The review explores the sensor technologies and odor-localization algorithms behind this capability, along with real-world applications ranging from search and rescue to healthcare. It also outlines the technical challenges that must be addressed to bring these systems out of the lab and into everyday use.
Why Smell Matters in Robotics
Olfaction plays a vital role in biological survival, but human smell is limited, inconsistent, and ineffective for detecting odorless or low-concentration threats like certain volatile organic compounds (VOCs). To address these shortcomings, engineers have developed e-noses: sensor arrays combined with algorithms that mimic the way animals detect and interpret smells.
While earlier research has produced various gas sensors and source-localization algorithms, many of these systems still struggle in the unpredictable conditions of the real world. This review tackles those gaps head-on, examining how recent sensor innovations and smarter algorithms are pushing robotic olfaction toward broader, more practical deployment.
Sensors and Algorithms That Power Robotic Smell
At the heart of robotic olfaction are two main components: the gas sensors that detect chemical signatures, and the algorithms that help the robot locate where those odors are coming from.
The review compares four primary types of e-nose sensors. Metal oxide semiconductor (MOS) sensors are widely used for their high sensitivity and affordability, though they struggle with selectivity and long-term stability. Electrochemical (EC) sensors offer better selectivity and lower power use, which is beneficial for portable systems, but they tend to have a shorter lifespan and are sensitive to environmental conditions.
Optical sensors provide excellent stability and accuracy, and recent research is pushing them toward smaller, multifunctional formats. Meanwhile, field-effect transistor (FET) sensors are highly sensitive and well-suited for integration with Internet of Things (IoT) platforms, but face challenges with detecting multiple gases and maintaining stability.
Alongside these, niche sensor types - like catalytic, capacitive, or quartz crystal microbalance (QCM) sensors - serve more specialized roles in detecting combustible gases or offering ultra-high sensitivity in controlled environments.
Once odors are detected, localization algorithms help robots trace them back to their source. The review outlines three main approaches.
Bionic algorithms, modeled on the behavior of insects like moths, are simple and efficient but often too rigid for dynamic environments. Multi-robot collaborative methods use coordination strategies (such as particle swarm optimization) to improve search efficiency, though scalability can become an issue. Bayesian probabilistic algorithms stand out for their accuracy in uncertain conditions but demand more computational power.
To overcome the trade-offs of each category, researchers are developing hybrid approaches. For instance, combining bio-inspired local search with Bayesian inference to improve robustness, accuracy, and speed in more complex environments.
Real-World Applications
This fusion of advanced sensors and intelligent algorithms opens the door to a wide range of applications. In search and rescue, robots outfitted with both tactile and olfactory sensors are being developed to locate people buried under rubble. Biohybrid drones, incorporating insect antennae, have been used to detect pheromones with extraordinary sensitivity. In healthcare, wearable e-noses are being explored for real-time respiratory monitoring by detecting CO2 and other biomarkers.
Each of these applications demands not just sensitive sensors or clever algorithms, but tight integration between hardware and software and reliable performance in settings far removed from the lab.
Barriers to Deployment
Despite promising progress, deploying robotic olfaction systems outside of controlled environments remains a challenge. Much of this comes down to the interdependent limitations of sensor technology, robot integration, and algorithm design.
For instance, sensors must balance sensitivity, selectivity, stability, and energy efficiency, often with trade-offs. MOS sensors may detect trace gases in turbulent outdoor environments, but their broad sensitivity can make it hard to distinguish between different chemicals. Rapid response and recovery times are essential for tracking fleeting odor plumes, but they’re often at odds with long-term stability, as sensor materials degrade over time due to environmental exposure.
Miniaturization introduces another layer of complexity. While smaller, low-power sensors are essential for mobile robots, downsizing can reduce performance or create thermal management challenges, especially for optical sensors.
Even when sensors perform well individually, integrating them into robotic platforms presents new issues. For example, a robot’s own movement disturbs the air and scent plumes around it, complicating odor tracking and requiring airflow-aware designs. Synchronizing slow chemical sensing with faster systems like vision or lidar also demands efficient edge computing.
Algorithmically, many current approaches don’t yet incorporate fluid dynamics, which limits their ability to navigate complex, shifting environments. And while combining smell with other inputs like thermal or visual data can improve navigation and mapping, most existing systems don’t fully account for uncertainty or changing dispersion patterns.
As the authors note, robotic olfaction is still largely untested in the field, where fluctuating wind, temperature, and complex chemical mixtures pose real and unpredictable obstacles.
What’s Next: A Unified Design Approach
To move forward, the authors argue for a comprehensive co-design strategy. One that doesn’t treat sensors, algorithms, hardware, and integration as separate challenges, but rather addresses them together. That includes the development of flexible, miniaturized hardware; active sniffing systems designed with aerodynamics in mind; and lightweight, timing-aware AI algorithms capable of operating at the edge.
Algorithmically, the path forward lies in embedding simplified transport models into 3D navigation and mapping, allowing robots to better predict how odors move through space. But above all, future systems need to be rigorously tested outside the lab, across diverse, uncontrolled environments, to validate their real-world performance.
Conclusion
This review paints a detailed picture of where robotic olfaction stands today, and what’s needed to take the next step. Powered by increasingly capable e-noses and smarter algorithms, these systems hold enormous potential for improving safety, health, and situational awareness in environments where human senses fall short.
But realizing that potential will require more than just better sensors or faster computation. It demands a tightly integrated, field-tested approach that combines hardware innovation, adaptive algorithms, and an understanding of the complex physical dynamics of odor dispersion. With that foundation, the next generation of robots may not just see or hear the world, but smell it too.
Journal Reference
Zhang, W., Tang, W., Wan, Z., & Fan, Z. (2026). Advanced electronic noses for future robotic olfaction. Npj Robotics, 4(1). DOI:10.1038/s44182-025-00071-y. https://www.nature.com/articles/s44182-025-00071-y
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.