As reported by The Physics arXiv: If you've ever watched huge flocks of starlings wheel and
pulsate in the early evening sky, you’ll know how spectacular this
phenomenon can be. Starlings probably flock to reduce their chances of
attack by predators such as peregrine falcons.
But
their are many other reasons why this kind of group behavior is
useful—more effectively navigation and foraging, for example. So it’s no
surprise that various teams want to reproduce the same kind of
behavior in artificial systems.
That’s turned out to
be harder than it sounds, particularly for aerial flocking. So while
various groups have created semi-autonomous flocks, and one group
created a flock of slow-moving blimps in a gymnasium, nobody has
successfully demonstrated a fully autonomous flock of robotic flyers
that can fly outdoors.
Until now. Today, Gabor
Vásárhelyi and pals from Eötvös University in Budapest, Hungary, reveal
that they have successfully demonstrated the first autonomous flying
robots capable of flockin
g outdoors, just as starlings do.
Flocking
is hard because of each member of the flock has to be able to sense its
environment and respond to changes quickly and accurately. That means
each flyer has to measure its own velocity and position, the position
and velocity of those around it, and use this information to calculate
what to do next.
Even then, it must be capable of
carrying out the necessary flight adjustments quickly. These machines
must be able to hover, to fly in a specific direction at a specific
speed and to change tack rapidly.
That’s
a big ask but it has looked increasingly possible in recent years given
that stable flying machines such as helicopters and quadcopters are now
commercially available.
Vásárhelyi start with a
commercially available quadcopter known as the MK Basicset L4-ME from
the German company MikroKopter. This is capable of self-stabilized
flight and is controlled using a handheld remote.
An example of the navigation module used to help provide the flocking, formation, target tracking, and obstacle avoidance features. |
The
team’s first step was to modify these machines to make them autonomous.
They did this by attaching an extension board carrying a variety of
navigational devices such as a gyroscope, accelerometer, GPS receiver
and so on as well as a wireless communications unit and a minicomputer.
During
flight, each flyer constantly broadcasts its position and velocity to
the others which then determine their own actions using the team’s
flocking algorithm. This essentially implements two rules, a short range
repulsion that prevents adjacent flyers from colliding, and a rule that
aligns their velocity and keeps adjacent flyers going in roughly the
same direction at the same speed.
Simulation of actual data generated from target tracking exercises. |
Two reduce the
amount of calculating each flyer has to do, the quadcopters all fly at
the same altitude so that the flocking problem becomes a 2 dimensional
one.
There is also a ground-based PC monitoring what’s
going on and this can make real time changes to the algorithms
controlling each flyer. But crucially, the flock does not rely on any
centralized control for its behavior.
The results are
impressive. These guys have successfully flown flocks with up to ten
quadcopters in the air simultaneously. “We successfully established the
first decentralized, autonomous multi-copter flock in an outdoor
environment, with swarms of up to 10 flying robots, flying stably for up
to 20 minutes,” they say.
They also find a number of
interesting behaviors. One of the key problems these guys have to
handle is the inevitable delay each flyer experiences between receiving
information, processing it and then performing the necessary changes in
flight.
Target tracking a moving vehicle. |
This kind of delay can lead to all kinds of
interesting oscillations within the flock. But it can also destroy the
flock if it gets out of control. So the flight algorithms must be
carefully fine-tuned to damp out the destructive effects.
But
with this in hand, Vásárhelyi and co say they were able to fly their
copters in various formations such as ring shapes and in lines. They've
also observed self-organised behavior in which the quadcopters fly in
lines and circles within pre-determined boundaries, just as locusts do
in a similar circumstances.
Obstruction avoidance simulation - the flock is attempting to reach the large geofenced area, while staying out of the smaller obstacle areas. |
There are limits on the
behavior of the flocks, of course. The speed of he flyers determines
the required braking distance between them to prevent collisions. And in
any case, the average distance between the flyers is between 6 and 10
meters. This is determined by the positional accuracy of the GPS sensors
which is to within 2 meters or so. Tighter formations will require
better positioning accuracy.
Nevertheless, this is an
exciting first step in this area. The potential applications for flocks
are numerous. The researchers imagine using them for large-scale,
redundant observations over wide areas, perhaps for farming, traffic
monitoring and, of course, military purposes. And equipped with sensors,
flocks could monitor large volumes of the atmosphere for signs of
pollution. A real advantage in all this is the redundancy that a flock
offers if one unit malfunctions.
With the first
success in outdoor autonomous flocks, it shouldn't be long before we see
more of these in the real world. Perhaps one day we’ll even be able to
watch them gather at dusk and demonstrate their own spectacular aerial
displays.
Ref: arxiv.org/abs/1402.3588 : Outdoor Flocking And Formation Flight With Autonomous Aerial Robots