Scent Creation with Artificial Intelligence for Olfactory Displays

Olfactory displays can add smell to virtual and mixed reality, but most systems have a fixed odor library. They can only present scents that were selected, prepared, and loaded before the experience starts. This limits their use in open or interactive environments, where the required odor may depend on user input, scene content, or events during the experience.

This talk presents AI based aroma generation as a way to reduce this limitation. The goal is to generate scent recipes from odor descriptions, instead of choosing only from predefined aromas. The talk will explain a pipeline that starts from a verbal or semantic description of an odor, converts it into odor descriptors, generates a target aroma profile, and reconstructs this profile as a mixture of real scent materials. It will discuss generative models for aroma profiles and nonnegative reconstruction methods for converting computational targets into usable recipes.

A key system issue is the connection between the generated recipe and the delivery hardware. Automatic generation can expand the range of scents that a display can produce, but it does not remove the physical constraints of scent delivery. Channel number, latency, volatility, residual odor, concentration control, calibration, and sensory validation remain important engineering problems.

The talk will use work from the Nakamoto Lab as a case study. It will discuss what current systems can do, what remains difficult, and how AI based scent generation could support more flexible olfactory experiences in future immersive environments.