Assistant Professor Audrey Olivier
When Audrey Olivier is staying in New York City, she likes to take advantage of the city’s parks to go for an early morning run. As an assistant professor of civil and environmental engineering at USC, Olivier’s research combines physics-based modeling, data analytics and uncertainty quantification to support resilient urban infrastructure and bring about the “smart cities” of the future.
Soon enough, recreation becomes research. A recent experience in Riverside Park relates to a NSF-funded project Olivier is working on with researchers at Columbia University, Henry Lam in operations research and Andrew Smyth in civil engineering, in partnership with the New York City Fire Department.
During her morning run, Olivier witnessed an accident – a crowd was gathering around a fallen cyclist, waiting for the emergency services to arrive. As specialist in probabilistic data analytics, Olivier started doing rapid calculations in her head, identifying the unknowns and missing links. How long would it take for an ambulance to reach the patient, and what invisible chain of decisions was shaping the timeline?
Designing for uncertainty
When someone calls 911, dispatchers decide which ambulance to send, where to route it, and which hospital can handle the patient – all while faced with constant uncertainty from unpredictable emergency call volumes and traffic conditions, among other factors.
“Every delay in care carries risk,” Olivier reflects. “We can’t eliminate uncertainty from the world, but we can design systems that account for it – systems that ensure patients get the help they need, when they need it.”
Supported by a Smart & Connected Communities grant, the goal of the current research study is to develop a probabilistic simulation (“digital shadow”) for emergency medical systems (EMS) agencies to test new policies in a safe virtual setting, enabling more informed and adaptive decisions.
Why is this so important? Faced with crucial choices in complex and rapidly changing situations, EMS agencies typically rely upon simple, rule-based approaches – for instance, to always send the nearest available ambulance. Olivier points out the flaw. “What’s best for one call might not be best for the system as a whole,” she explains. Dispatching the closest ambulance now may leave a neighborhood uncovered minutes later, or send too many patients to a crowded hospital. The rules that keep the system running smoothly in theory often break down in real-world contexts.
Data-driven decisions
Mapping of ambulance speeds across New York City informs probabilistic EMS decision-making
Unlike a fully integrated “digital twin,” which continuously updates and adapts policies in real time, a digital shadow primarily runs simulations offline. However, Olivier’s group is pushing the model to absorb real-time data feeds such as traffic conditions. This allows them to explore how ambulance travel times, surges in demand and hospital bottlenecks might play out under thousands of possible scenarios – cascading effects that fall under the radar of standard rule-based approaches.
At the heart of the project is a shift in perspective. EMS performance is often measured in averages: travel times, coverage and hospital loads. Olivier wants to go further. “Averages hide the risks that matter most,” she says. “We need to also take into consideration the rare but critical events – those times when an ambulance takes far too long to arrive.”
Another challenge rests in the data itself. EMS records reflect the policies in place at the time they were collected – predicting how the system will behave under a new policy therefore requires extrapolating beyond existing data. The team is developing probabilistic models to quantify uncertainty, rare-event simulation methods to explore unlikely but dangerous scenarios and risk-averse optimization techniques that can recommend strategies designed to avoid catastrophic failures – not just improve everyday performance.
That said, Olivier maintains a critical attitude to her own methods. “We want our models to not just produce numbers, but to be transparent about how much confidence we can place in those numbers,” she explains.
From equations to emergency situations
The collaboration with the New York City Fire Department is central. “Few research projects gain this level of access to a large urban EMS system,” says Olivier. “Working with FDNY inspires the research and provides the necessary data to make an impact,” says Olivier.
This project builds on a successful past collaboration between Olivier’s academic team and FDNY, in which the team developed algorithms to fuse information from multiple data sources and better inform decision-making for a specific piece of the EMS system, namely to-hospital transports. This initial project was key to building trust in the methods – FDNY implemented at scale the hospital recommendation algorithm developed by the team, which yielded a ~1 minute decrease in average to-hospital transport time for affected transports, as well as a load balancing algorithm to proactively mitigate hospital overloading.
This new SCC project further pushes the boundaries of EMS optimization by taking a holistic approach, fully integrating within the digital shadow the multiple complexities and decisions of EMS operations. “Ultimately,” says Olivier, “this research project will yield viable policies that can work for FDNY. That’s very unique.”
EMS is just one example of a cyber-physical system: a large, complex network where human decisions interact with physical infrastructure under uncertainty. “The study has the potential to impact not just EMS, but the study of cyber-physical systems more broadly,” Olivier projects. “For instance, power grids, transportation and other complex systems facing uncertain environments.”
Methods pioneered in EMS could ultimately improve resilience across many of these domains. A digital shadow serves as a prototype for data-driven decision making in any number of fields – helping humans to focus on the actions and relationships that foster connected communities.
Published on December 3rd, 2025
Last updated on December 3rd, 2025
