DEVELOPMENT OF THE END-EFFECTOR OF A PICKING
ROBOT FOR GREENHOUSE-GROWN TOMATOES
Y. C. Chiu, P. Y. Yang, S. Chen
ABSTRACT. The objective of this study was to develop a grip-type end-effector for use in a picking robot designed to
harvest greenhouse-grown tomatoes. The end-effector was designed with four fingers that have foam sponge pads inside to
reduce damaging the fruit when bending and gripping. By controlling solenoid activation, the fingers are bent to reduce
the opening of the tip, securing the fruit inside the end-effector. The center of the end-effector is a fruit suction device,
which assists the fixation of fruit to the inside of end-effector, enhancing the holding performance. Test results show that
the best suction performance is achieved using vacuum suction nozzles 15.0 mm in diameter and a suction force of
8.1 N/cm2. The average suction attachment success rate is 95.3% and the average picking time is 74.6 s for each fruit.
When picking the suction-attached fruit, the end-effector makes vertical inching movements over 60 mm for two cycles,
achieving an optimal fruit picking success rate. Various twisting angles are examined in this study, and the picking force
required is analyzed. The results show that before picking tomatoes, the end-effector should be rotated clockwise for 60°
followed by a counterclockwise turn of 120° until the fruit is 60° counterclockwise to the starting alignment, repeated
three times in relation to the fruit. The initial laboratory tests show promising results. Future integration with robotic
tomato-harvesting systems for actual field tests of tomato picking should increase the feasibility of developing automatic
robotic picking systems.
Keywords. Greenhouse, Automation, Harvesting, Vacuum suction.
A
utomated production is an important current
trend and future development in the field of
agriculture.
With
gradually
maturing
technologies,
employing
robotics
for
agricultural production is a significant development
direction. Research into the design, construction, and
application of agricultural production robots is also a
significant factor that enables the construction of a quality
agricultural produce production environment.
Tomatoes are an important crop with a high economic
value; they are grown widely by farmers in Taiwan, with a
total growth area of 4,762 ha. However, greenhouse-grown
tomatoes have long been harvested manually, which is a
substantial drain of manpower. Thus, developing automatic
harvesting robots for tomatoes is extremely necessary. The
design of the picking end-effector is a crucial aspect of this
development.
The research community has had a sustained interest in
designing robots for agricultural harvesting. Van Henten
et al. (2003) designed an autonomous robot for harvesting
Submitted for review in September 2012 as manuscript number IET
9913; approved for publication by the Information & Electrical
Technologies Division of ASABE in August 2013.
The authors are Yi-Chich Chiu, ASABE Member, Professor,
Department of Biomechatronic Engineering, National Ilan University,
Taiwan and Pen-Yuan Yang, Master, Department of Biomechatronic
Engineering, National Ilan University, Taiwan and Suming Chen,
ASABE Member, Professor, Department of Bio-Industrial Mechatronics
Engineering, National Taiwan University, Taiwan. Corresponding
author: Y. C. Chiu, 1, Sec. 1, Shen-Lung Rd, I-Lan 26041, Taiwan;
phone: +886-3-9357400 ext. 7804; e-mail: [email protected].
cucumbers in greenhouses. This robot comprises a six-axis
robotic arm and a collision-free motion path planning
algorithm. The average success rate for harvesting highwire grown cucumbers was 74.4%, with an average harvest
cycle time of 65.2 s/cucumber. Ling et al. (2004) used fruit
identification technologies and identified the mature
tomatoes using sensor signals. A lightweight four-fingered
triple-jointed end-effector was attached to the robotic arm.
A vacuum system was affixed to its central axis. The
extension and retraction of the vacuum suction nozzle was
coordinated with the bending of the fingers. The average
time required to harvest one tomato was approximately
3 min 37 s. Tanigaki et al. (2008) examined cherry
harvesting robots. Lasers and infrared (IR) lights were used
to precisely position the fruit to prevent collisions with
other fruit when the end-effector performs a deep cut.
Johan et al. (2008) devised various automated apple
picking robots. These robots used an industrial-grade sixaxis mechanical arm and specialized end-effector for
picking. The end-effector comprised a silicone funnel and
vacuum suction equipment. A camera was also installed to
identify, position, and assess the maturity of the fruit. The
average image processing time was 8 to 10 s. The harvest
success rate of this robot was approximately 80% for fruit
60 to 110 mm in diameter. Hayashi et al. (2010) developed
a strawberry-harvesting robot for application to an elevated
substrate culture. The successful harvesting rate of the
system was 41.3% when fruits were picked using a suction
device before cutting the peduncle, while the rate was
34.9% when fruits were picked without suction. The
execution time for the successful harvest of a single fruit,
Applied Engineering in Agriculture
Vol. 29(6): 1001-1009
© 2013 American Society of Agricultural and Biological Engineers ISSN 0883-8542 DOI 10.13031/aea.29.9913
1001
including the time taken to transfer the harvested fruit to a
tray, was 11.5 s. Furthermore, Hayashi et al. (2013)
established a structural environment suited to the operation
of this strawberry-harvesting robot, including a rolling-type
hanging bench and travelling platform.
The terminal manipulation and operation of robotic arms
and end-effectors designed to perform specific movements
is crucial for successful harvesting using robots. Customdesigned end-effectors that satisfy the need to perform
various movements using robotic arms are crucial to
successful harvesting. Appropriate end-effector designs can
enhance the operational stability and efficiency.
Considering the end-effector designed by Kondo et al.
(2010) for harvesting a string of tomatoes, the end-effector
comprised two fingers that gripped and one light sensor kit.
The upper finger set also included a cutter. During
harvesting, the end-effector approached the main stalk until
detected by the light sensor. The upper and lower fingers
then clasped the main stalk of the tomato before the upper
finger moved downward, cutting the string of tomatoes.
The harvest success rate was 50%. Applying this design in
a high-wire cultivation system should increase the success
rate. Reed et al. (2001) developed an automatic robotic
mushroom harvesting rig that employs a monochrome
camera for locating and measuring the fruit. The endeffector design includes vacuum suction and two fingers
with differently sized suction nozzles that can be rotated
and used for mushrooms of different sizes. Over 80% of the
attempts to pick mushrooms were successful. Arima et al.
(2004) developed a harvesting robot for strawberries grown
in hanging plant beds. This end-effector employs vacuum
suction, three pairs of photo interrupters, and cutters.
During harvesting, the photo interrupters determine
whether the suction head has reached the fruit. The fruit is
then sucked into the tip of the end-effector. Next, the endeffector tip rotates to guide the peduncle into the corner
with the cutter, which then cuts the fruit from the plant. The
fruit is then delivered to the tray via the transport tube.
Using machine vision to identify and locate fruit is also
essential when harvesting. Studies such as Plebe and
Grasso (2001) explored precisely locating oranges using
machine vision to facilitate manipulation by robotic arms.
Bulanon et al. (2005) developed a system for recognizing
and positioning apples in a tree. Once the size and threedimensional location of the apples are known, spatial
information is sent to the robotic arm for fruit harvesting.
These studies were referenced for this research. Ji et al.
(2012) developed an automatic recognition vision system
guided for apple harvesting robot. The tested results
showed a recognition success rate of approximately 89%
and average recognition time of 352 ms.
This study designs an end-effector system for picking
greenhouse-grown tomatoes, which can be installed on the
self-propelling automatic tomato harvesting robot
developed by Chiu et al. (2012). The end-effector was
designed as a four-finger gripper. Electromagnets are
employed to bend the fingers, which enhances the endeffector’s gripping capability. A vacuum suction device
immobilizes the fruit to isolate it from the plant before
picking.
1002
MATERIALS AND METHODS
ROBOTIC SYSTEM FOR TOMATO PICKING
This study designs an end-effector to be integrated in the
automatic tomato picking robot system. The system
includes a five-axis robotic arm, a picking end-effector,
machine vision, moving vehicle, and a control system. The
system architecture is shown in figure 1. The control
system is constructed using the graphic-based programming
language LabVIEW version 7.1.
DESIGN OF THE PICKING END-EFFECTOR
Figure 2 is a diagram of the tomato picking end-effector
design. The designed end-effector is a four-fingered gripper
comprising fingers, a spring plate, a fruit suction device,
and finger bending mechanisms. Each finger has four
sections and a total length of 138 mm. The lengths of the
four sections are 36, 32, 37, and 33 mm. All the joints
between the sections are flexible and provide cushioning
when grabbing the tomatoes. Soft foam rubber was placed
on each finger section to prevent damage to the fruit
surface. The four fingers were divided into two sets,
installed on the left and right open-close plates, and
controlled using the sideways open-close chuck at the tip of
the robotic arm. The open-close plates measured 140 ×
Figure 1. Diagram of a tomato picking robot.
Twin-axis cylinder
Finger
Suction nozzle
Foam rubber
Limit switch
Open-close chuck
Solenoid
Figure 2. The tomato picking end-effector prototype.
APPLIED ENGINEERING IN AGRICULTURE
139 mm and were designed to fit together during sideways
movements. The distance between the opened and closed
positions was 30 mm.
Figure 3 shows the external dimensions of the picking
end-effector. The distance from the base of the robotic arm
axis to the tip of the gripper fingers is 254 mm. When the
open-close plates are open, the opening at the tip of endeffector is oval-shaped, with a horizontal width of 90 mm
and vertical height of 60 mm. Because the open-close
plates move sideways at a distance of 30 mm, the opening
of the tip of the end-effector is 60 mm wide and 60 mm
high when the open-close plates are close, forming a round
opening that is 60 mm in diameter. When the solenoids pull
the spring plates on the inside of the fingers, the fingers
bend, reducing the opening of the tip of the end-effector to
a 40-mm circle. This secures the tomato within the endeffector. Because the maximum and minimum apertures of
the grippers are 90 and 40 mm, the diameter range of the
tomatoes from 40 to 90 mm is suitable to be harvested by
the developed gripper.
To pull tomatoes inside the end-effector during picking,
a twin-axis pneumatic cylinder drives the vacuum suction
nozzle to extend and retract, pulling the fruit from the plant
into the gripping area. This reduces collisions between the
end-effector and the stalk, preventing damage. Using
vacuum suction to secure fruit and prevent them from
shaking enhances the holding performance of the endeffector by enabling the fruit to remain in the gripper with
more stability. The fruit suction device includes an air
compressor, five-way two-position solenoid valve, twinaxis pneumatic cylinder, vacuum generator, and suction
nozzle. The main movements are divided into two
categories, fruit suction and fruit traction. A vacuum
suction nozzle is installed on a round platform, which is
connected to the twin-axis pneumatic cylinder. Through the
action of the twin-axis pneumatic cylinder, the suction
nozzle telescopes forward and backward, drawing the fruit
from the plant for rotational picking maneuvers, thereby
reducing damage to the plant.
The fruit suction device includes an air compressor
(MB0204, Puma, Chunghwa, Taiwan), a five-way twoposition solenoid valve (AM520-01-S, Shako, Taoyuan,
Taiwan), a twin-axis pneumatic cylinder (BD16X70,
Chanto, Taichung, Taiwan, the cylinder diameter is 16 mm,
and the travel distance is 70 mm), a vacuum generator
(EV10HS-G, EDCO, Fenton, Mo.), and a suction nozzle
(triple-fold type, Yuen Chin, New Taipei, Taiwan).
Figure 4. Finger bending mechanism.
Figure 4 shows the finger bending mechanism, which
includes a spring plate, wire, and solenoid. One spring plate
is inserted into each of the four fingers, with a wire
connecting the tip to the solenoid. When the solenoid is
magnetized, it pulls the steel wire on the inside of the
finger, drawing the tip in and forcing the spring plate to
bend. This causes the four sections of the fingers to curl
similarly to human fingers when gripping something. Limit
switches that control the finger curving extent were inserted
into the base of the fingers. When the four fingers curl
together, the opening of the tip of the end-effector shrinks,
enhancing its grip of the fruit. This study used a tubular
solenoid (SH-T2551L, Shih Shin, Changhua, Taiwan). The
nominal voltage was DC 12V, with a nominal current of
0.83 A. The magnetic force is 1.40 kgf with a 10 mm travel
distance.
PICKING ACTION FLOW
Figure 5 is the overall picking action flowchart. First, an
image is obtained using a binocular vision system to
determine whether the fruit is sufficiently mature. If yes,
the three-dimensional coordinate of each mature fruit is
calculated. The coordinate information is then delivered to
the robotic arm, which begins the picking operation. The
robotic arm is first positioned in front of the fruit; then the
open-close plate drives the fingers to open. The twin-axis
pneumatic cylinder pushes the vacuum suction nozzle
outward to grasp the tomato. To enhance the suction
success rate, the mechanical arm drives the end-effector to
perform vertical up and down inching movements to
increase the likelihood of a good fit between the suction
nozzle and the fruit surface. Once the fruit is held by the
suction, the twin-axis pneumatic cylinder retracts the
vacuum suction nozzle, pulling the fruit inside the endeffector. Then, the open-close plate closes and the solenoids
pull the wires, causing the four fingers to bend and clamp
simultaneously, thereby securing the fruit. Next, the endeffector detaches the fruit through rotation, and then places
the fruit inside the fruit basket, completing one picking
cycle.
EXPERIMENTAL METHODS
Tomato Physical Property Check
The main item picked by the end-effector in this study is
the Momotaro tomato. The major axis, minor axis, height,
and weight of several tomatoes were measured as shown in
figure 6. Fifty tomatoes in total were measured. The mean
and standard deviation of each measurement were
calculated and used as the foundation for this experiment.
Figure 3. External dimensions of the picking end-effector.
29(6): 1001-1009
1003
Image is obtained
using binocular
vision system
The robotic arm is
positioned in front
of the fruit
The vacuum
suction nozzle
stretches to grasp
the fruit
The end-effector
performs vertical
up and down
inching movements
The vacuum
suction nozzle
pulls the fruit
inside the
end-effector
The gripper clamps
and detaches the
fruit through
rotation
Places the fruit
inside the basket
Figure 5. Action flow of the picking end-effector with vacuum suction device.
Figure 6. The tomato physical property measurement items.
Test of Vacuum Suction Force on Fruit
To determine the appropriate degree of vacuum for the
fruit suction device, a triple-fold type suction nozzle 15 mm
in diameter for the suction tests was used. Five vacuum
suction powers were tested on five tomatoes of varying
size. The tested suction forces were 9.1, 8.6, 8.1, 7.7, and
7.2 N/cm2 (683.0, 645.5, 608.0, 577.9, and 540.4 Torr). The
tomatoes used in the experiment weighed 112, 148, 186,
215, and 278 g. The hanging locations of the tomato test
samples were known and fixed. The robotic arm was used
to perform the picking action. The preliminary test shows
that 70 mm travel distance of suction nozzle can pull the
fruit inside the gripper if the end-effector is moved in the
front of tomato. The suction nozzle was retracted once the
end-effector was successfully attached to the tomato. The
criterion of suction success was the attachment of the fruit
sample after the suction nozzle had been retracted for 3 s.
Each test was repeated 30 times to calculate the suction
success rate.
Suction Nozzle Diameter Tests
Different suction nozzle diameters produce varying
levels of suction power. Furthermore, because fruit is not
1004
entirely spherical, the diameter of the suction nozzle affects
the ability to create an air-tight seal between the suction
nozzle and the fruit. Previous studies have shown that
suction nozzles with a smaller diameter generate superior
air-tightness and a higher suction attachment success rate
(Ling et al., 2004). Therefore, for this study, four suction
nozzles of varying diameter were examined during the
experiment, namely, 15.0, 17.6, 18.2, and 19.5 mm. The
suction nozzles were made from rubber and tested during
the vacuum power tests explained in Section 2.4.2. The test
tomatoes weighed 112, 148, 186, 215, and 278 g. Each test
was repeated 30 times to calculate the suction success rate.
The Effect of Various Inching Movement Types on the
Suction Success Rate
Because the suction nozzle may not always attach to the
fruit properly because of the contact position, angle, or
surface undulation, the arm to perform an inching motion
was designed when the suction nozzle contacts the fruit
(vertical, sideways, and back and forth). Thus, the surface
of the fruit and the suction nozzle have a greater likelihood
of achieving an air-tight seal, increasing the suction success
rate. The factors examined included the inching motion,
inching distance, and fruit weight. Three types of inching
movements are performed, namely, up/down, left/right, and
forward/backward as shown in figure 7. The three tomatoes
examined weighed 112, 148, and 186 g. The major axis
length of each sample was 64.3, 72.4, and 81.8 mm. The
experiments conducted in this study adopted a Latin square
design; the results are shown in table 1. Each combination
was tested 30 times to calculate the suction success rate. In
table 1, “A” denotes the fruit that was 64.3 mm in diameter,
“B” denotes the 72.4 mm fruit, and “C” denotes the
81.8 mm fruit.
APPLIED ENGINEERING IN AGRICULTURE
Table 1. Latin square test of the effects of various inching types,
inching distances, and fruit sizes on the suction success rate.
Inching Type
Inching Distance
Vertical
Sideways
Back and Forth
(mm)
20
A
C
B
40
B
A
C
60
C
B
A
End-Effector Holding Force Test
Figure 8 shows a diagram of the holding force test. The
primary goal of this test was to examine the end-effector’s
ability to hold fruits of various sizes and weights. For this
test, a test ball is held by the end-effector, and the minimum
instant pull force that can pull the test ball from the grip of
the end-effector is determined. One end of the force gauge
was held by hand while the other end holds the test ball in a
net. A video camera was used to record the change of the
gauge indicator to assist in reading the data for every test
record.
The end-effector was tested in three modes with
different suction strengths: (1) gripper hold without
vacuum suction; (2) gripper hold with 8.1 N/cm2 of vacuum
suction; and (3) gripper hold with 7.2 N/cm2 of vacuum
suction. Each test combination was repeated 20 times and
assessed using analysis of variance (ANOVA). The
correlation curve between the volume and weight regarding
the holding capability was also established. The test balls
were divided into two types according to their weight and
volume:
(I) Test balls of the same weight but different volumes:
Styrofoam balls with diameters of 60, 70, and 80 mm were
embedded with a 150 g weight for grip tests.
(II) Test balls with the same volume but different
weights: Styrofoam balls 80 mm in diameter were
embedded with 100, 150, and 200 g for grip tests.
Different Twisting Arrangements Test
To validate the suction force of the end-effector and the
force required to pick tomatoes, the tomato picking force in
actual plantation areas was examined. In normal picking
operations, farmers first twist and then bend the peduncle
of the stalk to harvest the fruit. Therefore, this study
examines two methods of fruit picking, that is, with
twisting and without twisting. The force gauge holds the
tomato using a net, and then pulls the tomato to determine
the required force. The twisting operation was categorized
into three levels according to the rotation angles: (1)
clockwise for 60° followed by a counterclockwise turn of
120° until the fruit is 60° counterclockwise to the starting
alignment, repeated three times; (2) clockwise for 90°
followed by a counterclockwise turn of 180° until the fruit
is 90° counterclockwise to the starting alignment, repeated
three times; and (3) clockwise for 120° followed by a
counterclockwise turn of 240° until the fruit is 120°
counterclockwise to the starting alignment, repeated three
times. The twist angles were measured using angle plates
(fig. 9). Each test was repeated 20 times.
Picking Test with Actual Tomato Plants
To examine the suction effect of the fruit suction device
on fruits growing from actual plants, the vacuum force,
suction nozzle diameter, and inching method investigated
in Sections 2.4.2, 2.4.3, and 2.4.4 were used for suction
tests on actual fruit. The robotic arm determines the
coordinates of the tomato to be picked using the visual
system, which drives the robotic arm to perform the picking
operation. The success criterion was the adherence of the
fruit to the suction nozzle, and the removal of the fruit from
the plant.
The tomato samples were Momotaro tomato plants
purchased from Jin Yong farm in Guanxi Township,
Hsinchu County. The plants were cultivated in pots and had
an average height of 1.5 m. Each pot was 0.6 m in length.
During the experiments, the four pots were employed
together for a total length of 2.4 m. Tests were conducted
on 25 fruits, and the success rate was calculated. The
harvest range is shown in figure 10. A video camera was
used to film the actual harvesting operations. The time
required to perform each action was recorded and averaged.
Figure 7. Various inching motions.
Figure 8. The end-effector holding force test using test balls.
29(6): 1001-1009
Figure 9. Tomato twist angles were measured using angle plates.
1005
100
Carrier movement
Platform movement
Photograph scene
Success rate of suction, %
90
80
70
112 g
148 g
215 g
278 g
186 g
60
Figure. 10 The integrated harvesting test.
9.1
8.6
8.1
7.7
7.2
Vacuum suction, N/cm2
RESULTS AND DISCUSSION
ANALYSIS OF TOMATO PHYSICAL PROPERTY
The tomato samples used in this study were obtained
from the Zhou Deng Xiang farm in Zhuang Wei Township,
Yilan County, and measured in a laboratory. Fifty tomatoes
were randomly selected as the study samples. The results
show that the mean and standard deviation was 74.5±
4.9 mm for the major axis; 68.7±6.1 mm for the minor axis;
67.5±4.5 mm for the height; and 165±23.2 g for the weight.
Figure 11 shows the regression relationship between the
major axis (x) and the weight (y). The regression equation
was y = 4.3185x-156.68, with the coefficient of
determination R2 = 0.8368. This shows that the volume and
the weight of tomatoes are correlated. In other words, larger
fruits are heavier. According to the results of the physical
property test, the end-effector designed in this study is
suitable for fruit with a diameter ranging between 60.0 and
90.0 mm and a weight ranging between 100 and 200 g.
ANALYSIS OF FRUIT VACUUM SUCTION FORCE
Figure 12 shows the suction success rate when using
various suction forces and fruit weights. The curves
indicate that higher suction forces provide a superior
suction rate. When the vacuum suction force was increased
to 8.1 N/cm2, the success rate reached 86.7%. The average
240
220
200
Weight, g
180
160
y = 4.3185x - 156.68
140
R2 = 0.8368
120
100
65
70
75
Major axis, mm
80
85
Figure 11. The regression relationship between the major axis and
weight of tomatoes.
1006
Figure 12. Suction success rate for various fruit weights using
different vacuum forces.
success rate was 94.0%. However, suction forces greater
than 8.1 N/cm2 offer no significant improvement. The
ANOVA results show that both the suction force and the
fruit weight significantly affect the suction success rate.
The results also indicate that heavier fruits reduce the
suction success rate. This is because the fruit tends to loose
suction when the vacuum suction attachment is retracted.
No significant difference was observed for vacuum suction
forces over 8.1 N/cm2. Thus, the vacuum suction force of
8.1 N/cm2 was used in this study.
EXPERIMENTAL RESULTS OF SUCTION NOZZLE
DIAMETER
Figure 13 shows the suction success results for various
combinations of tomato sizes and vacuum suction nozzle
diameters. The results indicate that regardless of the fruit
size, the suction nozzle with a 15 mm diameter had the
highest success rate, with a low of 86.7% and an average of
95.3%. The average success rates for the suction nozzles
with a diameter of 17.6, 18.2, and 19.5 mm were 92.0%,
91.3%, and 88.0%, respectively. With larger suction nozzle
diameters, larger tomatoes have higher success rates. This
is because suction nozzles with a comparatively larger
diameter require a greater fruit surface area, such as
provided by larger tomatoes. The ANOVA results indicate
that both the weight of the tomatoes and the vacuum
suction nozzle diameters significantly affect the suction
success rate. The laws of physics define pressure as the
normal force divided by the contact area. Thus, if the
vacuum suction force is fixed, smaller action areas generate
greater negative pressures. Consequently, reducing the
suction nozzle size increases the vacuum strength. For this
study, the 15 mm suction nozzle was selected based on the
above results.
EXPERIMENTAL RESULTS OF VARIOUS INCHING
MOVEMENT TYPES
Table 2 shows the results from the Latin square test of
various inching movement types and their effects on the
suction success rate. The results show that the vertical
APPLIED ENGINEERING IN AGRICULTURE
1000
100
Without vacuum suction
Vacuum force of 8.1N/cm^2
95
Vacuum force of 7.2 N/cm^2
90
Pull force, gf
Success rate of suction, %
800
85
Nozzle diameter 15.0mm
Nozzle diameter 17.6mm
Nozzle diameter 18.2mm
Nozzle diameter 19.5mm
80
75
600
400
70
200
112
148
186
Tomato weight, g
215
278
60
70
Diameter, mm
80
Figure 13. Suction success rate of various combinations of tomato
weight and vacuum suction nozzle diameter.
Figure 14. Results of the end-effector holding force test using test balls
of the same weight but varying diameter.
inching movement provides the highest suction success rate
at 88.9%, followed by horizontal inching at 84.4%. The
average success rate of back and forth inching was the
lowest at 72.2%. Regardless of the inching movement type,
the success rates are higher when the inching travel
distances are further. The ANOVA results show that the
inching movement type and inching travel distance
significantly affects the success rate. The diameter of the
fruit, however, has no significant influence. Thus, for this
study, the vertical inching movement and a 60 mm travel
distance were used.
the vacuum suction force and test ball diameter significantly affect the holding force of the end-effector.
The results of the holding force test using balls with the
same diameter (80 mm) and different weights show that
when using the griper without suction assistance, the force
required to pull balls weighing 100, 150, and 200 g were
531 ± 12.8 gf, 539 ± 10.4 gf, and 547±11.3 gf, respectively.
The average pull force was 539 g. When the gripper was
combined with a vacuum suction of 8.1 N/cm2, the pull
force was 647 ± 10.9 gf, 654 ± 11.0 gf, and 660 ± 13.6 gf,
respectively. The average was 653 g. When the vacuum
suction was increased to 7.2 N/cm2, the pull forces were
815 ± 14.0 gf, 819 ± 11.4gf, and 821 ± 11.7gf. The average
was 818 g. These results show that test balls of varying
weight require similar forces to pull them free.
EXPERIMENTAL RESULTS OF THE END-EFFECTOR
HOLDING FORCE
Figure 14 shows the results of the end-effector holding
force test using test balls of the same weight but different
diameters. The figure shows that for the 60, 70, and 80 mm
test balls held by the gripper without vacuum suction, the
force required to pull them free from the end-effector were
311 ± 9.5 gf, 413 ± 10.7 gf, and 539 ± 10.9 gf, respectively.
When the gripper’s hold was augmented by a vacuum
suction of 8.1 N/cm2, the forces required to remove the test
balls of varying diameters were 413 ± 10.2 gf, 530 ± 9.5 gf,
and 647 ± 10.3 gf. If the vacuum force was increased to
7.2 N/cm2, the required pulling force was 507 ± 9.8 gf, 673
± 10.3 gf, and 829 ± 10.4 gf. These findings indicate that
larger balls require greater forces to be pulled away.
Vacuum suction of the fruits increased the end-effector’s
ability to hold the fruit. Greater vacuum suction offers
greater holding force. The ANOVA results show that both
Table 2. Latin square test of various end-effector inching
movement types and their effects on the suction success rate.
Inching Movement Type
Inching Travel Distance
Vertical
Horizontal
Back and Forth
(mm)
20
A=83.3%
C=80.0%
B=70.0%
40
B=90.0%
A=83.3%
C=73.3%
60
C=93.3%
B=90.0%
A=73.3%
Mean (%)
88.9
84.4
72.2
Standard deviation
5.09
5.09
1.92
29(6): 1001-1009
EXPERIMENTAL RESULTS OF THE TOMATO PICKING
FORCE
Table 3 shows the results of the tomato picking force
test. A degree of 0 indicates no twisting. The results for infield tomato picking tests without twisting indicate that the
average forces required to remove the fruit from the plant
was 2,110 g, with a standard deviation of 149 g, and
maximum value of 2,450 g. The minimum value required
was 1,900 g, which is also a substantial force. Compared
with the holding force tests in Section 3.5, the above results
indicate that the end-effector cannot pull the fruit from the
plant. Therefore, the tomatoes must first be twisted to
enable picking.
Table 3 shows that when the twisting angle was 120°,
the average pulling force was 401 g. However, this declined
to 360 g with 180° twists. When the twisting angle reached
240°, the average pull force required was only 260 g. This
indicates that significant twisting angles reduce the force
required to pick tomatoes. Compared with the holding
forces presented in Section 3.5, these results indicate that
the pull force required is less than the forces available. This
means that the end-effector designed in this study can
remove the tomato from the plant regardless of the twist
angle. However, greater twist angles require longer twisting
times, which then increases the average picking time. Thus,
1007
Table 3. The tomato picking forces required
under various twisting angles.
Twisting angle (degree)
0 (no twist)
120
180
Mean (g)
2110
401
360
Standard Deviation
149
175
145
Max (g)
2450
690
610
Min (g)
1900
150
90
240
260
123
530
60
the 120° turning mode was selected, that is, clockwise for
60° followed by a counterclockwise turn of 120° until the
fruit is 60° counterclockwise to the starting alignment,
repeated three times.
EXPERIMENTAL RESULTS OF PICKING REAL TOMATO
PLANTS
Among the 25 sample tomatoes, 23 were picked
successfully and 2 were not. The success rate was 92%.
The test results indicate that the vacuum suction device can
generally attach to the fruit securely. The two failures were
because the robotic arm was positioned too low. This
placed the suction attachment point near the lower edge of
the fruit, preventing the suction nozzle from fitting with the
fruit surface correctly, leading to suction failure. Because
the fruit could not be pulled into the end-effector correctly,
the picking attempt failed.
The automatic picking operation including robot carrier
movement, platform movement, and picking was controlled
using the LabVIEW program. From the time that the
robotic arm extends to pick the tomato until it retracts is
60.3 s. Within this time, end-effector operations, such as
extension, vertical inching, twisting rotation, retraction, and
depositing comprised the majority. The total time required
to pick 25 fruits was 1865 s, averaging 74.6 s for each fruit
in the condition of 40 mm/s for the robot carrier movement
speed, 135 mm/s for the platform ascent speed, and
100 mm/s for the descent speed. The overall test results
show that no damage occurred during gripping, pulling, and
twisting actions. The end-effector designed in this study
achieved its intended function and is suitable for practical
application. Through further integration with the overall
robotic picking system, the goal of automated tomato
picking should be achieved in the future.
CONCLUSIONS
In this study, an end-effector for picking greenhousegrown Momotaro tomatoes has been developed. The endeffector was designed with four fingers and a centrally
located fruit suction device. Using suction, the robotic arm
was designed to remove the fruit from the plant, reducing
damage caused by the end-effector to the plant and the
other fruits. The suction force also helps secure the fruit to
the inside of the end-effector, reinforcing the holding
capability.
During this study, the suction success rate was examined
for tomato fruits of varying weight. The results show that
larger vacuum suction forces offer higher suction success
rates. When the vacuum suction force was increased to 8.1
N/cm2, the suction success rate reaches 86.7%, with an
average suction success rate of 94.0%. However, no
1008
obvious improvements were observed with vacuum suction
forces greater than 8.1 N/cm2.
The suction success rates of various combinations of
tomato size and vacuum suction nozzle diameter show that
regardless of the fruit size, the 15.0 mm diameter suction
nozzle has the highest suction success rate at more than
86.7%. The average suction success rate was 95.3%. The
average suction success rates for the suction nozzles with a
diameter of 17.6, 18.2, and 19.5 mm were 92.0%, 91.3%,
and 88.0%, respectively.
The suction success rate of suction nozzles contacting
fruit under different end-effector inching movements was
experimented. The inching movement types were vertical,
horizontal, and back and forth. The inching movement
travel distances were 20, 40, and 60 mm, performed twice.
The test results indicated that inching the end-effector
vertically provides the highest suction success rate at
88.9%. Greater inching travel distances also provide higher
suction success rates.
The force required for picking by the end-effector under
various twisting angles was also tested and analyzed. The
results indicate that turning the end-effector clockwise for
60° followed by a counterclockwise turn of 120° until the
fruit is 60° counterclockwise to the starting alignment,
repeated three times before picking can effectively remove
the fruit from the plant. The initial laboratory tests have
shown promising results. Future integration with robotic
tomato harvesting systems for actual tomato picking field
tests can increase the feasibility and practical applications
of robotic automatic picking systems.
ACKNOWLEDGEMENTS
This project was financially sponsored by the Agriculture and Food Agency, Council of Agricultural Executive
Yuan, Taiwan. The authors would like to thank the research
assistances from Jia-Feng Lin and Rou-Jing Li, as well as
supports and advices from Mr. Cai-Fu Hsu from Qun Hui
Mechanics and Mr. Chien-Chung Liu from Tung Chian
Co., Ltd.
REFERENCES
Arima, S., N. Hayashi, and M. Monta. 2004. Strawberry harvesting
robot on table-top culture. ASAE Paper No. 043089. St. Joseph,
Mich.: ASAE.
Bulanon, D. M., T. Kataoka, H. Okamoto and S. I. Hata. 2005.
Feedback control of manipulator using machine vision for
robotic apple harvesting. ASAE Paper No. 053114. St. Joseph,
Mich.: ASAE.
Chiu, Y. C., S. Chen, and J. F. Lin. 2012. Development of an
autonomous picking robot system for greenhouse grown
tomatoes. In The 6th Intl. Symp. on Machinery and
Mechatronics for Agr.and Biosystems Eng. Korea: Jeonju.
Hayashi, S., S. Yamamoto, S. Saito, Y. Ochiai, Y. Nagasaki, and Y.
Kohno. 2013. Structural environment suited to the operation of a
strawberry-harvesting robot mounted on a travelling platform.
Engineering in Agriculture, Environment and Food 6(1): 34-40.
Hayashi, S., K. Shigematsu, S. Yamamoto, K. Kobayashi, Y.
Kohno, J. Kamata, and M. Kurita. 2010. Evaluation of a
strawberry-harvesting robot in a field test. Biosystems
Engineering 105(2): 160-171.
APPLIED ENGINEERING IN AGRICULTURE
Ji, W., D. Zhao, F. Cheng, B. Xu, and Y. Zhang. 2012. Automatic
recognition vision system guided for apple harvesting robot.
Computers & Electrical Engineering 38(5): 1186-1195.
Johan, B., D. Kevin, B. Sven, B. Wim, and C. Eric. 2008.
Autonomous fruit picking machine: A robotic apple harvester.
Springer Tracts in Advanced Robotics 42: 531-539.
Kondo, N., K. Yata, M. Iida, T. Shiigi, M. Monta, M. Kurita, and H.
Omori. 2010. Development of an end-effector for a tomato
cluster harvesting robot. Engineering in Agriculture,
Environment and Food 3(1): 20-24.
Ling, P. P., R. Ehsani, K. C. Ting, Y. T. Chi, N. Ramalingam, M. H.
Klingman, and C. Draper. 2004. Sensing and end-effector for a
robotic tomato harvester. ASAE Paper No. 043088: St. Joseph,
Mich.: ASAE.
29(6): 1001-1009
Plebe, A., and G. Grasso. 2001. Localization of spherical fruits for
robotic harvesting. Machine Vision and Applications 13(2): 7079.
Reed, J. N., S. J. Miles, J. Butler, M. Baldwin, and R. Noble. 2001.
Automatic mushroom harvester development. J. of Agricultural
Engineering Research 78(1): 15-23.
Tanigaki, K., T. Fujiura, A. Akase, and J. Imagawa. 2008. Cherryharvesting robot. Computers and Electronics in Agriculture 63:
65-72.
Van Henten, E. J., B. A. J. Van Tuijl, J. Hemming, J. G. Kornet, J.
Bontsema and E. A. Van Os. 2003. Field test of an autonomous
cucumber picking robot. Biosystems Engineering 86(3): 305313.
1009