Your browser is no longer supported

For the best possible experience using our website we recommend you upgrade to the newest version of your browser.

Your browser appears to have cookies disabled. For the best experience of Construction News, please enable cookies in your browser.

Welcome to the Construction News site. As we have relaunched, you will have to sign in once now and agree for us to use cookies, so you won't need to log in each time you visit our site.
Learn more

How generative design turns 10,000 ideas into one project

Half human, half computer: generative design can produce thousands of concepts then scientifically identify the optimal end-product – with major implications for the future of construction.

Think about how architects and contractors design offices now.

They would first consult with the client and perhaps the future occupants to work out what they require.

Do they want high levels of daylight? A certain number of meeting rooms or break-out spaces? How do they need seating arranged? Do they want high environmental performance?

Once they have an idea of this, they would come up with a few design options, from which the client could then assess before construction begins.

But what if you could measure more complex, qualitative data around how people work, then produce up to 10,000 design options that could be whittled down to one optimised solution that best suits the client’s requirements?

This is now reality thanks to a process known as generative design.

Until now, it has largely been used in manufacturing – to optimise the strength of aeroplane parts while minimising the material required, for example. But now it’s being applied to building design, as demonstrated by a project to create a new office space in Toronto, Canada.

Six-dimensional goals

Generative design is essentially a process that mimics nature, using cloud computing to take an evolutionary approach to design.

The process begins by defining a set of goals for your design as well as parameters that will restrict it, then uses a computer to crunch through all of the possible permutations to find the best option.

When software developer Autodesk decided to move into the MaRS Innovation District in Toronto, it saw an opportunity to carry out research into how the generative design process could be applied to construction, having already looked at how it could benefit manufacturing.

Generative design Autodesk MaRS GD Overview

Generative design Autodesk MaRS GD Overview

Design studio The Living, owned by Autodesk, took on the job. The first stage was to settle on the goals for the new office, which would define the aims for the software to work towards.

The team settled on six measurable goals: adjacency preference (how close staff wanted to be to other staff, meeting rooms and other facilities); work style preference (the suitability of the environment to the team’s preferred light and acoustic levels); buzz (how often shared spaces are activated through individual and team movement); productivity (the visual and audio distraction at each desk); daylight; and views to the outside (from both desks and circulating paths).

The latter four are all easily measurable and common considerations now when designing offices. The former two – adjacency preference and work style preference – were more complex to measure.

“We thought there might be some ways to do it that wouldn’t occur to a human, that were maybe even better, or certainly as good but slightly different, to what a human would come up with”

David Benjamin, The Living

“We asked what we needed for this project and whether we could possibly build that into a generative design model,” founding principal David Benjamin explains. “A normal project would probably have some surveys or focus groups at the beginning, and the information from that would be discussed and summarised by all of the stakeholders, then pushed into a design direction.

“In this case, we thought we could capture the raw data about, say, what does each one of 300 employees want to be near, in terms of other people, equipment and spaces like meeting spaces. And what does each individual or team of people want to have as the qualities of their workspace – more active and social or more quiet and heads-down?”

The ‘perfect’ computer problem

Through surveys and consultation, the team created a complex spreadsheet with all of the staff’s preferences and desires, deciding in particular to use generative design to help achieve these two ‘softer’ measurable goals. The software was given the task of computing the complex factors of human experience that are often overlooked or over-simplified (like a person’s position, seniority, attendance, feelings, preferred ways of working and levels of productivity).

“We could look at a layout of all of the people, desks and meeting rooms and ask if it was scoring high or low based on everyone’s combined preferences for adjacency,” Mr Benjamin says. “We could ask if it scored high or low on giving people the kind of work environment they wanted, the qualitative factors like social or active.

“Then things can get really interesting if you then have a computer model that can help navigate the trade-offs between things like daylight, which is pretty easy to calculate, and adjacency preference, which is much more difficult. There was no standard way to calculate it, so we invented our own.”

Generative design Autodesk MaRS GD Goals A

Generative design Autodesk MaRS GD Goals A

It is this combination of six goals, two of which were particularly difficult to measure, that made it a “perfect problem” for a computer to tackle, according to Mr Benjamin.

“If a human tried to solve that six-dimensional problem, especially with two of the dimensions involving a spreadsheet with 500 rows, it would be pretty hard for intuition to figure out the best design. [We] have been designing office spaces for hundreds of years, but we thought there might be some ways to do it that wouldn’t occur to a human, that were maybe even better, or certainly as good but slightly different, to what a human would come up with.”

A city in a building

So with the goals set, the team also had to come up with a geometric design for the space that the computer could work with – effectively setting the restrictions.

Through consultation with the stakeholders, the design team honed in on a concept of creating different work ‘neighbourhoods’, the idea being that each team has distinct characteristics and preferred ways of working.

Each neighbourhood would be self-sufficient, with its own medium-sized conference room and booths for making phone calls in a similar way that each neighbourhood in a real city has its own convenience stores, banks and social spaces. This led the team to divide the floor plate of the building into zones, with each zone becoming one of these neighbourhoods with a cluster of meeting rooms and shared spaces they could use.

“Nothing prevents you from going to another neighbourhood, but we wanted them to have their own unique characteristics and be efficient for the people inside them”

David Benjamin, The Living

“Just like a city, the idea is that you can go to a different neighbourhood to have a big meeting if you wanted – nothing prevents you from going to another neighbourhood – but we wanted them to have their own unique characteristics and be efficient for the people inside them,” Mr Benjamin says.

Each zone was now in a slightly different polygonal shape. The team then identified one side of that polygon where the meeting rooms would grow from, fitting functional rooms there and arranging the desks in the rest of the space. By allowing the shape, size and location of the zones and meeting rooms to vary, thousands of design options could be generated.

10,000 to 1

The process of actually generating the designs, using a prototype workflow on Autodesk software, took five days of computation.

The team ended up with 10,000 design options that it was able to evaluate and refine over time. It then used the computer to identify those that best satisfied the six goals, with options that organised the design into different clusters that satisfied certain goals over others.

“I like to say that, in contrast to traditional design where the architect might do a lot of thinking and come up with three different design options (each a fixed way of organising the space) in generative design each design option is a system of goals and constraints, and each system has potentially 10,000 design options,” Mr Benjamin says.

Generative design Autodesk MaRS GD Renderings

Generative design Autodesk MaRS GD Renderings

The set of 10,000 designs that The Living team looked at was actually the eighth set used. “In other words, we had earlier systems with different rules for the orientation of meeting rooms, or different rules for assigning people to spaces, and we had some that had a central spine organisation.

“In the end, we went with an option that had more of a clustered meeting room organisation. We did explore, in the way design always does, different ways of putting things together, but rather than each exploration being a single design, it was a set of 10,000 designs with rules and goals.”

The team then reviewed the designs with stakeholders to define which parameters were important, ensuring the human element was still at the forefront of the decision-making process.

“At one point on this project, when we were trying to decide between a final couple of designs, we had a really interesting discussion with the stakeholders where they said, ‘Although we have six goals, the productivity goal is probably most important to us. What would happen if we said the productivity goal got twice the weight of all the other goals?’” Mr Benjamin says.

“As we had a dataset of 10,000 designs, and since we have all of these algorithms to do this, we were easily able to score productivity twice as highly as all the other goals, re-sort the designs according to the highest-performing ones, and offer back to the discussion which design it would favour in that case.

“It gives us a way to have a data-informed discussion. It doesn’t replace human judgement or solve all differences of opinion, but it does provide transparency and common ground for having a discussion.”

What it means for contractors

The project is now under construction, with the first occupants set to move in during June this year. But the generative design process won’t end there – it will still be useful long after the building is in use.

After all, buildings change, as do the occupants – what happens if another team joins the business via acquisition and needs to use the space as well? Or what if a new building is constructed across the street, potentially affecting daylight levels? The algorithm can deal with it.

“We can use the exact model that created the high-scoring design [we selected], and we can see if the addition of a new team tips the scale enough in our score for adjacency preference that we should do a re-stack and rearrange where people are sitting,” Mr Benjamin explains.

“In addition, we’re going to continue to take surveys, and we’ve built a custom mobile app that will allow people to score as well as actively and passively contribute data to the system. That will allow us to look at certain assumptions we had – a certain way of scoring buzz, say – and see if that’s corresponding to how people are feeling.

“In generative design each design option is a system of goals and constraints, and each system has potentially 10,000 design options”

David Benjamin, The Living

“We can continue to refine our algorithms for the next time – or even say we’ve got something wrong about how we’re calculating productivity, so let’s re-run the whole dataset and see if we should move people, or convert one of the meeting rooms into something else, or something like that.”

For contractors, this process could also affect the way they work in the future. Plans are afoot to try to build a relevant cost model into the generative design goals, allowing the team to see if there are cost implications to pursuing a certain route. Likewise, sequencing calculations could be built in to see which design would be the quickest to build, or allow the most efficient sequencing of trades.

“We know that’s possible mathematically, but I think it would require a construction partner to help make sure we’re getting all the assumptions right,” Mr Benjamin says. “It has huge potential for being able to manage a lot of the complexity of the actual construction process.”

On this job, the contractor was not heavily involved in the process. But the design and construction teams did consult to get preliminary cost estimates and sequencing advice once the designs were narrowed down, leading them to change some of the model’s constraints.

“We looked at certain sharing of structures between two conference rooms, but in the end it would have been prohibitively expensive because it would require more custom steel, and we were able to recalculate the benefit we’d get from having two meeting rooms share a wall. We did a couple of things like that, but we think that’s just the tip of the iceberg.”

It’s still early days, then, for the use of this technology for construction. The implications for how companies work will take time to materialise.

But it’s safe to say it’s opening up new possibilities in building design – and therefore how buildings are delivered.

“This is a process that allows us to create designs that wouldn’t have been possible by a human along, and wouldn’t have been possible by a computer alone,” Mr Benjamin says.

“It’s this human-computer collaboration that’s so exciting and allows us to enhance our creativity as designers and go beyond some typical rules of thumb. I think it suggests this new world of possibilities for design outcomes.”

Have your say

You must sign in to make a comment

Please remember that the submission of any material is governed by our Terms and Conditions and by submitting material you confirm your agreement to these Terms and Conditions. Links may be included in your comments but HTML is not permitted.

Related Jobs