At first glance, it could be something from a David Attenborough programme. A writhing cloud of black dots stretches across the sky. It’s one of the great spectacles of the natural world and, as it draws closer, you can see why.
Inside the swarm, each dot darts around as though it’s going about its own business. Yet the overall effect remains one of unison. Thanks to ingenious navigation, or some form of organised chaos, they move as one.
As the swarm draws closer still, it becomes clear just how astounding it is. What had sounded like hundreds of beating wings is in fact something mechanical, motorised, man-made. What, from a distance, had looked like starlings or locusts are in fact tiny helicopters, each one small enough to land in the palm of your hand.
They’re robots, and their task is simple: find human beings. “A flock of autonomous helicopters like these could be extremely useful in searching for survivors after disasters,” says Professor Pei Zhang, who recently built the first prototypes for the swarm of robo-choppers, known as SensorFly. “After an earthquake, or during a fire, constant information about collapsing walls or immobilised survivors would be incredibly useful to human rescuers.”
SensorFly, which Zhang is working on at Carnegie Mellon University in the US, is the latest robot swarm to be set loose on the world (or at least the lab). It’s an approach to robotics that seeks to build something very different to the walking, talking tin-can servant we were all promised in the 1950s. In swarm robotics, the thinking is that 200 metal heads are much better than one.
Taking inspiration from industrious social animals such as ants and bees, researchers believe that the kinds of tasks that robots will be set in the future will be best tackled by a large number of bots all working together. Just like a food source is quickly found by a colony of ants all marching out to find it, so earthquake survivors are more likely to be found alive if the search party is made up of hundreds of miniature robots looking for them.
“A swarm’s main characteristic is that the robots are spread about in the environment. There is an enormous number of applications for them,” says Professor Alan Winfield, a roboticist at the University of the West of England. “If you want to explore the asteroid belt, don’t send one spacecraft, send a swarm. If you want to build a habitat on Mars, send a load of raw materials, and a swarm of robots to assemble them.”
Farmers and fighters
Robot swarms are not a new idea (work has been going on since the 1980s). But swarms are becoming more sophisticated, and the latest designs are being thought up to perform a bewildering array of tasks. From harvesting crops to monitoring the environment, from crawling through a rubbish dump looking for useful raw materials, to swimming through your veins delivering medicine, swarms could be very busy in the next decade or so.
Winfield believes the first robot swarm to escape the lab, however, will come from an arena that’s already crawling with robots: warfare. “I’m sorry to say, but I think the first swarms will come in the military domain, where the budget constraints are somewhat different. I think we’re likely to see flocks of unmanned aerial vehicles where a human operator flies one of them but, the rest are flocking like birds.”
The US military already has its first swarm squadron on order. In 2008, the US Army Research Laboratory awarded a $38 million contract to BAE Systems to lead a team of researchers in developing a swarm of spy-bots to be used in reconnaissance missions. The project, known as MAST (Micro Autonomous Systems and Technology), is now underway and BAE was eager to show off what it could do.
In an animated video of what it hopes to build a swarm of tiny robotic insects, including spiders like those in Minority Report, sweep through a building where two presumed terrorists are hiding. The bugs slip in unnoticed, track down the targets and send video back to soldiers outside. Seconds later, a missile reduces the building to rubble.
Presumably the robot spiders are destroyed alongside their human targets. But that’s okay – a swarm’s strength is in its numbers. If any single robot malfunctions or is somehow taken out, the swarm is able to continue. “It’s a ‘distributed system’,” Winfield explains. “There’s no mothership coordinating everything, which means that any of the robots can fail and the overall system will keep going. It gives you reliability.”
Dozens go out
It’s a system where everyone and no-one is in charge. Somehow, stuff just gets done. Think about an ant colony. When it builds a nest, there’s no architect or senior structural engineer barking orders. When the ants wage war on another colony, there’s no general giving the order to go over the top. Instead, it’s what scientists call emergent behaviour or swarm intelligence.
Each ant simply does what comes naturally. He goes about the little jobs hardwired into his head by millions of years of evolution. He reacts to cues in his environment. He interacts only with his closest neighbours. And when every ant in the colony does the same, mind-blowing feats of organisation emerge. The colony builds, feeds and functions.
Social animals like these are the ultimate inspiration for researchers attempting to build robot swarms. Professor James McLurkin, one of the world’s best, keeps his own colony of ants in his lab at MIT and follows the work of biologists like the rest of us follow movie stars or footballers.
“One of my favourites is Thomas Seeley at Cornell,” McLurkin says. “He looks at honeybees and how they decide where they’re going to live.” When a population of bees grows very large, it splits in two. The old queen and around half the worker bees pack up and leave. As scouts find them a new place to live, a new queen is installed at the old nest. “Dozens of scouts go out,” McLurkin says. “Somehow, as the scouts collect information, the group is able to reach a decision about where they want to live – there are a lot of subtle things going on.
“I idolise honeybees. It’s very humbling to see what nature can do because we’re not even close.”
Not that anybody is giving up, of course. Getting a swarm of robots to behave like a colony of bees or ants remains the Holy Grail for researchers in the field. But getting there is going to take a lot of caffeine and late nights, McLurkin says, because it’s a programming nightmare.
The main goal is to set a ‘global task’ for the swarm – finding survivors after an earthquake, say. You then have to break that down into individual robot tasks – such as avoiding giant slabs of concrete or following the heat signature of a human body. Each little robot only communicates (usually wirelessly or by infrared) with those nearest to it, but hopefully, through thousands of those tiny interactions, a swarm intelligence will occur among them.
“The magic happens when you run software to accomplish those individual robot tasks and the swarm somehow accomplishes the group-level task,” McLurkin says. “We can figure this stuff out, but it’s tricky.”
For now, much of the work in swarm robotics is very much fundamental. The swarm of robots in McLurkin’s lab, and others like them, are working on basic tasks such as understanding when they’re at the boundary of their network, and deciding whether to go beyond it.
On the horizon
Researchers are already imagining the next step, however. Along with roboticists from 10 European universities, Winfield is working on a robot swarm that can do things that even insects can’t.
“What we’re trying to do is build a swarm of robots that can work as a swarm when it needs to but, when the situation demands it, the individual robots can join together,” he says. “In other words, they self-assemble into a three-dimensional ‘organism’ that picks itself up off the floor and walks around.”
Yes, a bit like Power Rangers. The swarms, known as Symbrion and Replicator, would split up into different organisms suited to different tasks, each one specially adapted to the environment and problem the swarm is faced with. Winfield describes the process as a kind of artificial evolution. Going back to the idea of rescuing the survivors of an earthquake, he believes such a system would be invaluable.
“If the robots find survivors, some of them could self-assemble to lift off rubble. Others could self-assemble into a structure to perform first aid or pass water along a kind of robot chain.”
Like all other robot swarms, Symbrion’s strength will come from the fact that the group can achieve more than a single robot ever could. It might be 10 or 15 years before the technology is smart enough to be sent into the real world, but when it does buzz to life, the world won’t have seen anything like it. “Symbrion not only goes beyond anything that’s been attempted in the lab, it’s more or less beyond anything you see in nature.”