The scent of flowers arrives on the wind. Even when you know that something nearby is fragrant, finding its exact source is not easy. If an odor sensor is mounted on a drone, can it fly in the direction the scent came from?
In February 2025, a team from institutions including Chiba University and Shinshu University reported a small drone that tracks odors using a silkworm moth antenna. The innovations lay in equipment that makes use of the antenna’s sensitivity and in stopping rotation while surveying the surroundings.
An Odor in the Air Is Not a Single Path
Odor molecules are carried by airflow. Depending on wind speed and turbulence, their dispersion forms narrow filaments, with concentrations changing across space and time. It is not possible simply to draw a map in which the concentration always increases smoothly toward the source.
A strong response at a particular moment therefore makes it difficult to decide where to go on that basis alone. A drone also moves the air itself with its propellers. The researchers needed to consider where to position the odor-sensing apparatus and which way to point it.
Reading Electrical Signals from a Real Antenna
This study used a real antenna detached from a male silkworm moth. A single antenna was connected to electrodes to read the small electrical signals produced in response to an odor. The signals were amplified and sent to a computer, where they were combined with the drone’s orientation to determine its direction of travel.
The target was bombykol, the main component of the silkworm moth’s sex pheromone. It is a cue that helps males find a mate, and the antenna responds sensitively to it. This was not a study that created a machine able to distinguish flower species or perfume notes as a person does.
Because the antenna is biological tissue, signal strength varies from one antenna to another, and drying also needs attention. The team added grooves to the electrodes to help retain conductive gel and incorporated a circuit with adjustable signal amplification. These measures made weak responses easier to read while avoiding signals exceeding the measurement range when responses were strong.
Narrowing Down the Direction with the Shape of an Enclosure
An enclosure with an outward-flaring inlet was fitted around the antenna. Because placing it too close to the airframe would expose it to the propellers’ airflow, the antenna was positioned 10 cm from the drone’s center. The researchers measured air velocity and used smoke to check how air entered the enclosure.
When responses were compared as the airborne drone completed one rotation, an enclosure without a flared inlet produced responses concentrated roughly within 90 degrees on either side of the odor source’s direction. With the flared inlet, the range narrowed to roughly 45 degrees on either side. Greater responsiveness to odors from directly ahead, rather than equal responsiveness from every direction, provides a clue for finding the source’s direction as the drone turns.
Pausing Occasionally Instead of Turning Continuously
The next change concerned movement. The new method divided a complete rotation into three steps, inserting 4-second pauses along the way. A pause here does not mean landing or stopping the propellers. It means allowing time for the drone to hold its orientation in the air.
The electrical signals recorded during pauses and rotation were matched to the drone’s orientation at those moments. The drone estimated a direction of travel, taking response strength into account, and moved 0.7 m in that direction. It repeated this cycle of surveying the surroundings and moving forward.
In comparisons starting 3 m from the odor source, the previous method combining rotation and forward movement was judged successful in 3 of 17 trials, and the new method in 7 of 17. This was an increase from approximately 18% to approximately 41%. Success was defined as entering a radius of 1.3 m or less from the odor source. These figures do not represent the proportion of trials that pinpointed the exact source.
With the new method, estimated directions also tended to align more closely with the actual odor source. The authors suggest that pausing may make it easier for air to reach the antenna and reduce the influence of encountering an odor filament only momentarily. These explanations of the mechanism need to be distinguished from the observed increase in successful trials.
What the Search from 5 m Away Showed
The improved apparatus was also tested in searches starting 5 m away. These were indoor experiments with a wired power supply to avoid the flight-time limit imposed by batteries. In 2 of 13 trials, the drone approached to approximately 45 cm short of the odor source; in another 2, it reached approximately 1 m away. The result does not mean that it reliably found the source every time from 5 m away.
The paper also includes an experiment detecting the odor at a distance of 10 m. However, that experiment measured the response while the drone hovered in a corridor. Detecting an odor from 10 m away and flying from there to its source were investigated as separate capabilities.
All of these results were obtained indoors, with bombykol released from a specified apparatus.
Summary
Even a drone carrying an odor-sensitive antenna needs a way to decide which direction to take. In this study, refining the air-intake enclosure and adding periods of holding its orientation during rotation made it easier to investigate the direction of an odor and increased the number of comparative trials in which the drone approached the source.
As air moves, an odor arrives and then disappears. To use those changes as navigational cues, it matters not only what can be sensed but also how air is received and when measurements are taken. Pausing occasionally was a way of connecting odor-detection technology with odor-tracking technology.
Original Study
Chihiro Fukui, Tomoya Uchida, Sakito Koizumi, Yuta Murayama, Hao Liu, Toshiyuki Nakata, Daigo Terutsuki
Advanced bio-hybrid drone for superior odor-source localization: high-precision and extended-range detection capabilities
npj Robotics, 3, Article 4 (2025). First publication and version-of-record publication: February 5, 2025.
DOI: https://doi.org/10.1038/s44182-025-00020-9
Full text: https://www.nature.com/articles/s44182-025-00020-9
Original study license: Creative Commons Attribution 4.0 International (CC BY 4.0). The original Japanese article is an independent explanation based on the study’s findings. No figures or tables from the original study have been reproduced.
