Tip:
Highlight text to annotate it
X
Nature is full of fascinating group motion patterns.
These patterns are universal.
Mammals, birds, fish and even bacteria organize themselves
into very similar coherent flocks, herds or swarms.
This raises many interesting questions.
First of all, why and how do these wonderful animals exhibit
collective motion, and what can we learn from them?
Pigeons are known to fly in a very coordinated way
and have a great ability to navigate home.
We have recorder tracks of pigeons by miniature GPS devices they carried.
Animated videos of the recorded data points
vividly visualize the delicate collective decision making process taking place,
while the pigeons are trying to find the best way during their flight home.
It is almost like a permanent voting process.
We at Eötvös University, Budapest, are studying collective motion.
This research is supported by the European Research Council.
Collective motion refers to flocks of birds, schools of fish,
to all sorts of complicated flight or collective motion patterns in nature
exhibited by gregarious animals.
We came to the conclusion that
one of the best ways to understand how animals move together
is to build robots, flying robots, drones in this case.
We can control these robots
and through the control, we can understand
that a given factor how affects collective motion.
This way we understand better pigeons or other birds.
By watching and understanding animals,
we can build better robots.
So these things go back and forth.
As an excercise towards three-dimensional robotics,
we built a pool and experimented with a
system of cheap, radio-controlled toy boats.
These little robots were not autonomous.
Nonetheless, they still reached a global, spontaneously rotating motion pattern.
About five years ago,
we decided that the best vehicles (way) to study aerial flocking of robots
is to build autonomous drones.
We bought a dozen of relatively cheap,
commercially available quadrocopters
that can be controlled with a manual remote controller.
We equipped the copters with a home-made hardware layer
that turns these machines into self-steering drones.
This hardware layer integrates all the incoming data from the environment,
from other drones,
and creates high level steering commands,
like the actual desired velocity or position.
With the proper flocking algorithms fed to this new brain,
the copters are able to fly autonomously.
We could totally eliminate the need for manual control,
and the group of quadrocopters can preform flights and tasks on their own.
In spite of the sophisticated and realistic simulation framework
we still had a lot of crashes on the field.
So we had to create
mechanical protection around the copters.
For example a special landing gear was developed in our lab.
Thus we have decided to build
a flock of autonomous robots.
A car is not autonomous.
Of course, a car can also be autonomously behaving,
if it has a brain
either in the form of a person
or in the form of a huge computer, equipped with sensors.
We had to do this for small flying objects.
Prior attempts to produce flocks of quadcopers,
included an approach which was confined to a given area
and the individual robots had to communicate
with a central computer outside.
In our case, each quadcopter
is equipped with a little brain in the form of an on-board computer
They are completely autonomous
in the sense that all of the decisions concerning their
directions, flight positions, are decided by themselves,
calculated, based on the information they have
on their own position and on the position and the velocities of the other drones.
In this way, the whole flock becomes autonomous.
And in addition to being just a simple set of autonomous agents,
act(ing) together collectively
the performance of such a flock
is way beyond just being a simple sum of the individual performances.
After the drones were instructed to form a circle,
each drone finds its position along the circle.
Even the direction of rotation is decided collectively
by the positions of drones before the circle formation.
If drones receive an external command
to change the shape of their formation,
they are able to reorder autonomously into a line segment
or a grid.
Drones must actively compensate for all kinds of disturbances,
like wind, GPS outages
the vibrations induced by the rotating propellers
or the delay in the local communication network.
Therefore, we started our project with realistic simulations
that take into account all these sources of errors.
In our computer simulations the copters are shown as purple circles
and white arrows indicate their actual velocities.
With a suboptimal choice of the parameters
the motion is irregular.
However, if the system is tuned well,
oscillations disappear and the desired flight patterns emerge.
The quadcopters follow a so called informed individual
that leads them on a predefined trajectory.
This time, the drone, playing the role of the leader is semi-autonomous.
It is controlled a program to move along a given route,
but it makes autonomous decisions to avoid collisions.
Actual GPS track records of the flights are shown in the animation
for better visuaization of the process.
This is the case of the simplest hierarchical control.
It has two levels only, leaders and followers.
Extension to multi-level hierarchical control
what was observed for pigeons, for example
has a great potential and it is a very promising future direction.
A number of possible collective flight patterns
are reproduced by our flock of drones.
It is possible to define imaginary soft walls
that are not penetrable by the robots.
If the wall is cylindrical
or it is a rectangle around the flock,
then the flock moves around a self-organized semi-random trajectory.
If there are obstacles in the way
and there is only a narrow gap they can move through
then the robots show signs of jamming.
Drones are most commonly associated with war, terrorism and cyber-attacks,
but drones can be used in more peaceful, civil applications as well.
In the century of global warming,
low-cost environmental or agricultural monitoring is a must.
With a flock of drones you can create a self-organized
monitoring system from the air,
without any disturbance on the species,
animals or plants that you are observing.
But drones can be used in many other applications.
You can assist rescue operations, create ad-hoc communication networks,
or you can even deliver food or mail.
Our future is most probably overwhelmed with drones,
and it is our responsibility how we use them.
Indeed,
studying the flock of robots, of drones,
has allowed us to understand much better
the group flight of birds, for example.
It is by now clear that
the reaction times between the units
have to be extremely short,
in order to maintain coherent motion.
And in addition to that,
the units have to be able to predict the next move of the others
to maintain the high level of coherence.