Player is loading...
ecoRobotix with Steve Tanner
Steve Tanner, ecoRobotix
Thursday, April 19, 2018 · 1:35 p.m. · 17m 15s
Presentation ecoRobotix with Steve Tanner CEO and Co-founder
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
okay so if it's thank you for the intensive instruction
00:00:04
so i'm just see also to tell you that i was born in a uniform uh
00:00:09
my father was a far more on the uh he was gringos will be it
00:00:13
and when i was young i spend the uh quite a lot of time helping uh in weeding manual weeding with uh
00:00:19
'cause manual to uh it's quite a frustrating welcome sometimes frustration is at the start
00:00:24
of a innovation and this is why uh when i was in junior high
00:00:28
i came back to the idea of a improving this um this
00:00:33
task of reading uh uh to make it more um
00:00:37
uh of course uh uh it at the time it was not a man anymore it was
00:00:42
comical but to make it more uh i mean i don't know that for me
00:00:45
so what i call the chicks we are based in a um i like
00:00:49
this picture because it makes you think that uh it's always funny uh
00:00:54
whatever it is not always the case but um and we're active since two thousand fourteen we'll fifteen people
00:01:01
and we are active in position weeding with uh using active star intentions and i
00:01:05
will uh especially a focus on the spot a a i a definite
00:01:10
so this is the the way we we we it's too they eighty percent of the of the area with with big machines
00:01:18
this is our ah kind of how are we there was no uh intelligence to a to a selected applications
00:01:25
so the uh the intelligence now today easy in the money could we have developed a selective medical
00:01:31
which can uh destroy a some families of uh of uh plans and that's others
00:01:37
um but it's not um in fact that there are a lot of drawbacks uh in the solution
00:01:43
the point is that we make a and targeted application this mean
00:01:46
there is a lot of waste it's also quite the gusty
00:01:51
uh there are a lot of particular is you don't so what'll cop i knew
00:01:55
what we all all we have the problem of of a chemical weeding
00:01:59
and the also it's damaged the cop what we don't know is usually there is a five and
00:02:04
ten percent of uh you loss due to the chemicals because they are not directly a selective
00:02:11
and uh of course of course i couldn't decide is the is very complex and
00:02:15
posted to develop its three hundred million to put on the market the new
00:02:19
but it could this means that's the uh efficient and would make use off
00:02:23
for big crops and but for crops they don't have a good product
00:02:28
uh under there is an increased pressure to to ban uh this body cues
00:02:34
and uh another problem is that's the nature or is uh that has not wait either
00:02:39
to develop a resistance like a um a back then yeah with antibiotics
00:02:44
and now we have a a m. b. c. the resistant weeds and we were on the on the planet which is a huge problem
00:02:52
so the market offered chemical waiting it's about twenty five billion per year uh in the work
00:02:58
and we can divide it in into segment there's the high added
00:03:01
value crop segment like vegetables the medium i devalue crops
00:03:06
like a potential beats and the low id value group like
00:03:09
uh uh corn a week's a rice and so on
00:03:14
so of course the higher you you go up the uh more cost to your weeding will be
00:03:19
and this is why we interest first in the in in the upper part of the part where the farmers have more money to we
00:03:27
so also should was to develop the nautilus machine with a split screen
00:03:33
technology is means application of every side only on the weight
00:03:37
so the machine must have the ability to detect the way this is why it as division invasions and uh
00:03:44
we target the sent yet accuracy to really uh put as low as possible um quantity of every site
00:03:51
so this machine can save up to ninety five percent of chemicals or a row at the same time we
00:03:56
just got because every side is the most important part and the cost a lot of reading today
00:04:03
uh we wanted a a light weight machine simple safe affordable and uh that's can
00:04:10
also wrote simpler of people uh uh a critical uh application of any chemistry
00:04:17
this means we can use for example ascetic i see that as a every
00:04:21
side uh if we know how to apply it only on the weeds
00:04:25
uh because it's a little bit on the side so we we need to to put it on the right place
00:04:30
but uh when we are speaking of uh uh spraying
00:04:35
it can also be a uh like a freak is uh told
00:04:39
before it can also be with a organic a product
00:04:45
um
00:04:47
and uh we can reduce the um the point of n. b. c. the resistant we the problem
00:04:51
because we of puts a stronger those is and the the weeds cannot develop resistance is
00:04:58
so i have a small you know to show you the the machine operation it's output will work for both
00:05:07
never works ideals
00:05:22
so this is an old the a video of for a two thousand sixteen
00:05:26
um where we see the the the principle of uh operations all the
00:05:31
robot is is fully i'm older missing energy because it's all powered
00:05:36
and the there are these two arms below that's willa move uh
00:05:40
the uh the nozzle to the right place and people's it's
00:05:44
that's very small amount so we we make us battering uh impulse expose praying
00:05:50
and the camera the camera i become our ease used for navigation but of course
00:05:54
also for a wee bit action plan that action and the then we
00:05:58
have a a computer uh on board that classifies that make the consideration of
00:06:03
the of of the plants everything is control with a with a nap
00:06:08
uh and the machine is fully autonomous once you have switch it on
00:06:12
configured you can go back home machine will work uh would they
00:06:16
uh without any human intervention so it's like the the
00:06:20
the all the little was more more machine
00:06:24
and it's also a easy to uh to deploy with this that off track
00:06:27
or we don't need a special equipment to uh to use this machine
00:06:35
so although how is um in three main topics the first is uh
00:06:40
a i was uh the planning realtime technology we have our own
00:06:45
image data bank and we reach currently eighty five to
00:06:48
ninety percent uh of plant recognition rate uh in the most advanced got that we have this is should be
00:06:56
we also have uh the centre to accuracies battery in this means the um
00:07:02
a a combination of the movement of the robot with the with the on the robotic arm
00:07:08
uh i with all the problems that there is in a in a in the rain and uh we have
00:07:13
these autonomous also rubber platform with a result of investigation we haven't educated yes sentiment the level of accuracy
00:07:20
uh and uh we also autumn synergies so this means that the the
00:07:24
former does not need to recharge the battery of the machine
00:07:27
of course he needs to be filled the tank the tank as a a capacity of one day or one to two days of abortion
00:07:34
this is dan is more than the whereabouts we have a nap we have several hours uh to perform uh that
00:07:40
analysis uh and to exchange data with other systems and we
00:07:43
have also reported for a configuration of the machine
00:07:47
for so that the user can a computer for example the there's all of the workings on in a more efficient way than a moment up
00:07:54
i will not go into detail uh about the robots uh how director we have developed everything
00:07:59
from scratch and we have forty miles from the start the the cost of the machine
00:08:03
um but what i will do tell room give somebody there's some about the the the didn't part
00:08:10
so we have uh in fact the image
00:08:13
the processing pipeline with uh first
00:08:17
of course camera a control image acquisition acquisition which is very important
00:08:23
and then we have to buy plants one which is a dedicated to
00:08:26
a guidance navigation of the machine and yellow we detection and classification
00:08:32
so about the uh navigation we have a refined or by plan b. c. this means not
00:08:38
going that's that finds the rules on the robot will fall was uh not uh
00:08:44
a crash the the core of course if the robot words on the on the crappy can
00:08:48
damage so by this way we ensure that the robot always follow the the trucks
00:08:54
um
00:08:57
we have to keep the precision the sentimental level president is me this means we have uh we use weasel automatically
00:09:03
so we have a points on the means that we track and we can people a set of
00:09:08
a centimetre accuracy a wider roads moving even if they're awfully pages or out of a phenomena
00:09:16
uh and the what's the positioning we have all integrates a g.
00:09:19
p. s. wooden colour management the undeserved retreat into a common
00:09:24
to make a an estimation of the of the uh the movement to perform
00:09:28
to uh to keep the accuracy of the machine on the on the three
00:09:33
about with addiction pipeline we uh initially we had
00:09:37
a a standard classifier uh with features action
00:09:41
uh um but then we switched to a deepening so now we have a in fact
00:09:47
we still happen segmentation but then we go directly to uh to or network
00:09:52
and uh we have some post processing it um is that to improve the the results
00:09:57
and then we feed the robot to come with the the position of the weeds
00:10:03
the problem of uh uh the biggest problem in agriculture is the the conditions we all outdoor
00:10:10
we have a e. nation conditions that we don't want must or
00:10:15
and we didn't want to master from the start with his wanted an open machine with an up and comer
00:10:20
but the reader consequence is that we can have a very difference uh
00:10:25
images colour for example sorry shades the shade of the robot
00:10:30
uh uh and the we see the difference uh between uh a standoff classifier
00:10:36
with handcrafted feature on the deep learning one with the standard classifier um
00:10:42
we asked that a stack to uh to have one factor of one five it's very hard to go
00:10:48
uh to go uh to be better than this because a standard classifier you divide it is so huge that
00:10:54
it's it's really hard to uh to be able to extract features in all conditions but with the planning we we
00:10:59
reach a much better results and eighty five percent of
00:11:03
precision uh this means that it's acceptable for weeding
00:11:09
uh we have a invented is on um on um and it yeah uh teach to perform
00:11:16
uh and we of course will time because we don't have a it's not a cloud are connected the machine or
00:11:23
it's of course it's connected to the car but the processing is performed really on on the on site
00:11:28
in real time and we stick to we can reach the the the one second for one
00:11:33
image which is uh what we need in fact every second we we've begun image
00:11:38
of the four classification we take five image possible for medication but one for classification
00:11:44
um this is a new um subject um robotic an
00:11:49
actress or the the off some few competitors
00:11:53
the most advances the believer in the us but the user is unattractive machine
00:11:58
we are so far the the only ones to um to uh be soon uh of the market
00:12:05
level reason that with machine using uh i think citations they are a lot of research machine
00:12:12
um but we see that it's really uh it's the beginning of adventure
00:12:17
there is room for other companies
00:12:20
so the market for us this is the first market is uh uh is really these high added value crops
00:12:26
in europe this means all the vegetables not all vegetables all suited for machine but uh the majority these
00:12:34
this is a very fragmented markets with a very specific uh news in terms of an decide on or machine uh
00:12:41
is a highly flexible so it respond very well to a small our ass but complex
00:12:47
a fragmented the uh i wrapped body control what we could put a quarter
00:12:54
the road map uh this year are we have twelve uh were but
00:12:59
uh we will have fifteen or what integration for countries uh uh
00:13:04
they will be begin their work next week so because it is the beginning of the winning season
00:13:10
uh then we will set up the production uh and still have the sale
00:13:15
and the or market lounge is planned for a end of the year with
00:13:18
um they are a lot of eye doctor for doing me though
00:13:22
because the farmers they have some time to to buy equipment in winter so we will do the the market lunch at this moment
00:13:30
uh and then we will expand to um to idaho uh countries in europe of course with and is not
00:13:36
is a is a very small market uh we need to go global to be uh
00:13:41
but it's economically viable
00:13:44
um so we ah about fifteen person uh to from those um
00:13:50
uh and uh we have a border fibres or uh in uh in the back and we also have a lot of collaborations
00:13:59
also with the yep and with the purpose of uh of it
00:14:03
um and the um so this is also very important to have a good um
00:14:08
good the environment in terms of a research in terms of agony a active changes
00:14:14
and we have the privilege changes and to have all these uh partly to
00:14:19
so what we have lots of lots of ah thank you for the inside
00:14:23
of adjusting tool police in what you have learned i think we
00:14:27
for most of the top of this is an adventure where we're from first adventure
00:14:33
there are a lot of uh was done on the the goal is not to make uh the same hole
00:14:38
to make a new house every time so the goal is to buy it it's or id very only
00:14:44
it use that to to go to try to sell and i didn't know but nobody will
00:14:47
buy and it's very important to uh to pass the market to ask the customers
00:14:53
uh is it good timing i think if we would have launched this this uh project the
00:14:57
ten years before it would not have been successful because the the technology was not right
00:15:04
uh uh are you ready if oh you much make it by any impact or by money i
00:15:08
think money is not a good motivation you when you when that last you need something
00:15:13
higher than money to uh to uh to become an adventure i think change the world is a good uh
00:15:19
easy with evaluation because it will allow to to perform an apple it maybe a a little bit um
00:15:26
boasting as they are today well but if if the i think the startups worth
00:15:31
are the ones a breeding we added value and id value means improving something
00:15:38
there are a lot of things to improve so uh this is uh i think
00:15:42
my opinion that motivation to have two launches pop up is improving something
00:15:47
and if you have a good idea money will come it's a consequence is that it's not the calls um
00:15:54
are you ready for a screen door on martin martin and i think it's tough to visit more map on the screen it
00:16:00
can last a lot of a lot of of yours and you must be ready you know in the long run
00:16:07
doing education i think can convincing carry something in common value when that convince the big
00:16:14
playoff the first time you reconvenes a small players on them uh take maturity
00:16:21
invest in people
00:16:23
that's what we had before uh look for help from specialist because
00:16:29
we the abilities are most of the time outside on we need need to look for them
00:16:36
not try to do everything yourself but uh look the for the good people
00:16:41
transparency in good communication uh a sale is as important as technology uh if if
00:16:48
we are coming from the research part we are more prone to technology but
00:16:53
so little stool solid top sales it becomes more important people
00:16:57
and the and the interest or is the custom okay
00:17:01
and the it takes always more time than planned maybe you with this you will uh
00:17:06
imagine than three years we do it but maybe we need six or nine years
Recommended talks
Discoverying Life patterns
Fausto Giunchiglia, from University of Trento, Italy
June 8, 2015 · 4:07 p.m.
278 views
Keynote: How Design Really Improve Events
Christof Breidenich, Designer - Scientist - Lecturer - Performer
June 22, 2015 · 9:12 a.m.
103 views