Scientists Build ‘Scentronic’ System That Can Detect Smells and Transmit Them Over the Internet

Researchers at The Ohio State University and Zhejiang Laboratory in China have built a system that can detect a smell in one place and recreate it at another, using the internet as the bridge between the two. The technology, described in a study published on 22 October in the journal Science Advances, is called Scentronic — and its creators say it is the first compact, fully automated platform that brings odour sensing and odour generation together for remote sharing.

The system has two halves. The first is an electronic nose fitted with six gas sensors that detect the chemical compounds escaping from a scent. A machine-learning model then classifies the odour from the pattern of signals the sensors produce. The second half is the receiver, a wearable device — built in wristband, necklace and handheld forms — that carries preloaded scented liquids and releases them through miniature vibrating atomisers when the digital data arrives over an online platform. The scent molecules themselves are never transmitted; only the information about the smell travels, according to Reuters.

“In simple terms, it connects ‘smelling something here’ with ‘recreating the corresponding smell somewhere else,’” said Chunyu Yang, an Ohio State doctoral student in materials science and engineering and the study’s lead author. “The idea is to add smell to the visual and audio information we already share remotely, making those experiences more engaging,” she added.

The prototype currently recognises and recreates just five scents: coffee, beer, red wine, cake and oranges. The team chose them, Yang told Reuters, because “they are familiar everyday odours with different scent profiles, making them useful examples for demonstrating the system.”

When 14 volunteers were asked to identify the smells released by the receiver, they got individual scents right about 73 per cent of the time, according to the reporting on the study in TechXplore. When smells were sent over in pairs — coffee and orange, beer and cake — identification held at roughly 71 per cent. The system responded to signals from the smell sensor in as little as 0.6 seconds, although classifying more complex food odours took several to tens of seconds.

What makes Scentronic different, Yang said, is that it brings odour sensing and generation together for remote sharing. “To the best of our knowledge, our study is the first to explore this integration in a compact platform for remote odour sharing,” she told Reuters.

Previous attempts at digital scent have relied on bulky desktop boxes, static scratch-and-sniff cards or virtual-reality headsets, sometimes requiring a person to manually trigger a scent. The new approach is hands-free: the AI e-nose detects and analyses the odour, transmits the data across the web, and the Scentronic receiver recreates it. The e-nose itself is roughly 15 centimetres in diameter — portable, though not designed to be worn, according to Reuters.

The team also built a prototype digital cup that dispenses both scents and flavoured liquids. Wearing a mixed-reality headset, a user could select a virtual drink and trigger the release of the corresponding flavour and smell in the real world, TechXplore reported.

The researchers have been careful about what they claim. “A broader scent library is possible in principle, but an unlimited repertoire is far beyond what we have demonstrated,” Yang told Reuters. “Recognising an odour and accurately reproducing it are separate challenges. Covering a useful range of everyday scents is a more realistic goal, although that would still require substantial development.”

The receiver cannot recreate every smell the system encounters. “To support additional odours, we would need to train the electronic nose to recognise them and provide corresponding scented liquids in the odour generator,” Yang said. And in the paper itself, the authors note that the current implementation classifies odour at the category level and reconstructs it through preloaded channels “rather than reproduced with full chemical compositional precision.” They also caution that the experiments were relatively simple with limited control conditions, which “may introduce bias” — meaning further testing is needed to see how the system performs outside the laboratory.

Commercialisation, Yang stressed, is not imminent. “At present, this is a research prototype,” she said, adding that moving toward a product would require a larger, more diverse training dataset and reliable recognition in real-world conditions, “where smells often occur in mixtures.” The generator would need to support more scents while staying compact, and the team flags questions around scent fidelity, refillability, durability, manufacturing costs and the safety of repeated exposure to the scented mist. No company was involved in the research or the device development.

Even so, Yang sketched the horizon in plain terms: “For example, someone could share the scent of a meal with family or friends who live far away. Other potential applications include more immersive virtual and augmented reality experiences and online shopping that allows customers to sample selected scents remotely.”

Asked whether the technology could one day live in smartphones — imagine a football fan at an Ohio State tailgate sharing the aroma of grilling hot dogs with a friend over a video call — Yang said it could in principle connect to a scent-generating accessory, though integrating everything into the phone itself would be harder because “the scent reservoirs, electronics and atomisers all take up space and add weight.” Miniaturisation and convenient refilling would be the key challenges. “The tailgate example illustrates the idea well,” she added — while cautioning that the current system has not demonstrated the transmission of a grilling-hot-dog scent or the exact mixture of smells at a tailgate party.

Analysis: Why It Matters

Smell was always the hard sense

Every other sense got digitised decades ago, and it is worth asking why smell took so long. The answer is chemistry. Vision reduces to three colour channels: combine red, green and blue at different intensities and a screen can reproduce almost any colour a human eye can perceive. Sound is pressure waves sampled thousands of times a second, reassembled faithfully by a speaker. Smell has no equivalent primitives. An odour is a soup of volatile organic compounds — ethanol from a beer, limonene from an orange peel — drifting through the air in mixtures that shift by the moment. There is no pixel for scent, no frequency to sample. Any machine that wants to transmit a smell must first recognise a complex chemical mixture and then rebuild an approximation of it on demand, which is fundamentally a harder engineering problem than rendering pixels.

That is why the graveyard of digital scent is so crowded. Smell-O-Vision piped odours through theatre seats in 1960 and never survived its second screening. In 1999, DigiScents promised the iSmell, a USB device that would mix 128 base aromas, and raised roughly $20 million before collapsing in 2001. Japanese and British prototypes, scent-enabled game controllers and VR aroma rigs have come and gone. Each effort failed at some combination of size, cost, reliability and — most fatally — the need for a human to push a button to release a pre-baked smell. Scentronic’s contribution is to remove the human from the loop: the machine smells, the machine decides, the machine transmits, and a wearable machine on the other end releases. That end-to-end automation in a compact form is the step that the 2001 generation could not take, because they lacked both the sensor sophistication and the machine learning to do classification at the source.

Seventy-three per cent is better than it sounds

A headline figure of 73 per cent correct identification might read as underwhelming — until you compare it with what human noses do. Ask a blindfolded person to name mystery smells from a set of five options without context and they will often score worse, particularly with close relatives like beer and red wine. The volunteers in the Scentronic test were not smell experts; they were ordinary people asked to match a recreated mist to one of five everyday categories, and they did it nearly three times out of four. Pairs arriving together — two scent mists at once — scored only slightly lower at 71 per cent, which suggests the system degrades gracefully rather than collapsing when mixtures enter the picture.

The classification times matter too. A 0.6-second response from the sensor means the system can operate near real time for simple scents, which is the territory where communication feels conversational rather than postal. The slowdown to “several to tens of seconds” for complex food odours points directly at the known frontier: real-world smells arrive as mixtures, and untangling them is the job of a more powerful classifier and a richer training dataset. The paper’s own honesty about category-level classification — it identifies “coffee” as a class, not the precise chemical fingerprint of this particular cup — is exactly the right frame. Nobody needs the system to reconstruct an espresso’s full molecular profile; the goal is a recognisable approximation that carries meaning. Category-level is where the usefulness begins.

The wearable is the real innovation

Strip away the AI branding and the most consequential design decision is form factor. Earlier scent systems were desk-bound: you sat near the box, the box puffed. Scentronic’s receiver is a wristband, a necklace, a handheld — objects people already wear or carry. That matters because smell is a proximity sense. It works within inches of the nose; a desktop box broadcasting into a room dilutes the scent, wastes the liquid and annoys bystanders. A wristband that releases a fine mist near the wearer’s face on command is the first architecture that respects how olfaction actually works in physical space.

The miniature vibrating atomisers are doing the heavy lifting here — pushing scented liquid through a cotton filter to create a fine mist on demand, powered over Wi-Fi and a cloud server that stores and transfers the scent data. None of that is exotic technology in isolation; what is new is the integration, miniaturised enough to wear. The prototype digital cup, paired with a mixed-reality headset to dispense flavour and aroma for a virtual drink, hints at where this integration is pointed: not at making phones smell, but at closing the loop between what you see in a headset and what your other senses experience. The researchers’ choice of the word “interface” in the paper’s title — an olfactory interface, not an olfactory transmitter — signals the same ambition. Interfaces sit between humans and computers. That is the lane they are building for.

The honest limitations are the point

It would be easy to file this story under “the future is here,” but the researchers themselves refuse that framing, and the limits they name are instructive. The receiver can only emit scents it has preloaded; each new odour requires both training data for the e-nose and a physical reservoir of the right liquid at the receiving end. An unlimited scent library is therefore not a software problem — it is a logistics problem, a miniature plumbing problem, and eventually a safety problem, since atomising fragrances for repeated inhalation needs toxicological scrutiny.

There is also the deeper philosophical snag the team flags: recognition and reproduction are separate challenges. A machine can learn to label a sensor pattern “coffee” without being able to make you perceive coffee convincingly. Perception is not chemistry alone; it is expectation, memory and context. The researchers’ stated goal — a “useful range of everyday scents” rather than chemical precision — is a quiet admission that the engineering target is human experience, not molecular fidelity. That is a harder, fuzzier target, but it is also the honest one. Every commercialisation hurdle they list — refillability, durability, manufacturing cost, repeated-exposure safety, real-world reliability with mixed smells — reads as a checklist for turning a lab demonstration into a consumer device. No company is attached yet, which means no hype engine is attached either. That absence of a product timeline is itself a credibility signal.

The commerce angle is coming regardless

Despite the researchers’ caution, the commercial logic is plain enough that the market will not wait for the science to finish. Online retail’s oldest unsolved problem is that you cannot smell what you are buying: perfume counters migrated online during the pandemic and never came back, but the scent category remains stubbornly dependent on physical sampling. If even a five-scent vocabulary can reach 73 per cent recognition, the industry will start asking what a 50-scent vocabulary with a subscription refill model is worth. The fragrance industry, which has flirted with digital sampling for decades, is the obvious first customer.

Beyond retail, there is a parallel track the researchers barely mention but the hardware suggests: telepresence and care. Video calls kept people visually connected through the pandemic years; adding the smell of a family meal is not a gimmick but a genuine sensory bridge, the kind of thing that matters to people separated by distance. The healthcare angles are also real — anosmia became a household word after COVID-19, and controlled scent delivery has applications in smell training, sensory therapy and conditions involving hyperosmia. The research comes at a moment when the sensory vocabulary of the internet is being rebuilt from scratch by the AI era. Text, image, audio, video, touch — each arrived in sequence. Smell’s long absence may simply have been a waiting game for sensors and classifiers to catch up. With Scentronic, the wait is over, at least at prototype scale. The question now is not whether smell will join the digital senses, but who builds the refill cartridges.

What to Watch Next

1. Scent library expansion. The critical test is mixtures: moving from five isolated everyday scents to a wider vocabulary that can survive real rooms, where coffee steam, rain on asphalt and someone’s aftershave arrive together. Watch for a follow-up study with layered and overlapping odours.
2. Smell training and medicine. Controlled, repeatable scent delivery is exactly what olfactory rehabilitation and sensory therapy need. Expect clinical researchers — particularly in anosmia and migraine-related hyperosmia — to start testing or licensing the hardware.
3. The smartphone accessory. Yang’s “in principle” accessory comment is a door left ajar. If a third-party manufacturer ships a clip-on scent module that pairs with video calls, the technology crosses from lab to living room — with refill subscriptions following fast.
4. VR and AR headsets. The prototype digital cup with mixed-reality integration is the clearest commercial signal in the paper. The next generation of headsets, already chasing presence through haptics, will be the battleground where a sense of smell either matters or dies as a feature.
5. Competing labs. A first in this field rarely stays a first for long. Watch for parallel efforts from South Korean and Japanese research groups, who have led much of the prior digital-scent work, and for whether any of the big platform companies — the ones already investing in haptic and neural interfaces — take a position on smell.

Sources

• Reuters — “New ‘Scentronic’ technology can detect and transmit aromas remotely”
• TechXplore — “Something smells different as scientists send aromas over the internet”
• The Times — “How sending smells over the internet could let you sniff before you buy”
• Science Advances — Chunyu Yang et al., “Scentronic: An artificial intelligence–assisted olfactory interface for remote odor sharing and transmission” (2026)