Predicting Mosquito Flight Behavior
Interactive companion to Predicting mosquito flight behavior using Bayesian dynamical systems learning (2026) — MIT, Georgia Tech & UC Riverside
Mosquitoes transmit diseases like malaria, dengue, and Zika, causing over 700,000 deaths every year. To understand how they find us, we tracked thousands of mosquitoes in 3D using infrared cameras, then used a Bayesian machine learning technique to build a mathematical model of their flight. This simulation lets you interact with that learned model and see, in real time, how mosquitoes respond to different sensory cues.
Sensory Cues Studied in This Work
Vision
Attraction to dark objects
CO2
Response to exhaled carbon dioxide
Vision + CO2
Combined cues, as with a real human
(mosquito response may be slightly delayed) Switch models to see different flight behaviors Drag the slider to add more mosquitoes
Key Findings from the Paper
3D data points collected — the largest mosquito tracking dataset ever created, covering more than 477,000 flight trajectories.
Visual and CO2 cues don't simply add up linearly. When combined, they trigger a distinct host-seeking behavior that can't be explained by either cue alone.
The learned model accurately reproduces real mosquito flight around humans, opening the door to better-designed traps and repellents.
About the Research
"Predicting mosquito flight behavior using Bayesian dynamical systems learning"
Christopher Zuo*, Chenyi Fei*, Alexander E. Cohen*, Soohwan Kim, Ring T. Cardé, Jörn Dunkel & David L. Hu
MIT · Georgia Institute of Technology · UC Riverside · 2025
Funded by the NSF Physics of Living Systems network, Schmidt Sciences (Polymath Award), the MIT MathWorks Professorship Fund, and the Department of Defense NDSEG Fellowship.