The smell of coffee could become part of a virtual catch-up, thanks to an experimental system that connects odor sensors to wearable scent dispensers over the internet.
Researchers at Ohio State University and Zhejiang Laboratory in China developed the automated system to identify smells using artificial intelligence and trigger their release near another person’s nose.
Described in Science Advances, the technology combines chemical sensing, machine learning, and wearable devices shaped as bracelets or necklaces.
In laboratory tests involving 14 healthy volunteers, participants identified individual odors with 73 percent accuracy and combinations with roughly 71 percent accuracy. The findings offer a step toward adding smell to digital experiences, although the system currently relies on preloaded scents.
Turning chemical signals into digital instructions
Recreating smell presents a different challenge from reproducing images or sound. Cameras and microphones capture light and sound waves, while odors come from chemical compounds that must be detected and physically released for someone to experience them.
Previous approaches have included bulky desktop equipment, scratch-and-sniff cards, and virtual reality headsets. Some also required a person at the receiving end to release a scent manually.
The new system uses machine-learning algorithms to recognize patterns associated with volatile organic compounds, the molecules responsible for smells, and classify them into odor categories.
That information travels over the internet to a receiving device called Scentronic, which releases the corresponding scent close to the user’s nose. The process sends digital instructions between devices; the odor itself comes from supplies already loaded into the wearable.
Testing coffee, cake and mixed scents
The researchers tested five odors: beer, cake, coffee, orange, and red wine. Volunteers encountered them individually and in combinations, including beer with cake and coffee with orange.
Participants correctly identified individual smells 73 percent of the time. Recognition remained close to that level for combinations, reaching approximately 71 percent.
Speed was another focus. The system responded to a scent signal in 0.6 seconds, although identifying more complex food odors required several seconds.
The distinction matters because responding to a signal and accurately classifying an unfamiliar chemical mixture are separate tasks.
The team also developed a version that dispenses scent-infused flavored liquids. This extends the concept beyond smelling an object remotely toward recreating aspects of a drinking experience.
In a potential virtual reality application, a user could select a digital drink and receive a corresponding flavored liquid at home.
Moving beyond controlled laboratory tests
The technology cannot yet reconstruct arbitrary smells from their basic chemical components. Instead, it selects from preloaded odor channels, limiting the range and precision of what it can reproduce.
The researchers plan to test the system outside controlled laboratory conditions, introduce more complex evaluations, and improve controls for potential biases among participants.
Possible applications include virtual social gatherings, online shopping, and healthcare research involving sensory perception or loss.
For now, the work demonstrates a limited but automated connection between detecting an odor and recreating a corresponding scent elsewhere. Expanding that connection to the variety of smells people encounter daily remains the next challenge.