Category: Technology | Published: 2026-10-29
An online coffee shop can show you the beans, describe the roast and play a video of a fresh brew. The aroma is still something you have to imagine. Researchers are exploring whether connected devices could bring a recognisable scent into that experience, without needing the customer to be in the same room.
Digital smell technology has taken an interesting step in that direction with Scentronic, an experimental system developed by a team including researchers at Ohio State University. Their work connects AI-assisted smell recognition to a device that releases fragrances somewhere else.
It is a promising demonstration, but it is not a way to send any smell through your phone. Understanding that difference makes the research more useful, particularly for businesses wondering where sensory technology might genuinely help.
Imagine placing a coffee sample near an electronic nose. Sensors react to chemicals released by the sample, and machine-learning software classifies the pattern as coffee. That result is sent over an internet connection to a receiving device.
The receiver then releases a coffee fragrance from liquid it already contains. No coffee molecules travel down the network, and the equipment does not build an exact copy of the original aroma from nothing.
Think of it as sending an instruction to select a prepared scent. If the receiving device does not carry a suitable fragrance, it cannot produce that smell. The achievement is linking detection and controlled delivery, rather than removing the physical materials involved.
The electronic nose uses six gas sensors. Their electrical responses change when they encounter different chemicals, giving the software a pattern it can compare with its training data.
At the other end, small electrically driven discs vibrate to turn scented liquid into a fine mist. Separate channels allow individual fragrances to be released or combined, with control over their strength.
The researchers explored wristband, handheld and necklace versions of the receiver. Cloud-based communication passes instructions between the sensing and scent-releasing equipment. These are specialised prototypes, not features that an ordinary laptop or smartphone can provide on its own.
The study, published in Science Advances, involved doctoral researcher Chunyu Yang and colleagues. Its published research paper sets out both the engineering and the experiments behind the demonstration.
The recognition experiments covered beer, cake, coffee, orange and red wine. The team collected 250 samples, with 50 in each category. Its best-performing model correctly classified all 50 samples held back for the final test.
That is an encouraging result within a defined experiment. It does not mean the system recognises every smell with perfect accuracy.
The work used controlled conditions and one product within each category. Recognising that particular coffee sample is different from distinguishing lots of coffee blends, dealing with background cooking smells or identifying a mixture in a busy café.
Adding more categories would require suitable training data as well as the corresponding fragrances at the receiver. Digital scent recognition and the range of smells available for release would both need to expand.
There is another important distinction in the results: the machine's classification score is not the same as a person's experience of the released fragrance.
In a separate perception test, participants correctly identified the emitted scents 72.86 per cent of the time after familiarisation. For combinations of two scents, identification accuracy was 71.43 per cent.
These findings suggest that the device can deliver recognisable cues. They do not establish that each fragrance is an indistinguishable substitute for the source, or that everyone will interpret it in the same way.
For an online retailer, that matters. A pleasant coffee-like aroma could make a demonstration memorable without accurately representing the particular beans being sold. An engaging experience and a dependable product sample are different achievements.
One possible application is assessing how someone's sense of smell changes over time. A future service might remotely choose a fragrance and its strength, then record whether the person recognises it or notices a difference.
That could be useful for research and, if properly validated, some forms of remote assessment. It could also reduce the need for repeated journeys to a clinic in situations where remote testing is appropriate.
However, the study's perception experiments involved healthy participants. They did not establish that Scentronic can diagnose a disease or replace an existing clinical assessment.
The researchers also acknowledged limitations in their experimental controls. Larger, more rigorous studies would be needed to establish dependable results across users, environments and clinical circumstances. This is an area for further investigation, not a medical service businesses should assume is ready to deploy.
The team also developed an experimental cup combining scent release with small amounts of flavoured liquid. In a mixed-reality demonstration, selecting a virtual drink triggered corresponding physical scent and liquid delivery.
That shows how an on-screen action could be connected to more than pictures and sound. Possible future uses include entertainment, education and sensory research.
Online shopping is another suggestion, but the practical requirements are substantial. Customers would need compatible hardware, the correct fragrance supplies and an experience that represents the product fairly. The research does not demonstrate a universal smell attachment for every online purchase.
A focused application with a few clearly defined aromas may be more realistic than trying to reproduce an entire supermarket or restaurant menu.
Before considering any future trial, start with the task rather than the novelty. What would scent help a customer or employee understand that an image, description or video cannot?
Useful questions would include:
- Can users reliably recognise the intended scents without extensive coaching?
- Does the delivered fragrance represent the real product closely enough for the purpose?
- Does adding scent improve understanding or decisions, rather than simply attracting attention?
- How often do the devices need refilling and cleaning?
- Can scent intensity be controlled, and can users easily stop or decline the experience?
- What evidence supports safe repeated exposure and suitability for the intended users?
- How much ongoing work is involved in maintaining the hardware and supplies?
Those are evaluation questions, not claims that the prototype has already answered them. A convincing laboratory demonstration still needs substantial work before it becomes a dependable everyday service.
This project is a useful example of AI solving a narrow recognition problem within a larger physical system. The software helps classify sensor readings, but the overall experience also depends on fragrance chemistry, device design, connectivity and human perception.
That is relevant even if your business has no plans to use scent. When assessing an AI proposal, look beyond an impressive accuracy figure. Ask what was tested, under which conditions, and whether the result measures the outcome that actually matters to users.
Our AI Services page explains how Cloud Smart Solutions helps businesses explore practical AI tools and automation. The aim should be a useful outcome supported by evidence, not adopting experimental hardware simply because it is new.
Scentronic remains a research prototype. Its contribution is showing that selected smells can be recognised and used to trigger controlled scent delivery remotely. Whether digital smell technology becomes useful in shops, healthcare or entertainment will depend on how well that controlled demonstration translates into reliable, manageable experiences outside the laboratory.
Photograph: coffee aroma assessment, illustrating one of the scent categories tested. It does not show the Scentronic prototype.