Researchers hope to produce chemical-detection robots

Automated robot sniffer dogs could one day be seen searching airports for illegal substances, thanks to research at York University.

The project funded by the UK government’s Defence Science & Technology Laboratory (DSTL) aims to develop chemical-detection technology that could identify smells of items such as drugs or foodstuffs in unknown or changing environments.

Researchers from York’s electronics department have already developed a program for a hand-held device that can identify chemical signals from certain foods in the air under controlled conditions, but it takes around five minutes to calibrate the system.

Over the course of the three-year project, they hope to improve the software to speed up the learning process and fit it to a small remote-controlled robot that could move around different rooms, adapting to new chemical smells as it goes.

Project lead Prof Jon Timmis told The Engineer: ‘Airports are one example where this might be useful but there are lots, ranging from car parks and basic surveillance jobs to even the food industry, with some different sensors.’

The technology works by testing a sample of air using an ion-mobility sensor — a similar device to a mass spectrometer — to map the distribution of different chemical signals in the environment.

‘The system learns the profile of what an area looks like chemically, so when there are changes in the environment and another chemical comes in that shouldn’t be there, it has a relatively high probability of finding it,’ said Timmis.

‘You’ll get lots of false alarms as soon as you move so the system needs to be able to learn to adapt to that new environment without triggering alarms.’

When it enters a new environment, the system’s artificial intelligence learns what the standard distribution of signals looks like over time and uses pre-determined thresholds to decide when to raise an alarm about a change in the signals.

In order for the system to adapt to new conditions by relearning what a ‘normal’ signal is — and so reduce the time before it can be used in a new environment — the system needs a way to evaluate its performance so it can adjust the thresholds.

Timmis said he didn’t yet know how long the adaptation time could be reduced to. ’I would like it to be seconds to minutes. How realistic that is we’ll have to wait and see. But that’s where we’re aiming.

‘We’re hoping that we won’t rely completely on the chemical detection but also augment that… with different sensors. For example, using vision you might be able to associate specific smells with object recognition.’