A flock of copter drones flies at night. Credit; A.Korbely.
A flock of copter drones flies at night. Credit; A.Korbely.

A Bird-Like Flock of Autonomous Drones

This story appears in shorter form at Nature News.

In a field outside Budapest, Hungary, ten quadcopter drones are flying as a coordinated flock. They zip through the great outdoors, fly in formation, or even follow a leader.

The little machines are the work of Hungarian scientists led by physicist Tamas Vicsek from Eotvos University in Budapest. They’re autonomous, meaning that they compute their flight plans on their own, without any central control. They can follow instructions, but they work out their own paths using GPS signals to navigate and radio signals to talk to one another. They’re the closest thing we have to an artificial flock of birds.

The copter flock is a real-life version of an influential computer programme called Boids, created by Craig Reynolds in 1986. He programmed virtual flying objects—the eponymous Boids—to move according to three simple rules. They aligned with the average heading of their neighbours; they were attracted to each other; and they also repulsed each other to keep some personal space. These three simple rules were enough to simulate a realistic bird-like flock.

Boids was massively influential for Hollywood animators looking to depict swarms of bats or stampeding wildebeest. But it also showed scientists that the behaviour of animal collectives could arise from individuals obeying similar simple rules, rather than hewing to some master plan or communicating telepathically.

Vicsek is one of several pioneers of collective motion, who have expanded on these principles over the last few decades. And for five years, he has been trying to apply them to actual robots.

It hasn’t been easy. Alignment, attraction and repulsion can keep a virtual flock together, but in the world of wires and rotors, they aren’t enough. “The big enemies are noise and delay,” says Vicsek. The GPS signals that the copters rely on are very noisy, making it hard for them to accurately discern their position. They also need time to receive and process those signals, and these lags mean they often get dangerously close to one another or overshoot their mark.

It took close to five years to solve these problems. The team even had to build their own bespoke electronics lab to make their own copters, since store-bought ones were too unstable and kept on crashing. “When they crash, they crash very quickly,” says Vicsek. “We could only do experiments in areas without people or animals.”

In the meantime, his competitors were building their own mechanical flocks but Vicsek says that most of the reported successes have cheated in critical ways. Some, for example, could only fly indoors. Others communicated with a central supercomputer that did all their processing for them and gave them precise flight commands.

Only Dario Floreano, based in Switzerland, has come close to a truly autonomous flock. He created a group of fliers that can move together in outdoor environments, but they were hardly manoeuvrable. They could only move at the same constant (and slow) speed, and they avoided crashing into one another by flying at fixed (and different) heights. They were autonomous and impressive, but they offered a pale comparison to the dynamic flights of birds or Boids.

By contrast, Vicsek’s drones are free in their movements. Tell them to form a rotating ring, or a straight line, and they’ll coordinate themselves into the right position. Tell them that they’re heading towards an imaginary alleyway, and they’ll queue up to squeeze through a gap.

This isn’t just about aesthetics. “Like natural groups, this flock of robots is very robust to the failure or death of individuals, changing group size, and environmental perturbations such as sudden gusts of wind,” says Iain Couzin, another leader in the study of collective behaviour. “These emergent properties make self-organized robot flocks ideally suited to a wide range of applications involving efficient search and object delivery, especially in inhospitable environments.”

There are obvious military applications too, but Vicsek prefers to focus about peaceful ones. His son envisions a flock of sprayer drones that eliminate pools of stagnant water where mosquitoes breed. Vicsek himself likes to imagine quadcopters as artificial pollinators. “I think of these as future bees,” he says.

Reference: Vicsek has submitted a paper about this work as a presentation at the upcoming IEEE/RSJ International Conference on Intelligent Robots and Systems in Chicago, Illinois.

For more on collective motion, check out my big Wired piece on the science of swarms.