Building With Numbers: Urban Development and Machine Learning
Machine learning and cloud computing are supercharging urban development projects by helping teams find the best design and improve both sustainability and quality of life in neighborhoods.
Finding the right design for an urban project can have huge implications for the way people live—for example, by providing easy access to parks and public spaces or by meeting energy demands in a sustainable manner.
The United Nations expects that another 2.5 billion people will live in cities by 2050, and to ensure people can live comfortably, development projects will need to reflect their changing needs.
But it can take architects, developers, project managers and urban planners months to develop feasible ideas for urban developments. Normally, different teams focus on their own goals in their own areas of expertise, so when they meet to present their concepts to each other, they may not always be on the same page.
“A developer will talk in terms of spreadsheets and an architect will want to see a 3D model, while someone doing traffic analysis [of an urban development] will want to see a huge diagram,” said Douwe Osinga, director of engineering at Sidewalk Labs. “Everybody regroups, talks to their other people, they come back two weeks later, and it takes a very long time. The number of actual plans that you consider for a project is very low because you spend all this time talking across each other.”
Millions of design possibilities
This has serious implications for projects as it means the design may not be the best economy of scale, and residents might miss out on living in a neighborhood that gives them the best standards.
“The most obvious problem we often run into is the attention between affordability and quality of life,” Osinga said. “A developer wants to deliver a number of units, while the people who live in the city have certain requirements about their quality of life.”
Over the past year, Sidewalk Labs has been working on this problem for project managers. Last year, the team created a design tool prototype to help urban development teams find better design options than what common time and cost constraints normally allow.
In October 2020, Delve was launched. It uses machine learning and cloud computing to sort through millions of different combinations for urban development projects by looking at the core components of cities such as buildings, open spaces, amenities, streets and energy infrastructure.
The model was stress-tested by developer Quintain in a project at Wembley Park in London, England. The company used Delve to help design thousands of multifamily, build-to-rent homes on a 12-acre, mixed-use site.
To make the project financially viable, the team needed a number of units to be built, while meeting local daylight standards and preserving open space. But they were struggling to find a solution using traditional urban development methods.
By using Delve, the project was able to analyze several different possibilities and find versions of the plan that could deliver more units and increase daylight and space for residents. Teams on the project could input their priorities in the system and get feedback about what potentially worked and did not.
“The developer can see this is the profit margin, this is number of units I can offer,” Osinga said. “The architect can look at the 3D version, and from a traffic analysis point of view, you can see where you expect a lot of people to gather and where more empty spaces are.”
Improving sustainability and quality of life
Osinga expects Delve will help make developments greener as well. The system has a built-in utility demand tool, which can give estimates on measures such as electricity use, waste and water, and rooftop solar intake.
For example, a project manager can see how much sunlight a roof gets and then decide which energy system is needed to give the development the best return on investment.
They can also experiment inside the system and find out what different designs would look like without it being a time-consuming, expensive process.
“If you are a project manager you might wonder what would happen if we increase the park space by 5%,” Osinga said. “What is the cost of that? Allowing people to play with the system should give them the freedom to figure out that they actually can afford to have five to 10 more parks.