Mosquitoes rely on specific environmental cues—like the silhouette of a human body and the carbon dioxide ($CO_2$) we exhale—to track down their targets.
Now, researchers from MIT and the Georgia Institute of Technology have discovered how these visual and chemical signals shape the insect’s flight path. Using flight experiments under various sensory conditions, the team has built the first-ever 3D model of mosquito flight.
Their model, published today in the journal Science Advances, identifies three distinct flight patterns triggered by different sensory stimuli:
-
"Flyby" Mode (Visual Cue Only): When mosquitoes can only see a potential target, they swoop in quickly for a pass. If they don’t detect any other cues confirming a host, they simply fly away.
-
"Inspection" Mode (Chemical Cue Only): When they can’t see the target but can smell chemical cues like $CO_2$, they slow down and weave back and forth to stay close to the signal source.
-
"Orbiting" Mode (Combined Visual & Chemical): Intrepidly, when both cues are present—such as seeing a silhouette while smelling $CO_2$—the mosquitoes switch to a steady-speed orbit around the target, circling like sharks before landing.
The researchers believe this new model can be used to predict how mosquitoes respond to other sensory cues like temperature, humidity, and specific odors. These predictions could be a game-changer for designing highly effective mosquito traps and pest control strategies.
"Our study shows that mosquito traps need specially calibrated, multi-sensory baits to keep the insects engaged long enough to be captured," says Jörn Dunkel, professor of physical applied mathematics at MIT and a co-author of the study. "We hope this work establishes a new paradigm in pest behavior research, using 3D tracking and data-driven modeling to decode movement patterns and solve major public health challenges."
MIT co-authors on the study include mathematics postdoc Yifei Chen, and Alexander Cohen (PhD '26), who recently completed his PhD in chemical engineering under Dunkel and Professor Martin Bazant. They were joined by Georgia Tech collaborators Christopher Zuo, Soohwan Kim, and David L. Hu (BS '01, PhD '06), as well as Ring Cardé from the University of California, Riverside.
Digital Flight
Due to their devastating impact on human health, mosquitoes are widely considered the world's deadliest animals. These blood-feeding insects transmit malaria, dengue, West Nile virus, and other fatal diseases, claiming over 770,000 lives annually.
Of the roughly 3,500 known mosquito species, about 100 have evolved to target humans, including Aedes aegypti. This species relies on a cocktail of sensory cues to locate its hosts. Traditionally, scientists have studied these attractants using wind tunnel experiments—releasing $CO_2$ and observing where and when the insects land. However, prior studies rarely captured the actual dynamics of how mosquitoes fly when searching for a host.
"The big question is: how do mosquitoes navigate to a human target?" says Chen. "Previous experimental studies looked at which cues might be important, but none had analyzed the behavior in a highly quantitative way."
At MIT, Dunkel’s research group specializes in developing mathematical models to describe and predict the behaviors of complex living systems—ranging from how worms untangle themselves to how starfish embryos grow and swim, and how microbial communities evolve.
The spark for this mosquito research ignited after Dunkel gave a talk at Georgia Tech. David Hu, an MIT alumnus and now a professor of mechanical engineering at Georgia Tech, proposed a collaboration. Hu’s lab was running mosquito experiments at a Centers for Disease Control and Prevention (CDC) facility in Atlanta to study sensory responses. Could Dunkel's team use this experimental data to identify key flight behaviors that might help scientists control mosquito populations?
"One of the initial motivations was to design better mosquito traps," Cohen says. "Understanding how they fly around humans can also help us better understand how to avoid getting bitten."
Tuning Into the Cues
For their new study, Hu and his Georgia Tech colleagues ran experiments with groups of 50 to 100 Aedes aegypti mosquitoes. The insects flew freely inside a rectangular, slightly tilted white enclosure where a network of cameras captured highly detailed, 3D trajectories of each mosquito. To test different sensory inputs, the team placed a physical object in the center of the room:
-
Visual Cue: A black Styrofoam sphere mounted on a stand, contrasting sharply against the room's white walls.
-
Chemical Cue: A white sphere connected to a tube pumping out $CO_2$ at a rate mimicking human respiration, with no visual contrast.
-
Combined Cue: A black sphere releasing $CO_2$.
-
Human Trial: The team also tracked how mosquitoes behaved around a human volunteer wearing a suit that was half-black and half-white.
Across 20 experiments, the team generated over 53 million data points and captured more than 477,220 individual mosquito flight trajectories. Hu shared this massive dataset with Dunkel’s group at MIT, who used the raw measurements to construct the mathematical flight model.
"We started with a very broad class of dynamical equations. Initially, the equations to predict flight paths were incredibly complex, with many terms factoring in the relative importance of visual vs. chemical cues," Dunkel explains. "By iteratively matching the equations against the physical data, we stripped away complexity until we had the simplest possible model that still accurately fit the observations."
Ultimately, the team distilled a streamlined model that precisely predicts flight paths under visual, chemical, or combined conditions. What fascinated the researchers was that when both cues were present, the mosquito's flight path was not a simple mathematical "superposition" (addition) of the two individual behaviors. Rather than just mixing swooping and hovering, the mosquitoes adopted an entirely distinct, cohesive orbiting path.
"Our findings show that traps need to offer highly calibrated, multi-sensory baits if they want to keep the mosquito's attention long enough to capture them," Dunkel notes.
Cohen adds, "Obviously, humans emit many other signals—like heat, humidity, and skin odors. For the species we studied, visual contrast and $CO_2$ are the heavy hitters. But this modeling framework can absolutely be adapted to look at how other species react to different sensory cocktails."
To make their findings accessible, the researchers built an interactive app that runs their mosquito flight model. Users can experiment with virtual objects, adjusting parameters like the number of surrounding mosquitoes or the types of active sensory cues, to see a real-time visualization of how the mosquitoes would fly.
"Our ultimate dream was to build a quantitative simulator for testing trap designs," Cohen says. "Now that we have this model, we can genuinely start designing smarter, next-generation mosquito control tools."
This research was supported, in part, by the National Science Foundation, Schmidt Sciences, LLC, the NDSEG Fellowship Program, and the MIT MathWorks Professors Fund.
