In 2024, Tornyol Systems showed how a 1.4-ounce drone could detect and hunt mosquitoes using ultrasonic sensors. Two years later, the Paris startup recorded its first autonomous kill, but the victim was 1 moth, not a mosquito

A mosquito is an awkward target for a machine. It is tiny, fast, difficult to see and constantly changing direction. Now, a Paris startup has shown a 40-gram drone chasing a flying insect through the air and striking it without a human guiding the final move. The catch is important: the insect was a moth, not a mosquito. The demonstration is a real engineering milestone, but it is still far from proving that tiny drones can reliably clear mosquitoes from homes, gardens or cities.

A palm-sized drone has just crossed a strange engineering threshold: it hunted down a flying insect and struck it in mid-air without a pilot guiding the final move.

The Paris startup behind it calls the test a major step toward automated mosquito control. But there is an important detail. The insect was a moth, not a mosquito. And the demonstration used some tracking and computing equipment that will eventually need to be packed into a much smaller autonomous system.

Paris Startup Tornyol Systems Builds Micro-Drone to Hunt Mosquitoes

Tornyol Systems, a Paris-based startup founded by Alex Toussaint and Clovis Piedallu, released footage of its first air-to-air insect kill on July 14, 2026. The company is developing miniature drones specifically for mosquito control and entered Y Combinator's Fall 2025 batch. Its stated goal is unusually ambitious: use inexpensive autonomous drones to reduce the cost of mosquito control and eventually remove mosquitoes from defined areas.

The target in the public demonstration was a flying moth. That detail changes how the result should be interpreted. A moth is generally a larger and easier target than a mosquito, so hitting one proves that the tracking and interception system can work on a flying insect, but it does not establish the same performance against mosquitoes. Tornyol has previously demonstrated mosquito detection with its ultrasonic system, but a successful live interception of a mosquito is a separate technical challenge.

There is another important qualification behind the word “autonomous.” Reporting on the demonstration says the test used an external motion-capture system, with sonar processing performed on a computer that then sent movement commands to the drone. Tornyol has said it intends to move the processing and control functions onto embedded hardware. That means the demonstration is best understood as a major proof of the interception concept rather than evidence that the final consumer system is already completely self-contained.

The strange part is that the drone does not mainly need to see the insect

The heart of Tornyol's system is not a miniature camera searching for mosquitoes against a complicated background. Instead, the company uses ultrasonic sonar and a large microphone array to detect echoes from objects in the air. The approach borrows an idea familiar from radar and sonar systems: send out a signal, listen to what comes back and extract information from changes in the returning wave.

The LeSonar2 system developed by Tornyol uses 380 smartphone-style digital microphones alongside an ultrasonic transmitter and an Artix-7 FPGA. The microphones operate as a phased array, allowing the system to use differences in the timing and phase of incoming sound to determine where echoes are coming from. Tornyol says its system can create three-dimensional spatial information and detect extremely small movements. Its current technology page lists a detection range of up to about eight metres.

That matters because an insect's wings are not simply moving objects. They create a repeating pattern in the reflected ultrasonic signal. As the wings beat, tiny changes in the echo produce what engineers call a micro-Doppler signature. Tornyol says it analyses these patterns using digital signal processing and spectrograms, allowing its system to distinguish insects according to their wingbeat characteristics. In principle, the same physical effect can provide information about both the target's movement and its identity.

Why wingbeats can reveal more than location

The Doppler effect is usually associated with a passing siren changing pitch, but the same underlying physics can appear at a much smaller scale. When an insect's wings move toward and away from a sensor during each beat, the reflected ultrasonic wave changes slightly. Those changes repeat rapidly, producing a signature that can be analysed rather than simply treating the insect as a generic moving dot.

Tornyol says its system has been used to identify mosquitoes from their wingbeat signatures and distinguish different targets. That could become crucial if the drone is ever used outdoors. A system that simply attacks every flying object could collide with harmless insects or waste its limited battery chasing the wrong target. Species recognition, therefore, is not a cosmetic feature. It is part of the basic control problem.

The technology also has an unusual advantage over ordinary optical tracking. Ultrasonic sensing does not require visible light. A mosquito moving through a dark garden does not become invisible simply because the sun has gone down. Tornyol says its system can detect mosquito wingbeat patterns in darkness, although the practical performance of the complete drone system under real outdoor conditions remains something that still needs to be demonstrated.

The hardest part comes after detection

Finding a mosquito is only the beginning. Once a target has been located, the drone has to predict where the insect will be a fraction of a second later, accelerate toward that position and make contact without hitting a wall, tree, person or another object. At the same time, the aircraft is carrying a battery, motors, electronics and propulsion hardware inside a package weighing only about 40 grams.

This is a difficult control problem because both sides are moving. The mosquito can change direction unpredictably, while the drone has its own inertia and limited thrust. A successful interception therefore requires more than accurate sensing. The software must continuously combine target position, velocity and predicted trajectory with the drone's own position and flight dynamics. Tornyol says its control algorithms are designed to ram the drone into the target while avoiding nearby walls and obstacles.

The public demonstration shows that this chain can produce an actual insect strike. It does not yet tell us how often the system succeeds. There are no publicly available independent field measurements showing mosquito hit rates, false detections, battery endurance during repeated hunts or performance in wind and rain. Those numbers will matter much more than a single successful interception when judging whether the concept can become a practical pest-control system.

Why a backyard is much harder than a laboratory

A controlled test environment gives engineers something that a garden does not: predictable surroundings. Outdoor spaces contain leaves, branches, walls, pets, people, other insects and constantly changing airflow. Ultrasonic signals can also behave differently around surfaces because sound reflects from objects and creates complicated echo patterns. Separating one tiny moving insect from that acoustic background is a much harder problem than locating a target in a carefully prepared test area.

The drone itself introduces another constraint. Tornyol's current product concept involves a base station that detects targets and coordinates the drone, which can patrol a defined zone, return to recharge and resume operation. The company says its goal is continuous mosquito protection around a home. Its website currently describes a system designed for autonomous patrol and mid-air interception, with U.S. shipping targeted for 2027.

That commercial vision is striking, but it should not be confused with proven city-scale mosquito eradication. Tornyol has said that 10 drones could eventually cover about one square kilometre and that its approach could reduce mosquito-control costs dramatically. Those are company projections, not results from a demonstrated city deployment.

Is killing mosquitoes this way is safe?

Chemical mosquito control already raises questions about effects on other organisms. A drone that physically targets mosquitoes could avoid some of those chemical exposures, especially if its recognition system becomes accurate enough to select specific targets. But that creates a different safety question: what happens when the machine gets the identification wrong?

Mosquitoes are also part of food webs, even though some species are important disease vectors. Eliminating mosquitoes from a small human living area is very different from attempting to remove them across an entire ecosystem. A useful system would therefore need strict target recognition, reliable obstacle avoidance and strong limits on where the drones can operate. The engineering goal is not simply to make a machine that can hit an insect. It is to make one that knows what to hit and what to leave alone.

For now, Tornyol's moth interception is best viewed as a small but meaningful step in that direction. The achievement combines ultrasonic sensing, micro-Doppler analysis, rapid tracking and autonomous flight in a machine smaller than many household objects. The real breakthrough, if it comes, will be much less dramatic than the first moth being struck on camera. It will be repeated mosquito interceptions outdoors, performed safely and reliably without a human or laboratory system quietly doing part of the work. That is the test that will determine whether a tiny drone can move from an impressive prototype to a practical weapon against one of the world's most persistent disease vectors.