An electronic nose modeled on insect antennae simultaneously identifies gas mixtures and pinpoints their three-dimensional origin by decoding the natural dynamics of odor plumes.
(Nanowerk Spotlight) When a honeybee navigates toward a flower, it does not merely detect the presence of a scent. Its two antennae, separated by a small but critical distance, register subtle differences in when odor molecules arrive, how quickly concentrations rise, and how strong the signal becomes on each side. From these disparities, the insect’s nervous system extracts both the chemical identity of the odor and the spatial direction of its source, all from the same stream of sensory data. Artificial systems have struggled to replicate this unified capability.
Electronic noses, sensor arrays designed to detect and classify volatile chemicals, can now identify gases at parts-per-million concentrations and distinguish between chemically similar compounds. Yet these devices typically treat the turbulent, time-varying nature of real odor plumes as unwanted noise, averaging out the very dynamics that carry spatial information. Separate technologies handle source localization, relying on wind sensors, gradient-following algorithms, or other hardware disconnected from the chemical sensing itself. No artificial system has combined these functions into a single sensing architecture.
A study published in Advanced Materials (“Receptor-Mimetic Stereo Olfaction for Simultaneous Odor Recognition and Spatial Localization”) presents a platform that does exactly that. The research team developed AROMA (Artificial Receptor-Olfaction Mimetic Array), a sensing system that captures the spatiotemporal dynamics of odor plumes and decodes both mixture composition and three-dimensional source location from a single set of measurements.
Conceptual schematic of the AROMA inspired by insect olfaction. Bees leverage their stereo olfaction to simultaneously detect both chemical and physical properties of odors. Chemical: Distinguishing floral scents from different flowers based on their distinct molecular types and compositional variations. Physical: Antennae exhibit rapid responses to fast-diffusing molecules and delayed responses to slow-diffusing ones, with molecules approaching from the left triggering the left antenna first. The system integrates mixed-ligand AuNPs arrays, with varied ligands for enhanced odor specificity, to emulate diverse olfactory receptors detecting chemical properties of odors, and distributed sensor arrays, deployed in a 3D layout or on a robotic platform, to capture physical diffusion cues analogous to spatially separated antennae. Time-series signals from these arrays are processed by a neural network for real-time analysis, delivering simultaneous mixed odor identification and precise 3D source localization. (Image: Reproduced with permission from Wiley-VCH Verlag) (click on image to enlarge)
AROMA’s sensing elements are gold nanoparticles smaller than 10 nm, each coated with mixtures of organic molecules called ligands. The team combined 12 different ligands in varying ratios through a ligand-exchange process, producing roughly 100 nanoparticle variants with distinct surface chemistries. These mixed-ligand shells undergo phase separation into nanoscale domains, creating heterogeneous surfaces where different odor molecules bind through varying combinations of hydrogen bonding, van der Waals forces, and hydrophobic partitioning.
Each nanoparticle type sits on a tiny interdigitated electrode chip measuring 8 mm by 5 mm. When odor molecules adsorb into the nanoparticle film, the electrical resistance shifts. The rate, magnitude, and temporal shape of that shift depend on both the molecule’s chemistry and its physical diffusion behavior.
Testing across 11 volatile compounds confirmed that each produces a unique fingerprint. Ethyl butyrate, diffusing at 0.671 × 10⁻⁵ m²/s, fills a test chamber in under 100 seconds. Methyl salicylate, at 0.0601 × 10⁻⁵ m²/s, never reaches its plateau within the same window.
Spatial sensing emerges from how the sensors are arranged. Three 10-channel arrays sit on perpendicular walls of an 18 cm cubic chamber, mimicking spatially separated insect antennae. An odor source at a specific position delivers a different concentration history to each array.
Closer arrays register earlier onset, steeper rises, and higher peaks. More distant arrays show delayed, attenuated signals. These disparities directly encode the geometric relationship between source and sensors.
A multi-task Transformer neural network extracts both chemical and spatial information from the 30-channel time-series recordings. The model processes 354 time steps across all channels through four encoder layers with six-head self-attention. Two output heads branch from a shared internal representation: one predicts three-dimensional coordinates through regression, the other classifies which of six target odorants a mixture contains.
The six target compounds, ethyl butyrate, ethyl hexanoate, 2-heptanone, α-terpinene, allyl methyl sulfide, and diallyl sulfide, were chosen to span multiple chemical families and to include structurally similar pairs with distinct volatility and transport kinetics. This design deliberately challenges the model to rely on both chemical interactions and diffusion-driven temporal cues.
The researchers trained the model on 50 plume trials involving random combinations of these six compounds released from random positions. It achieved 86.7% overall accuracy in identifying mixture components and localized sources with a mean error of 2.84 ± 0.87 cm.
Performance varied by compound. Ethyl hexanoate reached 100% recognition, while 2-heptanone and allyl methyl sulfide each scored 98% and α-terpinene 96%. Diallyl sulfide dropped to 72%, likely because its chemical profile closely resembles that of allyl methyl sulfide. Ethyl butyrate proved most difficult at 42% because its low-amplitude, fast-plateauing signals become masked by stronger co-present odorants.
Computational fluid dynamics simulations validated these results, confirming that observed sensor patterns match diffusion behavior predicted by established transport equations. The simulations showed robustness across an indoor temperature range of 15 to 25 °C.
The team tested the concept at room scale as well. Three sensors mounted on a mobile robot at the right-front, left-front, and rear recorded responses as commercially available liquor was poured at nine random positions in a 3 m by 3 m conference room. Temporal and amplitude differences among the three sensors reliably indicated both direction and distance. A trained classifier correctly identified test positions, and the robot navigated directly to each source.
AROMA reframes artificial olfaction as the decoding of a dynamic, spatially structured field rather than the classification of a static chemical snapshot. The principle applies wherever molecular information travels embedded in transport phenomena. Industrial leak detection, indoor air-quality incidents, and autonomous navigation in visually obscured environments all require knowing both what a chemical release is and where it originates. A single stationary network that infers both from passive plume exposure offers a simpler alternative to combining separate identification and localization hardware.
For authors and communications departmentsclick to open
Lay summary
Prefilled posts
Nanowerk Newsletter
Get our Nanotechnology Spotlight updates to your inbox!
Thank you!
You have successfully joined our subscriber list.
Become a Spotlight guest author! Join our large and growing group of guest contributors. Have you just published a scientific paper or have other exciting developments to share with the nanotechnology community? Here is how to publish on nanowerk.com.
