Designing for the Next Pandemic

How generative design and multi-agent simulation can help us respond to COVID-19 and future pathogens

Alex Tessier

Damon Lau

Rhys Goldstein


Architectural researchers around the world have explored generative design as a way of making better buildings, where “better” means achieving certain objectives like maximizing structural integrity, minimizing the use of material, or providing occupants with more natural light. As we head into the post-COVID era, another objective has now become top of mind everywhere: protecting people from viruses and other pathogens.

Autodesk Research has developed a generative design approach that addresses the complexity of virus transmission in buildings. Our method forgoes simple metrics and instead uses a full-fledged multi-agent simulation to capture the behavior of building occupants. The simulation also tracks the amount of virus transmitted among the occupants via air and surfaces, then uses the resulting statistics to produce safer designs. This research was recently published in a journal article titled Generative Design for COVID-19 and Future Pathogens using Stochastic Multi-Agent Simulation.

Why Simulate COVID-19?

When the COVID-19 pandemic first hit, people everywhere needed to change the way buildings were designed and used in the hopes of minimizing the spread of the virus. As a society, we did not have the technology to properly simulate COVID-19 transmission in buildings. So instead, we adopted simple rules like the following:

  1. Keep people 6 feet (or 2 meters) apart.
  2. Install plexiglass screens around desks.
  3. Make shopping aisles one way.

Unfortunately, simple rules like these do not fully address the complexity of a phenomenon like virus transmission. For example, staying 6 feet apart may protect a person from large, ballistic droplets that fall quickly to the ground, but not from small, aerosol droplets that mix throughout an indoor space. The now-famous Skagit Valley choir practice resulted in 53 attendees being infected, not all of whom could have been within 6 feet of the contagious individual.

What about plexiglass? An investigation of the Wellesley High School outbreak concluded that the installation of plexiglass screens had restricted air flow in an office and likely put people at greater risk. However, the investigators also stressed that plexiglass can be helpful in some situations.

The effects of one-way shopping aisles are also complex. Some modelers have reported that one-way aisles have the potential to reduce the risk of contracting COVID-19 while shopping. Other modelers find that one-way aisles put people at greater risk by increasing shopping time.

3D visualization of higher virus risk zones shown in red and orange

Simulation offers a potential alternative to relying solely on simple rules. Simulation allows us to account for the many factors that affect virus transmission, including the layout of a building, the policies that could be adopted by the facility managers, the ventilation system, and most importantly, people’s individual and social behavior.

We developed a custom simulation that tracks the movements and activities of individual occupants, or agents, within a digital model of an office. The simulation also tracks the amount of virus particles that are first emitted from contagious agents, then linger on surfaces and diffuse through the air, and either decay harmlessly or are ingested or inhaled by susceptible agents.

This type of multi-agent simulation allows us to predict the amount of virus that will likely be transmitted from infected to healthy individuals under different conditions, rather than relying on an indirect measure of risk such as the distance between desks or how much plexiglass has been installed.

This simulation allows for an architect to systematically test how risky a given design is. Hotspots where virus particles concentrate in high numbers are highlighted in a 3D visualization, and an overall risk score is calculated for the entire floor plan. The designer can also play an animation of building occupants moving through the space with the corresponding virus particles visualized as accumulating in the air and surfaces. This tool can help the designer learn which combinations of architectural geometry and occupant behavior tend to be riskier and which are safer.

Animation of virus particles accumulating in air and surfaces from contagious agents over time. Note the concentration of high risk zones in conference rooms and frequently trafficked spaces such as kitchen and lounge.

Generating Safer Office Layouts

However, manually designing multiple floor plans and exporting to a simulation can be time consuming, therefore limiting the number and quality of options that can be explored. But what if we can automate the process of generating a floor plan, testing it for virus risk, and discover new layouts which best minimize risk?

Our team created workflow to do just that.

We sought to redesign our Toronto office with the goal of minimizing COVID transmission. Leveraging DynamoBIM and Revit Generative Design, we built a model that generates thousands of building layouts, tests each one, and learns which types of architectural geometry result in lower virus risk scores. The system can then generate new layouts that take the best features of the highest performing designs. The resulting design options showed 20% lower virus air risk score than a baseline set of building layouts.

Generative Design workflow in Dynamo to create a new office layout, simulate for virus risk, and return analysis map.

Using RevitGD to generate and test hundreds of design options, seeking to discover the higher performing layouts in terms of virus air and surface risk.

Evolving Generative Design

The main novelty of our paper is the fact we extended an existing generative design approach with a detailed simulation capturing day-to-day actions of individual people in buildings. In a typical generative design workflow, design options are evaluated using simple metrics or spatial measures like the 6-foot distancing rule. These simple heuristics fail to capture the complex human-building-pathogen interactions that involve an element of randomness and unfold over time. This is why our team developed and employed a multi-agent simulation, which accounts for the uncertainty and time-dependent nature of human behavior.

Yet multi-agent simulation comes at a cost. To ensure our virus transmission predictions were statistically accurate, we needed to run dozens or hundreds of simulations for every design. In exchange for this computational demand, the approach is highly versatile and may one day become the gold standard for designing buildings as systems involving occupants and a host of other phenomena.

To learn more, please see our recent journal paper published in Sustainable Cities and Society.

Alex Tessier is Director, Simulation, Optimization & Systems at Autodesk Research.
Damon Lau is a Senior Research Scientist at Autodesk Research.
Rhys Goldstein is a Senior Principal Research Scientist at Autodesk Research.



Get in touch

Have we piqued your interest? Get in touch if you’d like to learn more about Autodesk Research, our projects, people, and potential collaboration opportunities

Contact us