Nowadays, robots are everywhere,
their importance is growing in science,
society and industry.
They appear in fields, factories, space,
sea depths, gardens and in living rooms.
Some people think that in the 21st century
they'll be a bit like the 20th century car.
Robots haven't just entered industry,
they are taking part in our culture's
transformation, and sometimes,
they even take part in how
we understand ourselves, humans, in a new way.
So, what's a robot?
First of all, a robot is a physical machine
provided with a number
of engines to produce actions,
to produce movements, it also has detectors
or sensors as we call them
that enable it to perceive things
in its environment,
for example, a camera to see images,
micros to detect sounds,
force sensors to sense
when it touches an obstacle
or when it is bumped into, for example.
infra-red sensors to detect things that humans
can't see, but animals can.
The most important in the definition of a robot,
is that information is gathered
with the environment sensors,
it is used to decide which movements
will be produced by the robot.
So there is a permanent interaction
between perception and action,
that's what defines robots.
As opposed to automatons,
for example, Jaquet-Droz
or Vaucanson's automatons in the 18th century,
which were amazing machines.
Nevetheless, action didn't depend
on what was going on around them,
they always did the same thing,
regardless of what was happening around them.
A robot doesn't operate like that,
it reacts differently according
to what is happening.
So in practice, this definition of robots
covers a wide range of forms of robots,
it can be imaginary robots,
humanoid or animaloid robots,
but in the end, today,
they can only be seen in science labs.
This also concerns a lot of robots we live with.
Nowadays, planes are widely automated,
cars have driver-assistance system modes,
there are robot vacuum cleaners, for example,
or machines that will mow the lawn autonomously,
some toys too,
some fun robots in toy shops,
some of which are provided
with very advanced technologies. 
And this diversity of forms covers a rather wide
variety of content, operation and mecanism too.
So we can talk about a few great dimensions that
enable to differenciate macanisms. 
A first dimension is the autonomy. 
Some robots which behavior is autonomous 
because actions are a done while
it is perceiving its environment 
without permanent instructions or sometimes
just partial instructions, given from humans.
For example, robots in an automobile factory
which assemble vehicles,
sometimes with other robots,
well if humans don't interfere in the assembling 
and robots do it alone, 
they will be called autonomous robots. 
However, robots in nuclear plants,
which are operated by operators 
to go in radioactive confinement rooms
where the operator has a remote control
that allows him to define 
which actions they have to do at each moment, 
those robots aren't autonomous.  
An other dimension that differentiates
several families of robots,
is adaptation and learning.
Some robots' performance is completely frozen 
beforehand by a program and, 
whatever the robot will
experience in its environment, 
it will always behave the same way.
On the other hand,
there are other robots equipped with
other family mecanisms
called learning algorithms,
that enables them to update their action modes, 
and strategy actions according to the events 
they sense in the environment, according 
to tests and mistakes they will do. 
So, for example, there are robots that can explore 
an environment and gradually build up a map 
of this environment that will enable them 
to navigate in it in a rather efficient way 
although at first, they couldn't.
Some mecanisms will allow them to learn 
to recognize new objects a human will name, 
to learn to recognize
humans' actions for example, 
or learn how to move their legs,
learn locomotion for example.
These mecanisms are called learning algorithms, 
they enable machines to detect regularities 
in what they experience, 
an experience is made
when testing an action and observing the result.
It can be repeated a certain number of times. 
When detecting regularities,
they can understand 
how to predict the result
of their action, for example. 
Concerning how to find a solution,
a behavioral strategy, 
for example to learn how to walk,
there are test-mistakes strategies for example 
where algorithms
called optimization algorithms 
will enable them
to gradually refine the settings 
of these strategies to progressively
improve these solutions efficiency. 
In a number of cases, the solutions found
by these machines are quite surprising
or weren't predicted well by the engineer 
who designed the robot
or the learning algorithm. 
This doesn't mean these robots will invent 
completely unthinkable things,
but for example
it can be a locomotion strategy, a way of using 
one's body to move
which can seem a little absurd for a human. 
But observed carefully,
it is a very efficient strategy, 
and by using test-mistakes,
it's the robot which found it. 
There can also be learning algorithms 
a little more sophisticated 
that encourage machines
to explore an environment
and to search for novelty or information
to enrich their knowledge.
This will allow a machine
to develop its know-how list
which wasn't completely
programmed beforehand.
Nevertheless, we must
be careful when saying this, 
because we could get the impression
that the machines are going to be 
quite intelligent and creative,
but in fact  
they are still extremely far
from adaptation capacities 
of humans, of animals even,
and of more simple mammals. 
There's still a great amount of work
to do before being able 
to design machines which have the adaptation
capacities of a 3 or 4 month-old child.