The Smart(er) City

Neighborhood_GridRotation_White_small.gif
 

A full “design space” featuring thousands of iterations for a potential neighborhood

 

Computational Urban Design and Analysis

At KPF Urban Interface, we are interested in both the future of cities, AND the future of how cities are designed. Whereas tech companies strive to rapidly prototype designs, large-scale master planning takes days, weeks, or even months to create a single option. With this bottleneck in mind, the KFPui team has developed a computational methodology capable of generating tens of thousands of designs, applying data science and machine learning workflows to aid analysis, and developing web apps (Scout, CityBot) to engage stakeholders and solicit feedback from a broader audience. This approach allows a cross-disciplinary team to quickly explore high performing scenarios capable of addressing the complexities of the 21st century city.

Our process starts with a question: Can the "smart city" be more than technology grafted onto a traditional design? By privileging performance over form, and merging the concerns of human experience (comfort, daylight, visual interest) with functional efficiency (sustainability, building efficiency, access to transit and green spaces), we are able to embed computational intelligence directly into the built form of the city. This iterative, analytical, simulation-based workflow is the future of architectural and urban design, creating cities that are resilient, flexible, functional, and livable.

Smart_Cities.gif

Learning from Cities

Recently, we’ve been working with Sidewalk Labs to apply our computational urban design methodology to their Sidewalk Toronto project. Together, we are learning from leading cities around the world, integrating their successes and urban DNA: New York City’s orthogonal street grid allows for a flexible, efficient, mixing of high-density uses; Barcelona’s rich variety of public spaces supports a wide range of activities; Tokyo’s subways serve as a model of Transit Oriented Development; and Rome’s medieval urban fabric generates such a pleasurable pedestrian experience that visitors come from all over the world to walk its winding streets. By layering computational strategies on top of KPF’s extensive global practice, we generate design strategies that optimize comfort, enjoyment, and livability as well as functional efficiency, helping to ensure that the “first city built form the internet up” feels like an authentic global capital, and not the tabula rasa condition of past attempts at creating smart cities.

 From top: New York City’s grid, Barcelona’s network of public spaces, Tokyo’s subways, and Rome’s winding streets.

From top: New York City’s grid, Barcelona’s network of public spaces, Tokyo’s subways, and Rome’s winding streets.

 

Building the Model

Working from global insights to local tactics, the next step is to identify the desired outcomes and how they relate to design. These can vary from one project to the next according to climate, culture, and context, but they will incorporate both evaluative metrics to quantify the relative performance of design candidates, as well as variables to define the physical characteristics of a design and how it is allowed to change. By understanding these three aspects of design (the urban morphologies that interest us, the factors that define formal variation, and the benchmarks against which we will measure success or failure), we have everything we need to build a comprehensive, iterative model to generate thousands, or even hundreds of thousands of design candidates.

 Key Design Drivers: selecting evaluative metrics such as Access to Parks and Transit, Energy Efficiency, and Urban Comfort, as well as physical variables like Street Grid, Amount and Type of Public Space, Distribution Activities, etc.

Key Design Drivers: selecting evaluative metrics such as Access to Parks and Transit, Energy Efficiency, and Urban Comfort, as well as physical variables like Street Grid, Amount and Type of Public Space, Distribution Activities, etc.

 Diagrammatic representation of Computational Urban Design Model

Diagrammatic representation of Computational Urban Design Model

Choosing Inputs

The foundation of the model is the inputs, the physical characteristics that are allowed to vary. These inputs include a wide range of features but below are two of the most fundamental:

The street grid, which can vary according to street width, rotation, location of intersections, and even type (orthogonal grid vs. radial nodes vs. medieval meandering).

A “Pixel Map” that manages the distribution of activities, including the total number of approved uses (such as residential, office, retail, manufacturing, education, parks, etc.), their relative proportions, and their organizational logic (is public space evenly distributed throughout the site in small amounts, clustered into large parks, or parallel to major thoroughfares).

 Rotating street grid with variable block size

Rotating street grid with variable block size

 Pixel Map with variations in green space organizational strategies

Pixel Map with variations in green space organizational strategies

 

Design Generation

Once the inputs have been determined, the model follows a set of procedural rules to automatically generate building geometry. These rules are sensitive to the activities that will take place in the buildings (i.e. Residential vs. Office), as well as myriad other factors such as building height limits, the existence of park or plaza space on the site, the size of the blocks and parcel, and the total amount of built area that should exist in the building. The resulting buildings can be broadly classified according to three main types: high density towers on podia, middle density buildings (either courtyards, bar buildings setback from their base, or large floorplate industrial warehouses), and low density infill buildings (such as townhouses and small offices). However, any building type can be used to populate the model as long as it can be defined by a set of rules.

 Parametric building typologies for high density (left), middle density (center), and low-density (right) parcels

Parametric building typologies for high density (left), middle density (center), and low-density (right) parcels

 In aggregate, the buildings, along with the streets and parks, form the geometric composition of each design iterations

In aggregate, the buildings, along with the streets and parks, form the geometric composition of each design iterations

 

Analysis and Results

Each iteration is evaluated according to a wide variety of performance criteria deemed relevant to a given project. KPFui has developed bespoke analysis tools to measure almost anything, from views, to daylight, comfort, sky exposure, solar radiation, wind, energy efficiency, visibility of buildings from/to landmarks, access to parks and transit, mobility, and even subjective characteristics like “visual interest”.

Once each iteration of the model has been analyzed, KPFui uses Scout, a web-based platform for data-visualization developed in-house, to sort through the results and identify trends. Scout was designed for collaboration and features both “beginner” and “expert” modes, so that design partners, clients, and other project stakeholders all have access to the data, and can participate in exploring the results.

For our most complicated projects, the UI team will employ machine learning models to predict the most promising candidates in a particularly large design space (i.e. 10,000 - 100,000 options), or to filter data sets of such complexity that human exploration proves inefficient.

 Clockwise from upper left: diagrams of tools for Outdoor Comfort, Daylight, Energy Efficiency, Visual Interest, Unobstructed Views, and Access to Parks & Transit

Clockwise from upper left: diagrams of tools for Outdoor Comfort, Daylight, Energy Efficiency, Visual Interest, Unobstructed Views, and Access to Parks & Transit

 The Scout Interface

The Scout Interface

 Using machine learning to look for trends in a complex data set with a clustering algorithm (left), and a Correlation Matrix for predictive analytics.

Using machine learning to look for trends in a complex data set with a clustering algorithm (left), and a Correlation Matrix for predictive analytics.

Together with project stakeholders, we can provide insight into complex challenges of designing urban environments for the 21st century, determining the optimal course of action for projects to navigate myriad, seemingly contradictory constraints. The rich variety and deep exploration provided by computational urban design positions KPF teams to create projects that build off of a strong foundation of embedded intelligence and informed decision making, creating industry-leading projects renowned for iconic design, sustainability, and human enjoyment.