FOOD SECURITY OUTLOOK

RIPE FOR THE PICKING

The Netherlands is famed for the efficiency of its
fruit- and vegetable-growing greenhouses, but
these operations rely on people to pick the produce.
“Humans are still better than robots, but there is a lot
of effort going into automatic harvesting,” says Eldert
van Henten, an agricultural engineer at Wageningen
University in the Netherlands, who is working on a
sweet-pepper harvester. The challenge is to quickly
and precisely identify the pepper and avoid cutting
the main stem of the plant. The key lies in fast,
precise software. “We are performing deep learning
with the machine so it can interpret all the data from
a colour camera fast,” says van Henten. “We even
feed data from regular street scenes into the neural
network to better train it.”
In the United Kingdom, Green has developed

THE FUTURE OF
a strawberry harvester that he says can pick the
fruit faster than humans. It relies on stereoscopic
vision with RGB cameras to capture depth, but
it is its powerful algorithms that allow it to pick a
strawberry every two seconds. People can pick 15

AGRICULTURE
to 20 a minute, Green estimates. “Our partners at
the National Physical Laboratory worked on the
problem for two years, but had a brainstorm one
day and finally cracked it,” says Green, adding that
the solution is too commercially sensitive to share.
He thinks that supervised groups of robots can
step into the shoes of strawberry pickers in around
A technological revolution in farming led by five years. Harper Adams University is considering
advances in robotics and sensing technologies setting up a spin-off company to commercialize the
technology. The big hurdle to commercialization,
looks set to disrupt modern practice. however, is that food producers demand robots that
can pick all kinds of vegetables, says van Henten.
BY ANTHONY KING The variety of shapes, sizes and colours of tomatoes,
for instance, makes picking them a tough challenge,

O
although there is already a robot available to remove
ver the centuries, as farmers have adopted more technology in their pur- unwanted leaves from the plants.
JAN WINDSZUS

suit of greater yields, the belief that ‘bigger is better’ has come to domi- Another key place to look for efficiencies is timing.
nate farming, rendering small-scale operations impractical. But advances Picking too early is wasteful because you miss out
in robotics and sensing technologies are threatening to disrupt today’s on growth, but picking too late slashes weeks off the
agribusiness model. “There is the potential for intelligent robots to change the eco- storage time. Precision-farming engineer Manuela
nomic model of farming so that it becomes feasible to be a small producer again,” Zude-Sasse at the Leibniz Institute for Agricultural
says robotics engineer George Kantor at Carnegie Mellon University in Pittsburgh, Engineering and Bioeconomy in Potsdam, Germany,
Pennsylvania. is attaching sensors to apples to detect their size, and
Twenty-first century robotics and sensing technologies have the potential to solve levels of the pigments chlorophyll and anthocyanin.
problems as old as farming itself. “I believe, by moving to a robotic agricultural system, The data are fed into an algorithm to calculate
we can make crop production significantly more efficient and more sustainable,” says developmental stage, and, when the time is ripe for
Simon Blackmore, an engineer at Harper Adams University in Newport, UK. In green- picking, growers are alerted by smartphone.
houses devoted to fruit and vegetable production, engineers are exploring automation So far, Zude-Sasse has put sensors on pears,
as a way to reduce costs and boost quality. Devices to monitor vegetable growth, as well citrus fruits, peaches, bananas and apples
as robotic pickers, are currently being tested. For livestock farmers, sensing technologies (pictured). She is set to start field trials later this year
can help to manage the health and welfare of their animals. And work is underway to in a commercial tomato greenhouse and an apple
improve monitoring and maintenance of soil quality, and to eliminate pests and disease orchard. She is also developing a smartphone app
without resorting to indiscriminate use of agrichemicals. for cherry growers. The app will use photographs of
Although some of these technologies are already available, most are at the research cherries taken by growers to calculate growth rate
stage in labs and spin-off companies. “Big-machinery manufacturers are not putting and a quality score.
their money into manufacturing agricultural robots because it goes against their cur- Growing fresh fruit and vegetables is all about
rent business models,” says Blackmore. Researchers such as Blackmore and Kantor are keeping the quality high while minimizing costs.
part of a growing body of scientists with plans to revolutionize agricultural practice. If “If you can schedule harvest to optimum fruit
they succeed, they’ll change how we produce food forever. “We can use technology to development, then you can reap an economic
double food production,” says Richard Green, agricultural engineer at Harper Adams. benefit and a quality one,” says Zude-Sasse.

2 7 A P R I L 2 0 1 7 | VO L 5 4 4 | NAT U R E | S 2 1
©
2
0
1
7
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
,
p
a
r
t
o
f
S
p
r
i
n
g
e
r
N
a
t
u
r
e
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 OUTLOOK FOOD SECURITY

ELIMINATING ENEMIES

AFIMILK LTD
The Food and Agriculture Organization of the United Nations estimates that
20–40% of global crop yields are lost each year to pests and diseases, despite the
application of around two-million tonnes of pesticide. Intelligent devices, such
as robots and drones, could allow farmers to slash agrichemical use by spotting
crop enemies earlier to allow precise chemical application or pest removal, for
example. “The market is demanding foods with less herbicide and pesticide,
and with greater quality,” says Red Whittaker, a robotics engineer at Carnegie
Mellon who designed and patented an automated guidance system for tractors
in 1997. “That challenge can be met by robots.”
“We predict drones, mounted with RGB or multispectral cameras, will take
off every morning before the farmer gets up, and identify where within the field
there is a pest or a problem,” says Green. As well as visible light, these cameras
would be able to collect data from the invisible parts of the electromagnetic
spectrum that could allow farmers to pinpoint a fungal disease, for example,
before it becomes established. Scientists from Carnegie Mellon have begun to
test the theory in sorghum (Sorghum bicolor), a staple in many parts of Africa
and a potential biofuel crop in the United States.
Agribotix, an agriculture data-analysis company in Boulder, Colorado, supplies
drones and software that use near-infrared images to map patches of unhealthy
vegetation in large fields. Images can also reveal potential causes, such as pests or
problems with irrigation. The company processes drone data from crop fields in
ANIMAL TRACKERS
more than 50 countries. It is now using machine learning to train its systems to Smart collars — a bit like the wearable devices designed
differentiate between crops and weeds, and hopes to have this capability ready for to track human health and fitness — have been used
the 2017 growing season. “We will be able to ping growers with an alert saying you to monitor cows in Scotland since 2010. Developed by
have weeds growing in your field, here and here,” says crop scientist Jason Barton, Glasgow start-up Silent Herdsman, the collar monitors
an executive at Agribotix. fertility by tracking activity — cows move around more
Modern technology that can autonomously eliminate pests and target agrichemi- when they are fertile — and uses this to alert farmers to
cals better will reduce collateral damage to wildlife, lower resistance and cut costs. when a cow is ready to mate, sending a message to his or
“We are working with a pesticide company keen to apply from the air using a drone,” her laptop or smartphone. The collars (pictured), which
says Green. Rather than spraying a whole field, the pesticide could be delivered to are now being developed by Israeli dairy-farm-technology
the right spot in the quantity needed, he says. The potential reductions in pesticide company Afimilk after they acquired Silent Herdsman last
use are impressive. According to researchers at the University of Sydney’s Australian year, also detect early signs of illness by monitoring the
Centre for Field Robotics, targeted spraying of vegetables used 0.1% of the volume average time each cow spends eating and ruminating, and
of herbicide used in conventional blanket spraying. Their prototype robot is called warning the farmer via a smartphone if either declines.
RIPPA (Robot for Intelligent Perception and Precision Application) and shoots “We are now looking at more subtle behavioural
weeds with a directed micro-dose of liquid. Scientists at Harper Adams are going changes and how they might be related to animal health,
even further, testing a robot that does away with chemicals altogether by blasting such as lameness or acidosis,” says Richard Dewhurst,
weeds close to crops with a laser. “Cameras identify the growing point of the weed an animal nutritionist at Scotland’s Rural College (SRUC)
and our laser, which is no more than a concentrated heat source, heats it up to in Edinburgh, who is involved in research to expand
95 °C, so the weed either dies or goes dormant,” says Blackmore. the capabilities of the collar. Scientists are developing
algorithms to interrogate data collected by the collars.
In a separate project, Dewhurst is analysing levels of

CROP ANGEL LTD

exhaled ketones and sulfides in cow breath to reveal
underfeeding and tissue breakdown or excess protein
in their diet. “We have used selected-ionflow-tube
mass spectrometry, but there are commercial sensors
available,” says Dewhurst.
Cameras are also improving the detection of threats
to cow health. The inflammatory condition mastitis —
often the result of a bacterial infection — is one of the
biggest costs to the dairy industry, causing declines in
milk production or even death. Thermal-imaging cameras
installed in cow sheds can spot hot, inflamed udders,
allowing animals to be treated early.
Carol-Anne Duthie, an animal scientist at SRUC, is using
3D cameras to film cattle at water troughs to estimate the
carcass grade (an assessment of the quality of a culled
cow) and animal weight. These criteria determine the price
Drones with precision producers are paid. Knowing the optimum time to sell
sprayers (insert) apply would maximize profit and provide abattoirs with more-
agrochemicals only consistent animals. “This has knock on effects in terms
where they are needed. of overall efficiency of the entire supply chain, reducing

S 2 2 | NAT U R E | VO L 5 4 4 | 2 7 A P R I L 2 0 1 7
©
2
0
1
7
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
,
p
a
r
t
o
f
S
p
r
i
n
g
e
r
N
a
t
u
r
e
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 FOOD SECURITY OUTLOOK

SILICON SOIL SAVIOURS

A. RUCKELSHAUSEN/UNIV. APPLIED SCIENCES OSNABRÜCK

the animals which are out of specification reaching the

abattoir,” Duthie explains.
And researchers in Belgium have developed a camera The richest resource for arable farmers is soil. But large harvesters damage and
system to monitor broiler chickens in sheds. Three compact soil, and overuse of agrichemicals such as nitrogen fertilizer are bad
cameras continually track the movements of thousands of for both the environment and a farmer’s bottom line. Robotics and autonomous
individual birds to spot problems quickly. “Analysing the machines could help.
behaviour of broilers can give an early warning for over Data from drones are being used for smarter application of nitrogen fertilizer.
90% of problems,” says bioengineer Daniel Berckmans “Healthy vegetation reflects more near-infrared light than unhealthy vegetation,”
at the University of Leuven. The behaviour-monitoring explains Barton. The ratio of red to near-infrared bands on a multispectral image
system is being sold by Fancom, a livestock-husbandry can be used to estimate chlorophyll concentration and, therefore, to map biomass
firm in Panningen, the Netherlands. The Leuven and see where interventions such as fertilization are needed after weather or pest
researchers have also launched a cough monitor to flag damage, for example. When French agricultural technology company Airinov,
respiratory problems in pigs, through a spin-off company which offers this type of drone survey, partnered with a French farming coopera-
called SoundTalks. This can give a warning 12 days earlier tive, they found that over a period of 3 years, in 627 fields of oilseed rape (Brassica
than farmers or vets would normally be able to detect napus), farmers used on average 34 kilograms less nitrogen fertilizer per hectare
a problem, says Berckmans. The microphone, which than they would without the survey data. This saved on average €107 (US$115)
is positioned above animals in their pen, identifies sick per hectare per year.
individuals so that treatment can be targeted. “The idea Bonirob (pictured) — a car-sized robot originally developed by a team of
was to reduce the use of antibiotics,” says Berckmans. scientists including those at Osnabrück University of Applied Sciences in Ger-
Berckmans is now working on downsizing a stress many — can measure other indicators of soil quality using various sensors and
monitor designed for people so that it will attach to a modules, including a moisture sensor and a penetrometer, which is used to assess
cow’s ear tag. “The more you stress an animal, the less soil compaction. According to Arno Ruckelshausen, an agricultural technolo-
energy is available from food for growth,” he says. The gist at Osnabrück, Bonirob can take a sample of soil, liquidize it and analyse it to
monitor takes 200 physiological measurements a second, precisely map in real time characteristics such as pH and phosphorous levels. The
alerting farmers through a smartphone when there is a University of Sydney’s smaller RIPPA robot can also detect soil characteristics
problem. that affect crop production, by measuring soil conductivity.
Soil mapping opens the door to sowing different crop varieties in one field to
better match shifting soil properties such as water
availability. “You could differentially seed a field, for
example, planting deep-rooting barley or wheat vari-
eties in more sandy parts,” says Maurice Moloney,
chief executive of the Global Institute for Food Secu-
rity in Saskatoon, Canada. Growing multiple crops
together could also lead to smarter use of agrichemi-
cals. “Nature is strongly against monoculture, which
is one reason we have to use massive amounts of her-
bicide and pesticides,” says van Henten. “It is about
making the best use of resources.”
Mixed sowing would challenge an accepted pillar
of agricultural wisdom: that economies of scale and
the bulkiness of farm machinery mean vast fields of a
single crop is the most-efficient way to farm, and the
bigger the machine, the more-efficient the process.
Some of the heaviest harvesters weigh 60 tonnes, cost
more than a top-end sports car and leave a trail of
soil compaction in their wake that can last for years.
But if there is no need for the farmer to drive the
machine, then one large vehicle that covers as much
area as possible is no longer needed. “As soon as you
remove the human component, size is irrelevant,”
says van Henten. Small, autonomous robots make
mixed planting feasible and would not crush the soil.
In April, researchers at Harpers Adams began a
proof-of-concept experiment with a hectare of bar-
ley. “We plan to grow and harvest the entire crop
from start to finish with no humans entering the
field,” says Green. The experiment will use existing
machinery, such as tractors, that have been made
autonomous, rather than new robots, but their goal
is to use the software developed during this trial
as the brains of purpose-built robots in the future.
“Robots can facilitate a new way of doing agricul-
ture,” says van Henten. Many of these disruptive
technologies may not be ready for the prime time
just yet, but the revolution is coming. ■

2 7 A P R I L 2 0 1 7 | VO L 5 4 4 | NAT U R E | S 2 3
©
2
0
1
7
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
,
p
a
r
t
o
f
S
p
r
i
n
g
e
r
N
a
t
u
r
e
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.