AI simulation suggests 14.5% shorter ambulance travel times

On a New York street, an ambulance with its siren on has to negotiate a traffic system built for ordinary drivers. Researchers at NYU Tandon have now built an AI framework that predicts how quickly it will move through city traffic, giving the FDNY a virtual way to test response strategies before changing streets in the real world.

The pressure is measurable. FDNY's average response time to medical emergencies rose from about 10.4 minutes in 2015 to nearly 13.7 minutes in 2023, an increase of almost 32%. An ambulance can legally pass through red lights and maneuver around stopped traffic, but it still depends on nearby drivers recognizing the siren and making room. Tools designed for normal traffic, including Google Maps, cannot reliably capture that behavior.

The research team built a high-fidelity Traffic Digital Twin covering West Harlem and Morningside Heights, FDNY's M6 dispatch zone. The model combines detailed traffic simulation with real-world traffic and GPS data from nearly 1,000 FDNY ambulance responses in 2023. Led in the simulation and calibration work by C2SMART senior research associate Fan Zuo, the team trained EMVAID to predict ambulance speeds on individual road segments. Joseph Chow led the research, with C2SMART director Kaan Ozbay as senior author.

The model found that background traffic speed was the strongest predictor of ambulance speed, followed by congestion. Two-lane streets generally helped ambulances move faster than one-lane streets, while adding more lanes brought little extra benefit. Streets with unprotected bike lanes were associated with slightly faster ambulance speeds, possibly because they gave emergency vehicles more room to pass stopped traffic.

Concretely, emergency planners could use the framework to compare strategies in a virtual neighborhood before putting them into practice. In one demonstration, an optimized set of ambulance base locations reduced expected travel time by 14.5% and increased the share of calls handled by neighborhood-based ambulances from 41% to 55%. The result is not an FDNY recommendation: the researchers say it likely overstates the improvement because ambulances may already be responding to calls or returning from hospitals. The findings stayed stable when the team deliberately introduced prediction errors, and C2SMART says the approach could be extended beyond one neighborhood to estimate speeds across the city.