Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
Hello, and welcome to this podcast
0:02
from the BBC World Service. Please
0:04
let us know what you think and tell
0:06
other people about us on social
0:08
media. Podcasts from the BBC
0:10
World Service are supported by advertising.
0:13
Hello, everybody. It's Tuesday the third of
0:15
January twenty twenty three. Happy
0:17
New Year. think we can say that with
0:19
Garik Mitchell here and Bill Thompson to bring you
0:21
digital planet for this week from the
0:23
BBC. Are you alright, Bill?
0:25
I
0:26
guess. You're indeed there. Yes.
0:28
Good. Yeah. To be back. The year has begun.
0:31
Yeah. I
0:33
hope it will be friendly.
0:36
Yes. Let us hope. Well, you mean between
0:38
you and I, it's always friendly. I was
0:40
thinking more broadly, actually. was thinking
0:43
Given our global audience, I hope it's friendly
0:45
for
0:45
everyone. Life beyond the little parochial
0:48
hearings on between your NII, Yes.
0:50
Of course. And well, just to get
0:53
us in that New Year spirit, I don't know
0:55
if we can posit it as a New Year's tech
0:57
resolution, but whether people
0:59
should be backing up their
1:01
Twitter profiles. We've spoke about it
1:04
at the end of last
1:05
year, didn't we towards the end of last year?
1:07
And
1:08
We did. Yep. It's triggered a
1:10
little bit of response. So
1:12
It may be a new solution for some people. I
1:14
don't know. But inspired
1:17
by your piece, Bill, where you did
1:19
say a bit about how we can archive
1:21
profile and
1:22
momentum. I just talked about the fact
1:24
that it's clear that like any
1:26
other online service Twitter is a little
1:28
bit fragile. And if you value what you've been
1:30
saying there, it's and a good idea now
1:33
to to just take copy of it. It should
1:35
be good good practice anyway, good digital
1:37
hygiene to keep your own copy of anything
1:39
you post anywhere that you see because of interest,
1:42
because you can't rely on any third
1:45
party provider to keep it, or to care
1:47
about it the way you care about
1:48
it, that's really that way. I get it. Well,
1:51
Francis Day says one reason
1:53
to do it might be you could marvel at why you
1:55
got caught up in things which now seem irrelevant.
1:58
But there is oh, yes. I mean, I think I think
2:00
if you did it if you crawled through my
2:03
Twitter archive, you climbed a whole load
2:04
stuff. Did I really waste
2:06
my time thinking about this? Yes.
2:09
Yes. Thinking about
2:11
it, then tweeting about it. And then thinking
2:13
about the things that people replied yet, we've
2:15
all done it. Peter Smurten
2:17
kind of related to this really says, well, my
2:19
thinking is that you ever place anything on
2:21
Twitter that is worth preserving, then you're
2:23
doing it all wrong. He's added a smiley face
2:25
there. Jessica, it's rid of
2:27
I
2:27
believe. But but but that's actually a
2:30
very interesting point because it
2:33
it definitely was a sense that Turtle was a
2:35
place where you would put stuff that was ephemeral
2:37
as it works intended just to go away. But
2:40
then when threads were created and
2:42
used, people started using it to put up
2:44
really quite interesting collections
2:47
of thoughts. That possibly worth
2:49
preserving. And so people found a
2:51
way to use this this medium that
2:53
that was sort of unstructured. In
2:55
a more structured way. And as a result, over
2:57
the years, I think there are
2:59
repositories of wisdom embedded
3:02
in people's treatifeds that may be worth preserving.
3:05
But absolutely right. That was not the original intention.
3:07
Mhmm. It was definitely supposed to just go away,
3:09
you know. Well, thinking about my early tweets from
3:11
Lindsay about here where I was drinking cups of coffee.
3:14
You know, what sort of a nice cake I had and
3:16
things like that. There were not exactly significant
3:18
insights into my life.
3:21
Alright. Folks, by the way, we have an agricultural
3:23
tech special coming up in just a few moments when we
3:25
get into the radio program. In case people were
3:27
impatient to get into that which we have two
3:29
more listener suggestions, then we'll get into electric
3:32
tractors and smart traps
3:34
and other stuff like that, Bill. You
3:36
go for it. Oh, let's go for it with Sonya
3:38
Livingston. He says, I think there's
3:40
a difference between conversations and
3:43
sharing information that has value at the time
3:45
and wanting to keep the records for
3:47
posterity. Yeah. Bit
3:49
of a
3:49
difference. Yeah. Yeah.
3:51
Now what was a good insight from Sonya? Yeah.
3:53
Thanks, And Caroline Talbot, just
3:55
to finish this little section for us, says,
3:57
a lot of innovative commerce, serious
4:00
professionals used Twitter to share
4:02
and get feedback on their work with fellow
4:04
professionals and the public, for instance, scientists
4:06
and medics studying COVID, probably
4:08
important for them in many ways to retain
4:11
their archives. And, yep, I
4:13
know a lot of scientists and they will say
4:15
that they find I think out of all the networks,
4:17
they I don't know if it's changed now, but they
4:19
will say that Twitter is the one for them in
4:21
terms of sort of peer
4:23
discussion of things going on in
4:25
science in their field
4:26
outside. Maybe the the more formal channels I
4:28
should have. Anyway, Yeah.
4:31
So my last point and I know we need to move into the
4:33
program now is the architecture that gives
4:35
you is a starting point when it's not enough
4:37
in itself. You need to use other tools to
4:39
get the lists of who your contacts are
4:41
and any links and stuff like that. So don't just
4:43
see you can download your archive from Twitter and then
4:45
you're
4:45
done. You do need to put a little more effort
4:47
into this. Oh, alright. I'm glad you
4:49
added that because I must admit I'm one of those people who
4:51
just wants to get it all done and I just downloaded
4:53
the archive and
4:54
said, right, done it brilliant. You can kill it off
4:56
now. I don't care. But it's not as simple that says
4:58
build on some on digital
4:59
panels. Absolutely not.
5:00
Okay. Absolutely not. Thank you, Bill.
5:02
Alright. Well, let's jump into the rodeo
5:04
program. This is some cultural
5:07
technology coming your way as it's handed
5:09
on the radio this week. Hello,
5:11
everybody. Happy New Year. I'm Gareth Mitchell,
5:13
and this is digital planet. Today,
5:15
we're talking agricultural technology.
5:17
Yep. It's a special edition with
5:19
an electric tractor, a
5:21
smart insect trap, and a
5:23
robot fruit picker. And I'll be harvesting
5:26
comment and analysis from Bill Thompson
5:28
today as it goes. Hello, Bill? Hello
5:30
there, Gareth. Nice to be here again.
5:32
Likewise, to be speaking to you.
5:34
Okay. Well, let's start with that electric
5:36
tractor. It's all part of the drive,
5:38
of course, to low emission agriculture.
5:41
And out in the field, the machine isn't
5:43
just harvesting crops. It's harvesting
5:46
data, thanks to an array of
5:48
onboard sensors to monitor everything
5:50
from plant health to pests.
5:52
The company, monarch tractors,
5:54
launched their autonomous electric smart
5:56
tractor just before Christmas, CEO
5:59
and cofounder Praveen Penmetze
6:01
says it's the first vehicle of its
6:03
kind to be commercially available.
6:06
Yeah. So our tractor is quite special in
6:08
the sense. It's not just an electric tractor
6:10
gear. It's also a driver optional.
6:13
And more importantly, it's also smart.
6:15
It tells the farmers what is going on in the
6:17
farm
6:17
today, so it gives them alerts. And
6:20
also the data that it collects means that
6:22
the farmers can use that to go back
6:24
in time and see what happened that
6:26
led to, you know, the the harvest
6:28
that they have had and use
6:30
that data. To
6:33
not only save money, but also make
6:35
more money by telling their story to the
6:37
customers. So the tractor does all those
6:39
three
6:39
things. So as an electric vehicle
6:41
I suppose my husband's question is we know that tractors
6:44
need to be out in the field
6:46
for long periods at a time.
6:48
And it's all very energy intensive
6:50
work. How long does it last for on
6:52
a single charge?
6:53
Yeah. So our tractor, even
6:56
if you're doing some light activities, we'll
6:58
ask for, like, fourteen hours and sometimes
7:00
even more. But if even if you're doing,
7:02
like, really heavy activities, Garrett,
7:04
like, say, tilling or other
7:07
land management activities. Even
7:09
then, we get the tractor to, like, five to
7:11
six hours of usage. So that
7:13
means that a power can, you know,
7:15
finish the shift. And while they're having
7:17
lunch or taking a
7:18
break, they get to recharge the
7:20
tractor or swap the battery out
7:23
in the middle of the field and then
7:25
get the tracker up and running just like you
7:27
do with your
7:27
power tool. No. I like that analogy.
7:29
The tractor is also driverless then. Yes.
7:32
And one of the things that we learned to have
7:34
this bigger infinity girth when we
7:36
first built our first electric back
7:38
in twenty seventeen. We took it to an Indian
7:40
village. And the farmer looked at
7:42
us and said, this is great.
7:44
Who's going to drive it? And I
7:46
was, like, surprised because I was, like, wait a
7:48
second. I thought you're going to drive it. And
7:50
the farmer said, no. There's only two people in the
7:52
whole village who can drive a tractor
7:54
because it's not just about driving the
7:56
tractor. It's about doing operations. Those
7:59
operations are now fully automated. It, which
8:01
means it's driver optional from that standpoint.
8:03
And that makes a big difference for followers
8:05
around the world. Right. And that's the key point.
8:07
Driver optional. So a driver can be in
8:09
the cab if they want to be,
8:11
but it has a driverless option. Absolutely.
8:15
And the reason for that is how many times have you
8:17
not seen a tractor on a road. Right?
8:20
Yeah. I do want to drive
8:22
around it. I seem to
8:24
be reassured. Yes. Now
8:27
you described this as a a smart tractor and
8:29
all the time on this break when we have people telling us,
8:31
oh, we got smart this stat and everything smart
8:33
these days. But let's unpack what that
8:36
means lots of senses on
8:37
there, lots of processing. Like, I
8:39
remember a conversation where I was talking to
8:41
this farmer about all this
8:43
fantastic data that they're gonna
8:44
get. And the farmer looked at me and said, if
8:47
you're gonna give me data, I'm gonna charge
8:49
you. And
8:50
he said, what I want to know
8:53
is what action needs to be taken
8:55
just giving data or telling
8:57
somebody what's happening with the
8:59
tractor or what's happening on the farm
9:01
is not
9:01
enough. We have to make it actionable
9:04
insight is what we call them. No.
9:06
Sure. I know I know these days farming is a
9:08
data rich industry. And
9:11
So you have the tractor and the
9:13
data aspect presumably must come
9:15
from sensors on the
9:17
tractor. So tell me about the
9:19
data gathering and then how it can be used.
9:21
We have a number of cameras on the
9:23
tractor. We have both three d
9:25
cameras as well as normal monocular
9:28
cameras. The three d cameras are
9:30
very interesting in the sense they create
9:32
this point cloud data. Very
9:35
much similar to what a radar or a
9:37
light artist where you see points
9:39
you go. But the advantage of three d
9:41
camera is is we get that point
9:43
cloud data along with
9:45
the color image. So now
9:47
our computers can process both those
9:50
together and really
9:52
identify objects, identify
9:55
the path, and all of
9:57
that data allows us not only to
9:59
control the tractor, also we
10:01
can give that data to the farmer
10:03
with actionable insights by saying, hey,
10:05
the color on this leaf is is
10:08
off. You need to send somebody over to
10:10
check it because it might need a pesticide or it
10:12
might need some additional fertilizer, etcetera.
10:14
We make this whole data
10:17
open to anybody, which
10:19
means that a university
10:21
student can write an application
10:23
that will help a farmer around the world
10:26
on coming up with some
10:28
insights on making farming better.
10:30
But we overlay that Garrett with
10:32
additional sensor data including
10:34
GPS. So for example,
10:36
we can answer the questions of
10:38
not only who, what, where,
10:40
but also how. So we can
10:42
collect all this into a structured data
10:44
lake. And we also make that data lake available to
10:46
other third party researchers who can
10:48
provide additional value to the
10:49
farmer. Where do you
10:50
see all this going in the future and
10:53
technology and farming generally. Yeah.
10:55
So we see a world where sustainability
10:57
in terms of how we scale our
10:59
food ecosystem is now top of mind
11:01
for everybody. So
11:03
that's an important aspect. Farmers are
11:05
now struggling not only with weather,
11:08
with
11:08
the shortage of labor on their farms.
11:10
They're also having to meet sustainability
11:13
demands from state and federal
11:15
agencies around the world. But the
11:17
beauty of it, Garrett, is and something that I'm
11:19
very excited about is for the
11:21
first time, I think, we as consumers,
11:23
you and I can actually see how our
11:25
food was grown and have a direct
11:27
connection to the
11:29
on farm operations that
11:31
went into putting the food on water table.
11:34
Right now, we know more about the
11:36
delivery person who tell you what our food than the
11:38
farmer who grew the food. Right?
11:40
So we want to fix that. And I think the
11:42
future is going to be amazing
11:44
where we have these very
11:46
customized, very localized nutritious
11:50
food available at scale for the world's
11:52
growing population, but grown in
11:54
a sustainable manner both from a
11:56
side and planet standpoint as
11:58
well. That's Praveen Penmetze.
12:00
So Bill Thompson, one thing that struck
12:02
me in that interview with the amount
12:04
of discussion about data, like the lovely
12:07
point crowd data from the sensors on
12:09
the camera, you know, three d flip
12:11
mapping of where it's
12:12
been. Yes. I mean, they're
12:14
they're collecting so much data that that could
12:16
be of enormous value just just in
12:18
terms of sort of reshaping the way people think
12:20
about, you know, the way food is grown.
12:22
There there's always a question about, you know, how you would
12:24
get access to that data. Whether people
12:26
would actually be interested in it? I know that
12:28
a lot of organizations have tried to do things like
12:31
trade disputes. Clothes and stuff like that.
12:33
And frankly, people,
12:35
consumers don't seem that interested. They sort of
12:37
like to know it's there, but they don't actually go and
12:39
investigate it. But this does
12:41
offer certainly authorities and regulators and
12:43
sort of the people who want to know that
12:45
food has been properly grown, access
12:47
to, in fact, superior information what's
12:49
been going on on the farm. I
12:52
would hope that whatever data is collected is
12:54
also available to the farmers themselves
12:56
using other ways it's like I'm really
12:58
focused invested in open data and how
13:00
farmers might find ways to exploit
13:02
or use that data that haven't been
13:04
thought of by the manufacturers as well.
13:06
Yes. Because I I think Praveen was talking there about
13:09
actionable insights, for instance, you
13:11
know, from the the
13:11
data, so one would hope. But
13:14
there's always this thing about in such
13:16
once you put these complex automation
13:18
systems in place and let's assume they can
13:20
be maintained and sustained going
13:22
forward. Where control
13:24
lies. There's some
13:26
controversy in the United States, particularly
13:28
about the tractor manufacturer
13:30
John Deere, claiming that the software that
13:32
runs its tractors, it belongs to
13:34
it is its copyright. And therefore,
13:36
farmers can't, in any sense, sort
13:38
of, reconfigure it. And
13:40
that's quite a controversial point of view because it's
13:42
kind of asking who owns the machine
13:45
and who owns the software that runs it? And
13:47
I think what we need to be looking for
13:49
is it a situation within which
13:51
these complex machines and the fact
13:53
are seen as a partnership between the
13:55
the owner of the machine and
13:58
the creator. To work collaboratively,
14:00
but it's not a question of locking
14:02
farmers out of access to the data they
14:04
did collected or indeed locking them
14:06
out of the ability to maintain
14:08
technology. They absolutely rely on to
14:10
harvest their crops. Well, I'm sorry,
14:12
Bill, is it just to make it familiar
14:14
to listeners outside agriculture. For instance, not
14:16
being able to change the battery in your smartphone.
14:18
Well, that sort of example, but actually, it's more not
14:21
being able to use third party printer
14:23
cartridges because your printer identifies
14:25
a cartridge that's being purchased from
14:27
the manufacturer and won't run with any other
14:29
one. We see this constant, if
14:31
like, tension between the
14:33
people who make himself and quite reasonably want to
14:35
make money out today and want to work with it,
14:37
and the people who own the devices about
14:39
who can maintain, who can service, who can
14:41
change it. And as we move forward,
14:43
it becomes more sophisticated. I think it's really
14:45
important we have this discussion and debate
14:47
in the open so that people
14:49
don't end up signing up for a service they don't
14:51
want to use. And and I'm sure that, you know,
14:53
Praveen and others would want to have something which the
14:55
farmers find a positive experience in their
14:57
lives and not something which is difficult for
14:59
them. Alright, Bill. Thank you.
15:01
So with the help of a driver
15:03
optional electric tractor,
15:05
The crops are planted and growing. Our
15:08
next problem is controlling pests,
15:10
and that's where we go now
15:12
with a smart Insectract, it
15:14
gives the farmer a real time pest
15:16
survey by capturing insects
15:18
and identifying them through deep
15:20
learning. What's a likely facial recognition
15:22
in your phone's photo library, you want to
15:24
get the pests, not the pollinators, so
15:26
the trap helps to better target
15:29
pesticides rather than spraying
15:31
chemicals or whatever, everything that we ate.
15:33
Pest destroy up to forty percent of
15:35
global crops and cost two hundred and
15:37
twenty billion US dollars worth
15:39
of losses worldwide annually
15:41
according to the Food and Agriculture Organization of
15:43
the United Nations, the FAA.
15:45
Well, matter Stefancik is the
15:47
chief executive officer of
15:49
Trapview, the Slovenian company
15:51
behind this trap. He wants to move away from
15:53
what many farmers are doing at
15:55
the moment.
15:56
What really happens is, I know that
15:58
simplicity always wins.
16:00
And at the moment, the most simple,
16:02
the most common way of
16:04
of dealing with best insects is
16:07
basically eliminate. However, that
16:09
causes quite some issues. Right?
16:11
You you run into resistance if
16:13
you're over pray. The residual
16:15
levels in in the food we are
16:17
eating is relatively high.
16:19
And that has been one of the
16:21
issues that is being tackled with with
16:23
new technology. Right? So how
16:25
can we better understand
16:27
what's happening with the pest insects?
16:30
How can we predict how they
16:32
will develop. So we can use
16:34
software products. We can use maybe more
16:36
biological crop
16:37
protection. Which brings us to
16:39
your solution. It's called trap you
16:42
say, tell me a
16:42
bit about how it works. Collecting the data
16:44
is a big portion of it. And for that,
16:46
we developed automated traps, which
16:48
are catching insects with
16:50
chard. Taking picture of what was
16:52
caught and sending this data
16:54
over the cellular network to the cloud
16:57
to really get rid of
16:59
this reliance on the
17:01
manual work to collect the data.
17:03
The next step is then what you what you
17:05
do then with this row data,
17:07
right, with pictures. And that's how
17:10
you do the processing with image
17:12
recognition. That's how you do the machine
17:14
learning forecasting of what will
17:16
happen with the past population in the
17:18
in in the
17:18
future. So this is automated in the
17:21
sense that you have pheromones that attracts
17:23
the And then you get I mean,
17:25
you'll tell me, like, hundreds or thousands
17:27
of insects over a period of time. And
17:29
then through image recognition across
17:32
the cloud, you can identify which
17:34
insects and which and then get a
17:36
snapshot as to which ones you need to
17:38
worry about and so on. So that's it's like an
17:40
insect survey that's automated.
17:42
How about the image recognition
17:43
though? How do you discriminate between the different
17:46
insects? The real technical
17:49
question we are dealing with is
17:51
how many of the males of
17:53
that species we are looking at
17:56
have been caught because
17:58
that's how you build the information about
18:00
what's really the best pressure and
18:02
how it differs from different locations.
18:06
And the extraction of this
18:08
information from the pictures themselves
18:10
comes in two steps. The first one is
18:13
image recognition. Here
18:15
we have really strong position because
18:17
we built the biggest database of
18:19
pictures of insects in the
18:21
world that that really allows us to use frameworks
18:23
or techniques like
18:25
like deep learning very efficiently.
18:27
And most people might
18:30
know this from how the
18:32
cars recognize the traffic
18:34
signs or how it brings them closer to
18:36
the autonomous driving. But
18:38
it's basically the same technology in
18:40
the background that we use to
18:43
identify the targeted insects
18:45
from the picture, which could be
18:47
really obfuscated, you know, with
18:49
leaves, with
18:50
debris, also with other insects that
18:53
have been caught One issue
18:55
here though is you're using mobile
18:57
data to, you know, get the images from
18:59
the trap into the cloud where they can
19:02
be processed. But farms by
19:04
their nature are often in remote places. And I'm
19:06
just wondering around the world is a
19:08
lack of mobile
19:09
connectivity, a deal breaker for how this
19:11
can all work.
19:12
Sometimes you have pure
19:15
mobile connections like Europe
19:17
is very well covered.
19:20
And of course, another part is is done in some
19:22
sort of a hybrid mode, you know,
19:24
where you would have cellular network
19:26
in an area, but those
19:29
sell towers with connect to our satellite
19:31
to send the data. So
19:34
connectivity, yes, in some areas, it could
19:36
be a a challenge. But
19:38
that is something, you
19:39
know, that that has been much more
19:41
of an issue in the past than it
19:43
is now. If you're gathering effectively
19:46
like a real time survey
19:48
of insect populations across
19:50
swathes of
19:51
land. How does that information help the
19:54
farmer? Consider this case. Know,
19:56
you have coddling mouth
19:58
on apples. And you
20:00
want to use some, let's
20:02
say, natural predator, some
20:05
other insect or or
20:07
species that is eating eggs
20:09
of the cuddling moth. You
20:12
cannot use it when the best has already developed,
20:14
when you already have Jarvis, when it
20:16
already went inside the Apple.
20:19
So it's very important to use this kind of
20:21
real time data and real time
20:23
or and localized data.
20:26
With accurate prediction of how the best will
20:28
develop. So you can use the right
20:30
product at the right time. In some
20:33
cases, It is eggs or early
20:35
stage larvae. Yeah. In some cases, it
20:37
could also be adults, but more
20:39
and more of a softer are
20:42
really targeting early stages of development
20:44
and that is before the
20:46
damage has happened. So it
20:48
is really very much about
20:51
prevention of the problems to happen and
20:53
reacting on time.
20:55
So moving yourself away from
20:58
being, you know, reacting to
21:00
the problems to to
21:02
preventing them.
21:03
That's Matteo Stefanciets. So
21:07
Bill Thompson, What about this aspect
21:09
of using machine learning here
21:11
in identifying the insects as part of
21:13
that proactive approach that we
21:15
just heard there from Cartag.
21:17
At
21:17
last, a positive use of facial
21:20
recognition. You know, insects don't have
21:22
data rights and therefore, it's fine to
21:24
identify them in your trap. I have to think
21:26
given all the controversy we have
21:28
over ML being used in wider
21:30
society, it's nice to see what clearly
21:32
has positive uses. And
21:34
I think that building this database
21:36
and being able to sort of possibly identify
21:38
the insights is going to be challenging
21:41
and trap you would be working for some
21:43
years. But it's absolutely a solvable
21:45
problem. But you, of course, put your finger on it
21:47
with your question about the the WiFi and the
21:49
connectivity and things like that, which is all
21:51
of these systems don't exist in
21:54
isolation. It's not that, oh, look, I've got this box
21:56
and it's got a fair amount of crap and it's
21:58
identifying the insects. And therefore, it's going to be
22:00
great. How does that information get
22:02
into the system to be used and
22:04
exploited. And you can't think of any of
22:06
these things in isolation. And I
22:08
like the fact that the the trap are
22:10
thinking about that broader context. Right? It's not
22:12
just we've got this cool little box, you can put it
22:14
on the tree. It's why is the overall system
22:16
that will be used to positively
22:19
help
22:19
farmers? Alright. And I need to spend a
22:21
lot of time just checking for bugs
22:23
in the software. And, Jake, there for the tech
22:25
people, but thank you very much here all
22:27
night. Okay. Moving on. Thanks, Bill, by the way.
22:29
So we've sown the
22:31
crop with our electric tractor. We
22:34
protected it with that smart insect trap
22:36
with its machine Now it's time to
22:38
harvest, but with a growing
22:40
shortage of workers waiting to do the
22:42
poorly paid back breaking
22:44
work, how can robots help?
22:46
Machines already do all kinds of harvesting,
22:48
but soft fruits needing a
22:50
delicate touch while they remain a challenge.
22:53
In Portugal though, a robot is now harvesting
22:56
raspberries. It's been developed in
22:58
Britain and the tasty fruits of its
23:00
labor could soon be on our
23:02
supermarket shelves potentially. Here it
23:04
is in action on a farm
23:06
in Odamira in the Southwest
23:08
of Portugal. And if you're expecting
23:10
it to be some kind of robotic hand
23:12
with fingers, then think
23:14
again.
23:16
Hi. We are in Portugal where
23:18
the robot is inside one of
23:20
the lines with brush pens at
23:23
Summerberry Farm and
23:26
Right now, the mobile base is moving
23:29
to kill it detect a barrier. A
23:31
barrier has been detected. The
23:33
arm is now moved to underneath it.
23:36
It has inflated
23:39
a plastic membrane and
23:41
after it pulls the berry and
23:44
then goes to deposit inside
23:46
of a
23:47
panicked. That's
23:49
Andre Martin. He's a field test
23:51
engineer with fieldwork Robotics.
23:54
Andre's colleague, academic founder,
23:56
and chief science officer, Martin
23:58
Stollen, is the brains behind
24:00
this robot. If you see the machine in
24:02
the field is maybe about the size of
24:04
AAA large kind of
24:06
American refrigerator on wheels, If
24:09
you can imagine that, it
24:11
has four robot
24:13
arms on it. And
24:15
it's basically is able to
24:17
navigate between the rows of raspberry
24:20
plants and then it stops
24:22
and picks out targets and
24:25
carefully picks one
24:27
rasp at a time and puts it into
24:29
pellets.
24:29
The mobile base is
24:32
moving till it finds another barrier
24:34
to
24:34
pick. It has found one.
24:36
The arm is gonna position
24:39
itself underneath it.
24:41
There are cameras at the end of the arm to
24:43
make sure the berries inside
24:45
the
24:45
cup. It's
24:48
moving
24:48
up. It's inflating the
24:51
membrane. It's pulling
24:55
and now it's going for another
24:57
berry.
24:57
And so the robot goes inflating
25:00
and deflating its cup like hand
25:02
as it works its way through the
25:04
crop. Martin Stolen says it's taken a whole load
25:06
of trial and error to refine
25:09
the design. Being a technology
25:11
that's in heavy development, we've gone through
25:13
quite a few different iterations of
25:15
QuickBooks. And
25:15
actually, the latest version of gripper
25:18
looks a bit more like a
25:20
cup that approaches the
25:22
rasp free from below and it then
25:24
inflates a membrane
25:26
around the raspy to to
25:28
gently squeeze it and be able to pluck it off
25:30
the plant. Right. So
25:31
this is the latest iteration then.
25:33
So you've tried the finger pincer
25:35
movement found maybe it's not quite
25:37
delicate enough. So now you have the membrane
25:39
almost like a is it like a bubble or
25:41
most it? You know when people blow bubbles
25:43
with gum. Isn't that kind of bubble, if you
25:45
like? Yeah. Just I guess That's
25:47
the rubbery. I guess
25:49
if you can imagine Well,
25:52
kind more like a doughnut, to be honest.
25:54
As you
25:54
imagine the raspberry entering into a deflated
25:57
doughnut and then the doughnut inflating around
25:59
it to even see the raspberry in the
26:01
first place. There is a vision system
26:03
on this robot, isn't there? So perhaps tell
26:05
me a bit about that. However, even knows to
26:07
pick a raspberry and not just pick a
26:09
leaf off the plant or something. Correct. So the the the
26:11
first challenge you have is to to
26:13
try to to pick out the
26:15
Rasprey from the surroundings. And
26:18
for that, we we
26:20
use cameras, of course, and we use different
26:22
types of cameras, cameras
26:24
that look in at colors. So
26:27
the color of the raspberries is a good
26:29
indication of how or where it
26:31
is compared to the leaves. But
26:33
also cameras are able to detect the
26:35
three d structure of the bush itself and the
26:37
raspy in relationship to it. And that enables
26:39
us also to pinpoint the
26:42
location of any potential targets,
26:44
any potential raspberries. And then we
26:46
have machine learning
26:48
algorithms that enable us to say
26:50
something about the maturity of that
26:52
recipe. If it's ripe for picking or not, if
26:54
it has the caesars, etcetera. How did it
26:56
take to develop the robot? Fumor Robotics
26:58
was spun out of the University of Plymouth when
27:01
I was I was a lecturer there and I was
27:03
back in two thousand sixteen.
27:06
So we've been working on it heavily since then.
27:09
We've of course been growing, started out
27:11
as a small typical spinout
27:13
company. We have some support from a
27:15
company called front TRIP that helped
27:17
us kind of get going. And
27:19
then we've gradually grown to about
27:21
fifteen employees now. And
27:23
we are at the stage where we now have
27:25
a robot in the field continually
27:27
in Portugal that is
27:30
partially used as developing a
27:32
new technology and partially demonstrating
27:34
that we can generate revenue. How
27:36
does the robot compare to a human
27:38
picker? There's four
27:39
robot arms on each platform, and
27:41
each robot arm you would say it's a
27:43
bit moving a bit slower than than a human
27:46
arm at the moment, but you in
27:48
sense recuperate that by having
27:50
more arms by operating much
27:52
more hours, you can operate almost twenty
27:54
four hours a day. And
27:56
also the key point for us is
27:58
that you need to look at the cost of the
28:01
technology. So if you compare
28:03
the technology with
28:06
robotics, that we put into the field,
28:08
the requirement for those robots
28:10
are quite different from the robots that you would
28:12
have in, say, a factory. In
28:14
a factory, you might need
28:16
very fast speeds. You
28:18
might need a sub millimeter accuracy.
28:22
Which are things we don't necessarily need
28:24
in robots that's going to pick raspberries.
28:26
So also then the costs come down
28:28
considerably. And where we compare
28:30
ourselves to the human
28:32
harvesters is really on the cost
28:34
per barrier or the cost per
28:36
kilogram, which also is
28:38
how
28:38
a lot of the business model around the harvesting robot
28:40
is centered on providing them as
28:42
a service. Yeah. I was wondering about the
28:45
economics of it. So
28:47
you'd say that there it is viable then
28:49
because these are not cheap, but you lease them
28:51
out, and then that gives a a good deal
28:53
to farmers or at least that's your
28:55
plan? Certainly. And we're already demonstrating
28:58
for the first time this year, we were able
29:00
to harvest dry sprees that
29:02
passed the quality control of one of
29:05
the largest raspberries growers in the U.
29:07
K. Which also has operations
29:09
in Portugal. And
29:11
demonstrate that we could sell these berries.
29:13
They're they're picked to spec. And
29:15
we are now gradually
29:18
rolling out more robots in the field and each
29:20
robot is gradually picking more
29:22
and more kilograms per
29:24
hour and we're quite confident that we will
29:26
be able to have
29:28
profitable operations not
29:31
in a very distant future. Yeah.
29:33
There you go. That's Martin Stolland.
29:35
We also heard that
29:38
from Andre Martin's.
29:40
Bill Thompson, I suppose there's a bigger question here
29:42
though, isn't it about where we want the robots to
29:44
stack in and maybe where we want to keep with the
29:46
human pickers or
29:47
growers? Or tractor they make it.
29:49
I think there always is,
29:51
Gareth, absolutely, that
29:54
with any of these attempts to
29:57
automate processes which have been traditionally carried
29:59
out by humans, we have to ask
30:01
ourselves, is it a job we want people to do?
30:03
Are we happy have it handed over to
30:05
machines. And what does that
30:07
do to local economy? There's absolutely
30:09
an issue that fruit picking isn't a
30:11
particularly rewarding job. It can be physically
30:13
very intensive and quite
30:15
damaging. But if it's the only job people
30:17
have, maybe you want it
30:19
to be still available to them. And there's
30:21
always that balance. It's
30:23
clear from just the description of the
30:25
technology that we're moving to the point where
30:27
these robots or similar robots will be able to
30:29
do this job in particular
30:32
contexts. And again, it comes back to this
30:34
point again. They have to be embedded in
30:36
quite complex systems. They need to be
30:38
maintained, you know, apart from things like
30:40
power supplies and charging. They need to
30:42
be repaired. They need to be
30:44
supported. And so there are only going to be a limited number of environments
30:46
within which they are absolutely
30:48
viable. That will be countries with a
30:50
very well developed industrial infrastructure build
30:52
to provide that support. The real
30:54
danger would be to try to offer them to
30:57
places where they would break down very quickly
30:59
and end up being useless. I I
31:01
think that looking at the development of
31:03
this picking technology, has that been
31:05
remarkable? And just the innovation in terms
31:07
of using, as we've described, the membranes and
31:09
other things to do the picking, The
31:11
fact that the robot hands are three d
31:13
printers and effort can be adapted to
31:15
to different produce over time. All of
31:17
that, it's a really brilliant engineering
31:20
project. But always the question
31:22
is, what's this engineering offer to
31:24
the people like the farmers who actually
31:26
need to get their job
31:27
done? Is it good for Alright. Thank
31:29
you very much. He's Bill. I'm Gareth.
31:31
The producers are Allan
31:33
Beech and also Andy Litterover. It's
31:35
the studio manager. It's Steve Greenwood. See
31:37
you soon. Bye bye. Alright. Well, let
31:39
let's carry on and we started the pod
31:42
intro with some New Year's Tech
31:44
Resolutions. And, of course, that was
31:46
all very positive and
31:48
forward looking. So I just think, Bill,
31:50
we may as well just go back to a bit of
31:52
tech winching which is the strength. I think we
31:54
nearly killed up before
31:54
Christmas, but there were just a few good ones, so I'm afraid I'm
31:57
gonna use them today. You you said
31:59
you were gonna come back to this. This is this is
32:01
turning into the marble calendar of
32:03
the nose. Period of our existence,
32:05
isn't it? Cara, please
32:07
share some tech nineties with us. I'm
32:10
fascinated and intrigued to know what
32:12
they might
32:12
be. Oh, thanks for reading that thing I put on the
32:14
script there, that note. Thanks, Bill. Right
32:16
on script there. So alright.
32:18
Then since you insist, Bill, we can have a bit
32:20
of Richard GaN. This is by the way,
32:23
folks. Just a quick reminder if you can
32:25
bear it that I having problems with
32:27
my broadband before Christmas and
32:29
as well as the funny moment
32:31
really when I was
32:33
on the line to the call center and saying, you know, I think
32:35
I've got problems and they said, yes, we do
32:37
have issues with the broadband in your area. And
32:39
there was, like, cable
32:41
that was cut, and I could literally see it
32:43
hanging loose just outside my door. So that's the
32:45
context I just said to you, dear,
32:47
listeners, just if there's anything else that's
32:49
been getting you down then we're nice
32:51
friendly people share it with
32:53
us, and at least we'll listen. So
32:55
but Richard G hadn't not so
32:57
much a tech solution actually, which is wonderful
32:59
and kind of waiting into my
33:01
broken cable kind of issue. He says, well,
33:03
you need a fiber spice
33:06
kit for your next birthday. It's a good
33:08
suggestion, Richard. But my
33:09
provider, I think I might need it.
33:12
Share it with my neighbors on the street
33:14
as well. Yeah. just get in touch
33:16
with Chris Condor at Baum
33:18
brought down from the rural North, which which just
33:20
goes around lain fiber in the north of
33:21
England. They they can help you out
33:24
with that. Oh, alright. I'll I might I
33:26
I may have to do that. And
33:28
I because I wonder what it's like to slice
33:30
a piece of fiber. I've never done that if
33:32
it's sounds quite tricky to
33:34
me. I bet you need to be quite
33:37
good, maybe not. Anyway, let's Can
33:39
I can I Here's what I'm right. They came from me over
33:41
the festive season, which is
33:44
like why oh, why is it so
33:46
hard to install Internet of
33:48
Things devices? as an
33:50
example, my mother has
33:52
a smart voice a voice assistant as
33:54
I think I might have slipped into the
33:56
end of year last week.
33:59
And she loves this voice assistant, and
34:01
so there's a bit of it where you can switch
34:03
the living room lights on and
34:05
off. And I've I've installed
34:07
some smart when she first
34:09
had this voice assistant and everything was working okay. But then
34:11
her good old fiber broadband
34:13
provider then went and changed her
34:15
router so now of course nothing
34:17
in her house works anymore. So I've had to had to spend a load
34:19
of time when I could have been celebrating the festive
34:22
season reconfiguring a load of stuff in her
34:24
house. But it's just putting in
34:26
smart plugs where you have the
34:28
voice assistant, which needs to
34:30
interface with a native app
34:32
that goes with the particular
34:34
make of smart plug.
34:36
And then In order to get the
34:38
smart plug working, the
34:40
broadband router because it's on two frequencies. What
34:42
is it? Two point four and five gigs
34:44
or something? Five. Yeah.
34:46
Okay. But you have to split the
34:48
band. Otherwise, the smart plug might
34:50
try and substitute into the five gig one, which
34:52
doesn't work for it. So then you have to switch off the
34:54
five gig and then just use the two point four. But then I
34:56
found when I did that because I was trying to configure for my
34:58
laptop. Of course, my laptop no longer
35:00
wanted to communicate with
35:02
the router.
35:04
So then I went through my mobile broadband. And about halfway through this, I thought,
35:06
why am I doing this? Why is it so hard? It
35:08
shows surely now with the technology we
35:10
have in the world at the moment.
35:13
We should be able to just buy an off the shelf
35:15
smart plug, plug it in, and then set
35:17
up some command on the voice assistant, and
35:19
it should just work. Why is it so
35:21
painful? Don't need to answer. I just want to get that off my chest. It's my
35:23
other tech win beginning
35:26
twenty twenty
35:26
three. I can
35:27
move on
35:28
if you will. Okay. And very briefly. I mean, the
35:31
answer is there is no effective standardization
35:33
yet. Other people have been working on it used
35:35
to be Zigbee, for example, report. So
35:37
there isn't there isn't a way of doing it. is still broken
35:39
in many ways. We haven't solved that
35:41
problem. And partly it's because the device manufacturers
35:43
want to innovate really
35:46
fast. And so they want the devices to
35:48
do things that aren't necessarily supported by current standards and so they implement something on top
35:50
of it. And so we've got this constant
35:52
tension between innovation and standardization. With
35:56
the router changing problem, in the past when
35:59
that happened with me on a smaller scale,
36:01
I got the new router
36:03
and changed its identified such
36:05
as an ID and password the same as the old
36:08
router. And then things just clicked the
36:10
new router and couldn't realize that it'd
36:12
been changed. Because the yeah. As long as as long as the WiFi standard
36:14
itself hasn't changed and it's supported by the
36:16
new router, it should work. As when the
36:18
device just looks for a particular identifier,
36:20
it says please can I
36:22
connect? Here's here's some credentials, and the
36:24
zone has got the right
36:25
credentials. It should work. Yeah. It should. It
36:27
should a
36:27
lot of work in that center. I
36:29
I get it. And that would
36:31
include changing the SSID on the
36:34
router as well. Oh, I yeah. Indeed.
36:36
I I did wonder about that. It all sounded a
36:38
bit scary to me, and so I just got
36:40
grumpy for a while. And then my work around
36:42
was just what did I do? Yeah. I just
36:44
switched off the five
36:46
gig to wrap band
36:48
on the on the router. And then the it's sure
36:50
enough the devices went for the two point four
36:52
and I expect it to all back up on the gain into the
36:54
normal configuration.
36:56
And it did work. But, yeah, that's a good point,
36:58
isn't it? Just change the credentials. Thanks, Bill. Yeah. I think
37:01
that's worth a
37:01
try. It's Well, I have to try.
37:04
Even, sir, it shouldn't need to be that
37:06
complicated. I will say it needs to be that
37:08
complicated. Right? I'll give you that.
37:10
Thank
37:10
you. But enough of me, let's have some of the Marty
37:12
says my fiber optics Internet
37:14
randomly disconnects around every forty
37:18
eight hours. And I need to reset the
37:20
modem. So I've decided to reset it every night when I don't need the Internet for
37:22
about a minute. My best
37:26
service provider continues Maddy is
37:28
flaking my apartment and its
37:30
vicinity. Sometimes if I move my phone by
37:32
half an inch, the
37:34
call drops. And if my phone says five g around here, that
37:36
means no cell internet.
37:38
And Monday goes on a bit of a
37:40
gripe about his university's IT
37:42
department that's a bit lengthy, so I
37:44
won't go into it now, but I hope you got it
37:46
fixed with the university there. A few
37:48
winters there. And
37:49
also, the Broken IT and New University
37:51
is a preparation for a
37:53
world work where the IT will be before broken. So so look at the
37:55
product as part of the training they're
37:57
offering. Yeah. Although, Marty's
37:59
actually a teacher, I think, is a professor
38:01
at the university. Okay.
38:03
In which case. Sorry. But no but no please
38:05
don't apologize because I think he can flip that and then
38:07
make that as part of his kind of lesson, his
38:10
life lesson as well as his
38:12
technology
38:12
lesson. Says very grateful students in what I think is mechanical
38:14
engineering. I used to do that
38:16
at a certain remaining unnamed
38:18
university in a certain On
38:22
a certain course explain that the really rubbish IT
38:24
systems were their preparation for
38:27
when they got killed.
38:29
They'll they'll be thanking you in later life,
38:31
I'm sure, or hopefully
38:34
just improving them if they go into that line of
38:36
business. Let's do
38:38
one from Chris, if we have times, as I had a similar situation
38:40
with my Internet provider years ago when I was
38:42
using a third party router. Oh,
38:44
yes. The good old third party router thing,
38:46
Chris. Yeah.
38:48
Thanks for I don't know where Chris continues. I finally found out that the phone
38:50
company sent a query down the
38:52
line to see if users were using a phone
38:54
company provided router, i. E. An
38:56
official one. If they didn't get
38:58
the right response, they throttle or stop
39:00
the service. Apparently, finishes
39:02
Chris. They could do this because
39:04
they provided their router for free
39:07
But my issue with that is it didn't didn't
39:09
have any WiFi. So this must have been
39:11
some time ago. Yeah. I'm
39:14
not surprised I bet a load of providers
39:16
do that. They'll just send some diagnostic when they're down the line
39:18
just to check that you're using their
39:19
kit. And because it it might be some
39:21
liability or indemnity things as
39:24
well. Yeah. It
39:26
might be that they claim they can't provide
39:28
service unless you're using their technology, but the
39:30
sort of people who are competent enough to install their
39:32
own Richard generally wouldn't want the sort
39:35
of service that provided because they were already beyond that point.
39:37
Mhmm. It's a it's it's a tricky
39:39
one as as we as we
39:41
know. Yeah. So What's
39:43
a control you should have over the technologies in your
39:45
home. Yeah. A lot. I think Yeah.
39:48
I I had a feeling you would think that. Yeah.
39:50
Mhmm. Alright. Simon
39:52
says, bear in mind the system the run up to
39:54
Christmas that this would have been written. Simon says,
39:56
does the season to be having a flaky
39:59
home network? Simon says I discovered that plugging
40:01
odd stuff into main power sockets, which have
40:03
a power line broadband network running through them,
40:06
can make the connection go bad. In
40:08
this case, was a smart
40:10
home power monitoring the
40:12
plug, but I'm still suspicious of the
40:14
Christmas tree lights. But if you're
40:16
suspicious of them, my guess, disconnect
40:18
them and see if it's fix the
40:19
problem. God, I don't I don't present a tech program but nothing do I with amazing
40:21
advice like that site over there building. So I
40:23
mean Christmas tree
40:26
lights always used to be the problem when they were incandescent
40:28
lights. Now they're LEDs, I suspect
40:30
they're less of a problem. Yeah.
40:33
I I can see why that Yes.
40:36
I suspect you'd be right. Especially if you get the battery
40:38
powered ones. Mhmm. Anyway
40:41
Well, yes. Well, actually so that it's partly about the power, and then it's
40:44
also just about the interference electron
40:46
interference. Yeah. Yeah. Because because a a
40:48
lot of tiny little lights
40:50
flickering away.
40:52
There's a there's a lot there's a lot of electromagnetic interference
40:54
coming from those. Yeah. I think now
40:56
it's as an RF person. It's not kind of the
40:58
kind of thing I'd want near any
41:00
a high frequency kit. Put it that
41:01
way. I will
41:02
always think who you as an RF person. Thank
41:04
you very much. I'll get the t shirt.
41:06
So I do a pretty person, Garrett.
41:10
But there we are. We both have we're both in our
41:12
clans, aren't we then? We we have our
41:14
tribes. We
41:17
do one that one of the most things seem to be going bizarrely
41:19
well, really, which just on the road, we could just take
41:21
a last one from David before we call it
41:23
today and limp out the studio. David
41:26
says I moved house last working day
41:28
before the first lockdown in March twenty
41:31
twenty. They sent an engineer out
41:33
the next only come as far as the green box,
41:35
but told me that should be enough. When
41:37
that person didn't finish the whole job, they
41:39
were then only able to send
41:41
someone else out to fix it in
41:43
June, presumably after you could then enter people's houses again. In the
41:46
meantime, David says he had to run
41:48
his phone,
41:50
burning hot using it as a
41:52
WiFi hotspot in order to keep his
41:54
job, utterly ruining ruining
41:56
the plane's battery, but not his
41:58
career. So David probably saw it as a trade
42:00
off worth doing.
42:01
Yeah. Recovery.
42:02
Yes. Yes. You need. Not bad
42:04
at all. Alright. Well, I'm I think that
42:06
might do it, mainly because I've run out of listener comments.
42:09
So without them, it's just you and I chatting away
42:11
and that might be too much for anybody to
42:12
bear. That that would and that could fill
42:15
quite a lot of time really. But let's not
42:17
inflict that on anyone
42:19
particular producer any no, that's true. I think we've got another four minutes of
42:21
studio time if you do want to fill, but I suspect we should
42:24
just wrap it up. And you're
42:25
agrees. Yep. There we are. But
42:28
I've had his voice. It's
42:30
been a life. It's been a joy. Your
42:32
technology winches have illuminated
42:35
my life. Thinking about in the air in such a
42:37
great way. Oh, you guys. Alright.
42:40
There we are. There'll be more
42:42
sincerity at the same time
42:44
next week. Here on the
42:46
digital planet podcast. Thanks for bearing with
42:48
us folks. And, yeah, see you
42:50
next week. I hope
42:52
cheers. Bye.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More