In this issue

What happens when a robot can smell a room, and remember it?

🤖 The pitch: Ainos published a patent for a self-clearing electronic nose that resets between rooms in seconds, turning a stationary lab sensor into something you can put on a moving robot.

⚠️ What's new?: Two air pipes feeding one chamber, one filtered and one raw, so the sensor can zero itself on the move without docking. That is the piece lab e-noses have been missing for thirty years.

💰 Follow the money: Osmo raised US$70 million for fragrance AI. Ainos, which holds this patent, reported US$313 in first-half 2026 revenue and US$1.4 million in cash against US$16.3 million in current liabilities.

📄 The paperwork: A robot that smells a room is also smelling the people in it. Smell is involuntary, and 80% accuracy is promising for a demo but thin for a gas alarm.

The dog at the airport

You are third in line with your bag at your feet. A beagle comes down the queue at a trot, nose a few inches off the floor, handler on the other end of the lead. It touches your bag for half a second, moves on, and stops dead at the one behind you.

Nothing else in that terminal can do what the dog just did. It read the air around a closed suitcase and found the one molecule responsible faster than any scanner in the building. It also took months to train, working in short shifts because a tired nose starts missing the faint things, and cannot be copied. There is exactly one of them.

What the dog is doing is chemistry. Molecules lift off the bag, land on receptors in its nose, and fire a pattern it has been taught to recognise. Engineers have been trying to build that in silicon for thirty years, and it keeps failing for an unglamorous reason. The machine version cannot clear its nose between bags.

Ainos, a company in San Diego, recently published a patent covering the hardware that fixes that.

We gave machines eyes and ears about a decade ago. Smell is the one sense we still outsource to a dog.

Ainos, Inc. is based in San Diego and sells a smell module to robot makers.

Before the patent, we will lay out the basics, because a machine being able to smell is simpler than it sounds.

A smell is molecules drifting in air. A machine catches them using a coated chip that changes electrical resistance when something lands on its surface. One chip on its own is not particularly informative, so a sensor is a small array of them with different coatings, each reacting its own way. Coffee pushes the array one way. Petrol pushes it another. The reading is a pattern across a handful of numbers.

The machine has no idea it is coffee. Software compares that pattern against a library of patterns people have already labelled and returns the closest match. Ainos calls its version a Smell ID. When nothing matches, the robot has smelled something and cannot tell you what.

All of that works while the sensor sits still, which is why lab e-noses have been fine for thirty years. Put one on a moving robot and it has really struggled. The coatings hold onto molecules after the source is gone, the way your own nose gives up on a kitchen after ten minutes. For example, where a robot has been in a room full of bleach, even when it leaves the room the sensor will still falsely register that bleach "smell".

The patent this week resolves that. Two pipes feed one chamber. The first runs air through a carbon filter, the sort in a cooker hood, so the machine breathes blanks until its readings flatten out and hold. That flat line is the sensor back at zero. Then it switches to the second pipe and raw, unfiltered air. The pump keeps running through the swap, so pressure and flow stay put and the only thing that changes is what is in the air. Cleaning drops from minutes to seconds, and the robot can do it in a corridor instead of at a dock. A nose that resets between rooms is a nose you can put on rounds.

Anything that smells for a living today either stands still or is a dog. A gas detector is a box on a wall reading the air in front of it, so a leak at the far end of a plant waits until it spreads far enough to arrive. Dogs cover the ground the boxes miss, at months of training and one handler per queue.

Put the box on something that moves and the readings stop being trustworthy. The sensing surface holds onto molecules for minutes after the source is gone, so the machine reaches the next room still reporting the last one, with nothing on the screen to say so. It calls bleach in a room that has none. Worse, it can arrive already loaded from the corridor and barely react to a real leak in front of it.

An alarm people learn to ignore gets switched off. That is the problem this patent takes on. Knowing what the smell actually is stays open.

Machines learned to see because someone built the reference first. Millions of labelled photographs, then models trained on them. Speech went the same way.

Smell has no equivalent. There is no agreed catalogue of what a sensor pattern means, so every deployment builds its own. Ainos ran its module in a Japanese semiconductor plant and reported recognising 22 compounds at close to 80% accuracy. Twenty-two, in one building, after months on site.

And still, that nose recognises little in a different environment. The labelling starts again, by hand, in every new setting.

Building the nose is engineering and it is nearly finished. Teaching it what it is smelling still takes fieldwork, one site at a time.

Digital smell splits into people who make scents and people who read them, and they rarely meet.

The best funded is on the making side. Osmo spun out of Google Brain in 2022 and works out how a molecule will smell from its structure, which lets it design fragrance far faster than a perfumer can. Health and environmental sensing are on its list for later. There is no Osmo sensor on a robot.

On the reading side, Aryballe in Grenoble builds the physical detector using biochemistry and optics, and its backers say a lot about who expects to need one, such as IFF in fragrance, Hyundai in cars, Samsung Ventures and Groupe SEB in appliances. It is the closest technical match to this patent, though its sensors are built for lab benches and production lines rather than something on wheels. Aroma Bit in Tokyo, backed by Sony, makes compact sensors on the same principle. Alpha MOS, Sensigent and Airsense have sold laboratory e-noses for years, which are accurate and plugged into a wall.

Koniku took the strangest route, growing living neurons on silicon and fitting them with odour receptors. It signed Airbus in 2017 and announced airport trials for late 2020 at Changi and San Francisco. Six years on, no terminal is screening passengers with them and the company has drifted into defence work.

The robot slot Ainos is standing in has been empty for a while, and that is the part worth chewing on. Seeing this smelling capability has generally been around for a while, Ainos is patenting around something their competitors didn't bother with, which raises the important question of whether nobody had yet found a customer who will pay extra for an alarm that walks.

Straits Research valued electronic noses at US$156.9 million in 2024. Maximize valued the same year at US$30.78 billion, considering a broader category of things that sense gas. Ainos quotes the large number.

Meanwhile, plain gas sensors were worth about US$1.78 billion in 2025, heading past US$3 billion in the early 2030s. Safety rules are a significant driver in these categories.

Robots are the layer Ainos is aiming at. It fitted a nose to a machine built by Japanese robotics firm ugo in April 2025 and called it the first robot with a working sense of smell. It also signed ASE, one of the largest chip packaging firms in the world, to try the technology across its plants. Ainos builds no robots, so its position depends on being the nose inside other people's machines and charging for the readings rather than the parts.

Which runs into the buyer. The safety manager with the budget already owns fixed gas detectors that work and never move. Selling him the same alarm again on wheels rests on the robot catching things the wall-mounted units miss, and no one has published that comparison yet.

The reading a robot takes of a room could be a reading of the people in it.

Smell is involuntary. You can step out of a camera's view or stop talking near a microphone. You cannot decide not to smell of the whisky, the cigarette, or the infection you have not been to a doctor about. The whole point of this patent is to make that reading hold up room after room, which is what turns a lab result into a log.

Then there is the question of who is liable for a wrong answer. The call gets made on the robot, by software matching a pattern to a database. What if it passes a room that has a gas leak, simply because that particular hazard is not listed in its database or the robot simply misses it? If a facility trusted the green light, the fault sits somewhere between the sensor licensor, the robot maker and the operator. In a market where safety rules are such a strong driver, how will those rules need to adapt to this novel, moving sensor?

The plainer risk is that the thing is not good enough yet to deserve any of this trust. Sensor coatings drift as they age, humidity could potentially move the readings, and similar molecules can look alike, which is exactly why the patent spends its entire length on holding a steady zero. Ainos's own reported accuracy is around 80%. Good for a demo. Poor for a gas alarm.

The money in digital smell is going into making scents rather than reading them.

Osmo raised a US$70 million Series B in February 2026, led by Two Sigma Ventures, bringing it to US$130 million after a US$60 million Series A from Lux Capital and Google Ventures in 2023. Fragrance design pays for it. Sensing could come later, if it comes.

The detection side raises smaller and stranger cheques. Aryballe has taken about US$19 million across four rounds, mostly from corporates like IFF, Hyundai and Samsung Ventures who want a look at the technology. Aroma Bit's early money came from Sony's innovation fund on the same logic.

Then the company that filed this patent. Ainos grew AI Nose revenue 499% in 2025 at 82.9% gross margin, which reads well until you find the base. Revenue for the first half of 2026 was US$313. At 30 June it held US$1.4 million in cash against US$16.3 million of current liabilities, including US$11.0 million of convertible notes now due, on an accumulated deficit of US$74.6 million. Part of this year's cash came from a US$2.8 million loan from ASE Test, the partner it is running semiconductor trials with. It held its Nasdaq listing in July 2025 only after climbing back above the minimum share price it had breached the year before.

Gas detection itself is a decent, consolidated business lead by Honeywell, Dräger and MSA. None of them has bought a digital olfaction startup. When incumbents with cash sit out a category for a decade, that tells you something about how close they think the revenue is.

For you it arrives as a light in the hallway, or as a line in a log that is considered irrelevant to flag to you fully. Both come from the same sample of air.

Watch the balance sheet more closely than the technology here. The patent describes a nose that could monitor a whole building. The company holding it has US$1.4 million in the bank. The family is already moving on, though, with a related patent granted in Japan that swaps the filter for a UV lamp and a catalyst that cleans the chamber with no cartridge to replace, and adds an idle mode that only samples when something in the air changes. These are different versions addressing the same mobility and accuracy problems, seeming to achieve more longevity (i.e. moving towards systems you can safely operate unsupervised overnight).

This week's patent is US 2026/0110671 A1, titled "Electronic nose with gas exchange system", published by Ainos, Inc.

Read the filing if you want to see how a robot forgets a room on purpose. Reply and tell us whether you would let one smell your house. We are on Instagram and LinkedIn.

•  Where the olfaction money went with Trellis: Explore why the best funded company in digital smell is designing perfume rather than building sensors.

•  The balance sheet behind the vision with the SEC: Read Ainos's own results and weigh the robotics story against the numbers underneath it.

•  How digital smell burned US$20 million once before with The Hustle: Learn what happened to the last company that promised to digitise scent, and why the money left.

•  The living-cell approach to smelling danger with CNN: Discover the rival that put engineered neurons in an airport, and why the trials never became terminals.