Published Oct 13, 2026

The smell of cinnamon buns can transport us right back to our childhood in an instant, but how does that work? Researchers from KTH, the Karolinska Institutet and the Max Planck Institute are developing AI models that can learn how people experience and perceive smells.

Project Grant: Knut and Alice Wallenberg Foundation

Photo: Magnus Bergström, Knut and Alice Wallenberg Foundation

Project Grant 2026: Multimodal Representation Learning of Human Olfactory Perception
Principal Investigator: Professor Danica Kragic Jensfelt, KTH Royal Institute of Technology
Co-Investigators: Professor Johan Lundström, Karolinska Institutet; Professor Jonathan Williams, Max Planck Institute
Funding Awarded: SEK 28 million over a five-year period.

“We want to create smells that make people remember things. How do people describe a smell, and what language do they use? We also want to understand the chemical perspective – how can we recreate a smell without using 100 different types of chemical compounds?” says Danica Kragic Jensfelt, Professor of Robotics at KTH.

The sense of smell gives rise to everything from taste experiences to emotional and social reactions, and it can also help us to perceive danger. Whilst it is one of the most fundamental of the human senses, the perception of smell has been left behind by the rapid developments in artificial intelligence and neuroscience. Researchers are now setting out to bridge this knowledge gap, at the intersection of artificial intelligence, neuroscience and chemistry.

"We are a data scientist, a psychologist and a chemist who will be studying various aspects of smell. A chemist will investigate how molecules and different types of substances can be combined to create smells, and how to do so. AI helps us to understand how to build models using data. The psychologist will investigate how people react when they smell something and how they describe what they smell. Ultimately, when someone says they want something that smells in a particular way, we aim to be able to generate that type of smell based on chemical principles," Kragic Jensfelt says.

In a large number of controlled laboratory experiments, participants will be asked to smell various samples and also to take the smells home with them. Large amounts of data on how each individual perceives the smells will then be used to train AI models.

"Your nose isn’t the same as mine. When it comes to different smells and tastes, we agree less than we do when looking at a picture. We can’t say that there are absolute smells; that’s what makes smell so interesting. Concentration also plays a part; people can tolerate smells to varying degrees and for different lengths of time."

This knowledge could make it possible to develop aids for people who have lost their sense of smell, to promote wellbeing through scents, and to help create olfactory experiences in digital environments – in much the same way that images already enable us to assess a product’s visual characteristics online.

"In the future, scents could be used on a television or computer. Imagine you want to buy a perfume and a scent corresponding to it appears, just like a picture of an item of clothing," Kragic Jensfelt says.

Text: Sabina Fabrizi ( sabina@kth.se )