AI and Machine Learning

Case study: Autodesk’s generative design artificial intelligence

Generative design sped up a building project for a research space

By Karen Pierce October 23, 2020
Courtesy: UNIFI Labs

While there are numerous examples of projects implementing both artificial intelligence and virtual design across the construction sector, generative machine learning is arguably one of the most interesting.

Generative design is an artificial intelligence-guided tool that mimics nature’s evolutionary process. A computer algorithm experiments with an initial design and then modifies it repeatedly to see whether it better fits the desired outcome parameters. After millions of attempts, it eventually produces a solution. Usually, it is better than anything that a team of experts could design, making it one of the most potent artificial intelligence applications.

Autodesk was quick to see the technology’s potential and decided to put it to use in designing its new office and research space in Toronto’s MaRS Discovery District. The company wanted to facilitate happenstance interactions between various innovators and personalities. The resulting cross-referencing and interaction of ideas, it hoped, would aid competitive advantage.

Nevertheless, the company knew that getting designers to manually produce such a configuration would be a challenge because it is not an easy thing for people to think through naturally. Designing a space to encourage random interactions is a big challenge for human designers, but artificial intelligence can do it with relative ease. Autodesk started setting up the parameters that would allow it to use artificial intelligence to design a space generatively.

The team had the concept set in their minds; It was just a matter of providing the generative algorithm with data that it could use to develop an optimal layout. Once the generative algorithm was developed, the artificial intelligence could experiment with room features and fictitious human agents to see what type of setup would achieve Autodesk’s goal.

The entire process was over very quickly. Using the latest supercomputers, Autodesk ran through approximately 10,000 iterations in just a few days. Once completed, the computer gave a short list of arrangements to meet the goals while complying with constraints. Human architects looked over the designs and decided which they thought was best from a broader perspective.

This innovation flipped the design process on its head. Architects and planners turned from artists to reviewers. Their role went from assembling the design to review it. They provided high-level input but no longer needed to drill down into the details. Autodesk recognized this was merely the beginning as they realized that they could use generative design to optimize a range of other factors vital to an efficient building project.

They concluded that generative design could maximize construction efficiency, budget efficiency and a range of other factors. Thus, Autodesk could apply this particular artificial intelligence technique to the final design and the planning process. The idea had the potential to slash their costs, reduce construction time and offer practical compromises with other end goals specified in the original parameters. In short, it made rational, data-driven, multivariable optimization feasible.

These high-level overviews would then feed into virtual design considerations, improving the information available to those involved in the construction and planning process. Clients would be able to see trade-offs between different directions in ways that were not possible in the past.


Karen Pierce
Author Bio: Karen Pierce is a senior BIM specialist at UNIFI Labs. She has 10 years of experience in the AEC industry and has led the charge on expanding BIM technology in various roles.