and
[Music]
or
today I have the opportunity to share
with you the work that we are
developing in recent months
this can not be seen especially can
make decisions do not have plans like
this commercial what we have access
about development and research
this is a video of a robot that
We built for many years in
2005 a contest with minimum capacities
bush and routes but its
channel to be able to pick up the and so
take it to the goal
how can you imagine the priority that
really that we had was that I could
float of the ten or eleven teams that
participated
this was the only one that floated then
only for having floated we had already
won the contest but it was a good
experience because this robot for me
illustrates the challenges we have that
face in the investigation so that
the drones the robots or any other
artificial robotic intelligent system
can play activities in a
autonomous and also illustrates the
importance that these systems can
see verses environment
This is a video for me motivates me
reasons why we would like the
gifts you can see is a fixed-wing drone
let’s fly
we realized that it was not so easy
fly it and we decided to land it then
the vehicle comes with an option in the
which one button is pressed
begin to spin in circles begins to
descend and there is nothing that can not
do then it is important that the
drones can see
it would have been very opportune that this drone
at the moment we ask you to
land they could see that the improvement and
obstacles that could hit
It would have been very opportune that when
We requested that you land
best place to land or just
if we mark a desired place of
weighing that I could do then today
I want to talk to you about the things we have
fact that we are working on the
which we are investigating to treat
to solve these problems
some of these problems
this is a drone which transmits
videos can see it end is the video that
is on board you get with the camera to
board and with this video then we can
do many things we can process the
pictures or the images that are captured
at every moment every second and
applying sophisticated techniques
computational view of robotics e
artificial intelligence we can do
many things for example we can do
that our do not recognize the windows
gaps between windows that can
go through it
this is in the context of a
competition that took place this year
in the Mexican robotic tournament and one
of the challenges will be precisely that the
drone autonomously could
traverse the largest number of windows
we also teach identify color
so here we have one of the
judges who volunteered to
take the object shows the
drone so the drone recognizes that
color is object
and then keep the operator in fact the
judge was a bit nervous because
I had a little press in which the
ball was going to stick
then it ‘s like a dance can see that
the operator moves as he wants and the
Drone limits his movements
in fact the judge did not have a route
determined I could do what I wanted
and the task
The challenge is for the drone to maintain its
field of view to this object and could
follow him and he did it pretty well
so what do we do what can we do
if our drone with its camera on board
can take photos can take videos
but you can also move in a
autonomous
then we have thought about our gift
browne cameraman and today I will not do
the exercise that are the sisters
shows this video but usually I
I ask the audience to tell me
that the two videos which of the two believe
that is being controlled by a human
and which of the two by a computer
here they have the answer but
curiously most of the time the
people tell me that this is being
controlled by the human and can
appear the android party is
following the person around here is the
another is being controlled by the
human and what does this drone what
that smart system is that it has a
follower
I do not know if the region was and has a
follower identifies precisely the
person and is able to predict the
address where is old then
with that the drone can decide on its own
only when staying static because
the person is in his field of vision and
when you have to move scroll
to keep the person in the field
vision of the human pilot is capable of
do that you can do it very well but not
is more and when we compare the drone
with his artificial intelligence with
with respect to the human
the drone reacts faster then
this is the system that is running at
board with which we can
estimate where the person is the
field of view
the borders that have been placed for
that the drone just perform
movements and tries to keep the
objective
within this field you can see this
line what you are doing
unions the direction to where
is moving the person then with
that we can predict towards
where the goal is going to move and that’s
what determines the policy of
navigation that allows the drone to
autonomous way can travel move and
maintain this objective
then using these techniques
using again vision techniques
computational artificial intelligence
we can do several things for example
in this exercise
we raise our great about 80 meters
and what we did is that with the video that
transmit us earth
we think about generating high images
resolution
this is the video in low resolution are
800 x 600 pixels
however when it is moving
these images begin to be recognized
you begin to observe what is the
coincidence between images and then
we generated what is known as
mosaic of images that is the result
is a high resolution image that
can go from 10 thousand to 10,000
pixels up to 20 years
depending on the scanning area and most
important is that we can do this
in real time
then imagine the impact of
applications that we can do with this
we can go to a disaster area so
our rum and that generates this photograph
large panoramic area also what
What we can do is analyze the video and
analyzing the images that we
transmit the drama and generate a
three-dimensional reconstruction of this
space with which we also
we can know where the drone is the
rock trajectory without using cepes
and that is very valuable because there are many
contexts where drones can not
fly if you do not have bosses though with
this system
we can get a model
three-dimensional scene we can
know where the drone is and with that
we can use it and exploit it for
apply artificial intelligence so that
the drone make decisions
These are examples of models
three-dimensional textures that
we can generate but the interesting thing is not
it stays there we can then if we have
a three-dimensional model of the scene
establish a safety radio for
that when the drone follows a
trajectory identify what there
structures with which you can
collide and with that include it
just in the policy of good for
that the drone can be worth the
station here we have also been
working on improving their algorithms
improve these capabilities and justly
we were a competition last year in
china that is called imac where you can
approve small vehicles in this case
they could see that it took off from a
platform the movement had to
enter this house simulated so
autonomous had to leave had to
land on that platform too
the movement as it is illustrating
I had to pick up objects and take them
an error is a competition
very important international is one of
the most important ones like the Olympics
of the drones
here you can see the lifting of
objects that we make
we are the team that carried out the most missions
unfortunately there was a penalty
Why do not we process on board for your
masters with the images that this transmits
It is a three-dimensional map that generated
but we were fortunate to get the
second place in an event in which it is
the first time a Mexican team
participate and of course it’s the first
Once a Mexican team gets on the
podium of course not everything was very
good
this video illustrates one of the
capabilities of our great see what
What does he want
generally automobile crashed
against that foot or for example in one of
the autonomous landing tests fell
down liquidity of the rail 36 and it
the cart took it but nothing happens
then we have been working
intensely to improve this now by
example we have developed algorithms
that allow us to identify a zone of
Landing and land just as
different heights and using the
ability to generate a representation
three-dimensional scene to know
where is the drone without relying on the gps
and that with the visual information that
acquires the drone can make decisions
to carry out this landing
we have also been specializing
in the evasion of obstacles and for that
we apply what we know how to do with
regarding color recognition with
regarding the location of the vehicle
without using gps generating a
three-dimensional representation of the
scene and with that
the drone can decide what to do here
for example it is going to the front
as soon as he sees an obstacle he decides to go to the
left because that’s where there is more
space continues to the front
again when they face
obstacles decide where you need to go and
they will see
we have done this flight 40 times
changing the order poles changing
the tone of the colors can see
warm colors and we have managed that
work of course there has been in the
investigation process there has been
some collisions with the poles
We have been refined so much brown as
the poles because as they are cardboard
people break it but it has worked has
worked pretty well
so we decided to try this in a
more realistic environment only with
single-painted cardboard posts
color or true color and we went to
taste
so the idea is that the drone can
navigate follow a path so
autonomous and as soon as a
tree
then here they have our system
running in real time
flying autonomously and
identifying how we are here in
yellow that there is a tree
taking the decision to go towards
right or left depending
where there is space and showing that
we have a generally autonomous drone
able to make decisions based on the
ability to power
however this is not the last word
there is a lot to do and one thing that we have
been working is how to teach you a
drone certain things just as
do with children how can you learn
a drone in this case we got to
think about this challenge because it’s okay to
group identify color but yes change
the lighting
if the sun is the sun goes down that
variation in color may affect the
system
so we need to think more ways
efficient more robust than this
it works and we go to just one of
artificial intelligence techniques
most popular that there is today that is
the deep learning that is based
in artificial networks neural networks
artificial and basically what there is
What to do is take several examples of
trees that are far away for example and
show them to the vehicle so that
garment
know that this state tree so
You can either follow your course or example of
that the tree is close and different
except then with this
we developed a system where the drone
learn these concepts and can execute
your autonomous flight without problems
this is an example with the drone taken to
hand where what they can see is that
he is saying that he will move fast
you have to today to right after I go
fast because the tree is far away when
the tree when it approaches the tree
supposed slow asks him to go slow and
go to the right when the moon
is already very close
then again he asks him to go
quick detects that the schooner near by
which he asked to slow down what
that I’m going to do right this is
running in real time is the system
artificial we use
then finally what is the result
we have our great one that takes off
we can see the images that are
getting with your camera on board
we can locate it
using this information you are
images and we start walking
so here he is asking you to go
to face because there are no trees what
which says here as soon as it detects
very close view and therefore has
to come then brown begins the
action begins to move to the right
until it no longer detects that there is a tree and
he asks again that they begin to arrive
forward again and again to be
same in every moment
the system is detecting seal bonus to
who say it is very close and has
that fly to the right they can see
where is the trajectory of the ball
again move forward in
when you find this tree move to the
right you can see that even
it maintained certain disturbances
again this flight is completely
autonomous
then to finish all this
work they have seen there could not be
Of course it was possible without the
enthusiasm of young people who are
working with us with those who
collaborated that they study masters and
bachelor’s degree and the list of fact is
long but specifically they went back to
work of all these guys for
which I’m grateful because there
certain apprehension of flying a drone and see
crash but the satisfaction of
see him make decisions in a way
autonomous
based on what I am also very
great thank you very much
[Applause]
how
[Music]
or
plus
”
Be First to Comment