There’s a swarm coming: how robots are learning to solve human problems

If you have ever lazed away a summer afternoon watching an army of ants work together to carry a leaf back to their colony, you will no doubt have marvelled at the industrious little beasts.

Their teamwork is something scientists have long been keen to emulate in robots, with so-called swarm intelligence believed to hold the key to solving a variety of human problems. 

Imagine an army of cyborg insects that are able to march along a human capillary in the way those garden ants transgress your patio, carrying not leaves but life-saving drugs to be delivered directly to where they are needed in the human body.

Wildfire detection

Or how about a team of drones that can monitor the hills for bushfires to protect cities from disastrous wildfires such as those which recently devastated communities in Los Angeles?

Sabine Hauert is Professor of Swarm Engineering at the University of Bristol. She presides over a team of scientists, affectionately known as “Swarmies”, and they are attempting to solve this very problem.

Working with a company called Windracers, which operates autonomous aircraft, the team is looking at using swarm robots to both monitor and extinguish fires. 

“We have the ability to detect things automatically, which we do using AI and image processing, and then we can use these aircraft to cover a huge area, which gives us the ability to extinguish fires before they grow into something bigger,” she tells BI Foresight.

Her lab has already done experiments in the UK with drones dropping water balloons and extinguishing a small fire, and alongside Windracers is taking part in an $11m, four-year competition for firms using technology in innovative ways to end destructive wildfire. Dubbed XPrize Wildfire, it is part of the wider X Prize Foundation, which is on a mission to solve some of humanity’s biggest challenges using tech. 

Prof Hauert is very aware that any such system would involve humans and robots working together.

“We’ve spoken with 50 firefighters about what it would take for them to embrace the swarm within their current work protocols and we are working on an interface, so they can just click a button. The swarms use the algorithms we design to split up and search different parts of an area. What was interesting in those conversations is that the firefighters wanted to have control over how the fire was going to be attacked, with a line on the screen to say how the robots would swoop down and deliver the water,” she says.

Hissing cockroaches

Large groups of robots are already operating in the world, perhaps most notably in places such as Amazon’s vast warehouses. But the key difference between these logistics robots and swarm robots is that while the first are controlled centrally, swarm robots are self-organising.

Utilising self-organising robots in the real world is complex. “In the past, swarm robots were very simple, minimal robots that could interact with their environment and through these local interactions beautiful swarm behaviours emerged. And because they only relied on those local interactions they could scale to huge numbers,” says Prof Hauert.

“But when we get them out into the real world, those same simple robots now need to be more sophisticated, because they need to understand the world around them. And the humans who are managing these robots at scale need to be able to give instructions to the swarm, to monitor what the swarm is doing and have confidence that what it is doing makes sense.”

“The idea is to have micro robots detect and treat diseases. They would be administered intravenously or orally and using external energies like ultrasound and magnetism…”

Assistant Professor Simone Schürle-Finke

Advances in swarm intelligence are being made all the time.  A recently published study in Nature Communications describes how scientists at the universities of Hiroshima, Osaka, and Nanyang had developed a new algorithm which allows a leader insect to give the intended destination to others, while also allowing them to adapt dynamically and help each other overcome obstacles.

For the experiment, the team used Madagascar hissing cockroaches, which were equipped with a battery-powered backpack containing a navigation system, circuit board and sensors. Those working on the study said that such a system could be used in search and rescue. 

Meanwhile in December it was announced that scientists in South Korea had shown off a swarm of 1,000 microrobots performing some Herculean tasks, including turning themselves into a raft that floated on water and wrapped around a pill weighing 2,000 times more than each individual robot to carry the drug through liquid. Another swarm was shown unclogging tubes that resembled blocked blood vessels.

“While the study’s results are promising, the swarms will need higher levels of autonomy before they are ready for real-world applications,” says author Jeong Jae Wie from the Department of Organic and Nano Engineering at Hanyang University in Seoul.

Bitcoins for robots?

Marco Dorigo is considered one of the fathers of swarm intelligence. His PhD, completed 30 years ago, looked at applying the way insects behave to software agents to solve difficult mathematical problems. Fast forward to today and now he is looking to human interaction to help advance the field.  

He understands all too well the conundrum faced by the South Korean scientists.

“One of the great limitations of swarm robotics is self-organisation, which means that no robot is in charge and they all have approximately the same status and only interact with robots that are nearby,” says Dorigo. “But we know from human societies that complete self-organisation is not the best option. There are hierarchies where people have different roles and responsibilities. So what we have been working on in the last five years is a logical structure that is kind of a hierarchy, so there is a leader but there are also sub leaders. For instance, if you have a leader and that robot breaks down, someone else takes the lead automatically, without any external intervention.”

Cooperation between humans isn’t always a given, and in the robot world Prof Dorigo is looking at ways to incentivise good behaviour using blockchain technology.

“When a robot co-operates with other robots they gain crypto tokens and if they act in a way that is not acceptable for the other robots, they lose them. If they act badly too often, they run out of tokens, so they cannot participate anymore.”

A harsh reward system, even for a robot.

Future farming

An important question about swarm robots is; when they will be useful in real-world scenarios and what industries they can transform?

In Rome, Vito Trianni, Researcher at the Institute of Cognitive Sciences and Technologies is looking to agriculture. He envisages a future where both drones and land robots could provide insights to farmers about where pest outbreaks may occur, or predict when is best to harvest crops, or how much water or fertiliser to use.

His work began by looking at how he could transpose algorithms from abstract scenarios into real-world applications.

“It turned out to be not that easy at all because of the huge variability,” says Trianni. “So now we are looking at adding features that make the system useful for farmers. And once you reach that level then the problem starts to be the individual robot’s abilities. If you have a ground robot that is able to navigate easily around a vineyard then you can scale it up to multiple robots but the issue is that the navigation for the single robot is not fully solved.”

Trianni adds that what he is doing now is trying “coordinated behaviour from multiple robots in order to improve the perceptual abilities of the single robot.”

He is also looking to apply his research to undersea robots, which could be used for monitoring ocean changes and pollution levels. In the vastness of the ocean, swarm robots make a lot of sense.

“The single robot would not have the ability to perceive anything sensible. It’s only the collective that can do that because it can gather information on a wider scale and detect patterns,” he says.

Marching out of the lab

A bit further out, in terms of being ready for the real world, is the use of micro swarm robots in healthcare. It is an area of research that occupies much of the time of Assistant Professor Simone Schürle-Finke at the Biomedical System Lab at ETH Zurich.

“The idea is to have micro robots detect and treat diseases,” explains Schürle-Finke. “They would be administered intravenously or orally and using external energies like ultrasound and magnetism, we engineer them in a way that if there is a certain disease marker present it will affect the way the robots respond to the ultrasound.”

“They can check how the liver or the gut is doing, or check for inflammatory bowel disease.”

Swarm robots are perfect also for drug delivery, she says.

“If it’s a single robot, it’s just too big, so you need to distribute the drug amounts to several that can go as a swarm. And they can change their shape to fit through vessels.”

She predicts such systems could be in use in hospitals within the next decade.

For Prof Hauert, it is a very exciting time for the field in general.

“We’re at a stage in swarm robotics where we are looking to get the amazing research that’s been done over the last two decades in the lab, out into the real world,” she says. “The reason it’s happening now is that robots are becoming more capable. They can detect that there is a wall, or there is another robot. Now we have the technical capabilities in terms of perception and reasoning on the robot itself, then it is just a case of getting it out there.”