 additional square feet
This simulation made more moderate changes to the baseline plan, altering the built form in each block while preserving the big open space in the upper-right. As a result, this design creates a small increase in open space, but a medium increase in density and a large increase in daylight access.
Generative design #01140
- Open space — 12.6% increase
- Daylight access — 8.6% increase
- Total GFA — 496,781 additional square feet
This run made rather significant changes to the baseline plan, adding significant height to some buildings while removing others to free up space. The result is a medium increase in daylight access but large gains in open space and density. Such open space gains could create intimate neighborhood spaces that play an important role in supplementing the city’s park network.
Potential future application: Facilitating community input
As this case study shows, generative design doesn’t provide answers — for example, weighing the objectives and assessing the trade-offs, the community might decide it ultimately prefers the baseline design to all the alternatives. What the tool can do is make the design process more holistic and efficient, helping planners and the community make the most informed decision possible.
And while generative design isn’t currently designed as a community engagement tool, it’s not hard to imagine that usage being developed down the line.
Consider a future where stakeholders voice priorities and concerns during a community meeting, and the generative design tool simulates how these factors might play out in the actual development. That type of real-time assessment and transparency could be a big boost for participatory planning.
The tool also has the ability to get smarter as time goes on. With machine learning, a generative design simulation can not only understand tradeoffs between various objectives like daylight and density, but also learn what’s worked (and what hasn’t) from years of existing neighborhood designs. Today, that type of institutional knowledge is trapped inside professionals, and difficult to pass on to communities.
In the end, if generative design does its job, it won’t just make the planning process more accessible — it can result in neighborhoods that truly reflect the needs and priorities of the communities they serve.
The authors wish to recognize the other members of the generative design team: Jack Amadeo, Josh Chappell, Betty Chen, David Huang, Okalo Ikhena, Amanda Meurer, Douwe Osinga, Kabir Soorya, Samara Trilling, and Dan Vanderkam. If you want to learn more, contact the team via this Google Form or email.