Machines see more and better, hear over a larger area and stronger, and move surprisingly efficiently (for robots). And yet they still have a problem with something that is obvious to us. The latest research review published in Nature shows that robots can also acquire a sense of smell. And it will be better than any other animal on Earth. Potentially.
In biology, smell is a survival tool. It was ignored by machines for a surprisingly long time… Humans are not the best reference point here – our sense of smell is subjective, unreliable and completely useless when detecting low concentrations of volatile organic compounds or gases that we consider “odorless”. This is a serious gap for robots operating in danger zones.
Noses built with the help of electronics are finally going to change that. Instead of one “odor sensor”, arrays of sensors that respond to various chemical signals are used, and then algorithms interpret their patterns. Everything is fine: until the smell starts to mix with the wind, temperature and movement of the robot itself. Then the algorithms have problems and start confusing the compounds that linger in the air.
Sensors versus reality
Overview in Nature compares several major classes of sensors used in electronic noses. The most popular are semiconductor sensors of the MOS type: because they are cheap, common and easy to implement. At the same time, unfortunately, they are characterized by poor selectivity and problems with long-term stability. A robot with such sensors will literally torment the operator with false alarms about escaping various gases.
Electrochemical sensors perform better in terms of selectivity and energy consumption, but are sensitive to environmental conditions and age much faster. Optical measurement systems, on the other hand, offer high precision and stability, but miniaturization and heat management are more difficult there. However, FET sensors are well suited to IoT systems, although it is difficult for them to detect many gases at the same time. So many possibilities, and such a huge problem with choosing.
How to find the source of the smell
Detecting the smell is just the beginning of the job. It is much more difficult to determine where it comes from. Researchers point to bionic algorithms, inspired by the behavior of insects, which are fast and energy-efficient, but perform poorly in dynamic environments. There is also a “team approach”, where robots with variously specialized sensors carry out coordinated actions: but let’s agree. This type of thing has a scaling problem. Everyone would like to have one robot that rules all… smells.
Probabilistic algorithms based on Bayesian inference offer the greatest precision. Unfortunately, there is a problem here: the high cost results from very difficult, heavy calculations. Hybrids are becoming more and more common – combining local, even “instinctive” search for a scent with probabilistic correction. That’s more or less how nature works.
The robot will find work everywhere
These types of robots can be used, among others, for: searching for people trapped under the rubble: there, human body odors are one of the few signals and at the same time very accessible to robots. Biohybrid drones that use the sensitivity of insect antennae to detect pheromones can operate on farms or apiaries. “Wearable” e-noses monitoring patients’ breathing in real time can analyze the level of CO2 or metabolites detectable in the breath, which would help, among others: diabetics. There are plenty of possibilities, as well as methods.
However, all this requires reliability outside controlled conditions. Unfortunately, we are still “limping” here and it is difficult to say when this will change and what new opportunities science will open to us.
Theory is its own, practice is its own
The movement of the robot disturbs the air flow and blurs the scent trail. Miniaturization of sensors reduces their efficiency. In the case of algorithms: these often ignore the physics of diffusion and turbulence, which causes them to get lost in changing conditions. Not to mention the time synchronization of such an olfactory sensor with, for example, lidar. In this respect, for example, dogs’ noses (and brains) are so amazing that we simply can’t keep up with biology. So we have systems that work great in a wind tunnel, but can fail completely in an open field.
The authors of the publication say directly: the future of robots with “electronic smell” lies in interdependent design. All factors must arise together, not in isolated teams. In addition, there is the need for lightweight AI algorithms running locally, without cloud support, and tests, tests and more tests – in wind, dust and complete chaos. Until we have this, robots will see perfectly, hear phenomenally, but “smell”… tragically.
Read also: Food delivery robots are causing more and more problems. This video says it all
Missing Machine Sense
If we can combine precise sensors with algorithms that understand the physics of smell, we will have systems that can operate where our senses reach their limits. We have the potential to overcome even the most sensitive noses in the animal world. However: not now, or even in five years. Maybe in a decade we will hear about electronic noses that will be able to match what biology has to offer. Besides, researchers are already working on it and they will probably surprise us very much one day. By the way, artificial intelligence plays a significant role in this, as it will help create solutions tailored to the freshly started second “quadrant” of the 21st century. century.
