Industrial Robot

Ind
us

Workspace analysis of Cartesian robot system for kiwifruit

tri

harvesting
al

Journal: Industrial Robot

Manuscript ID IR-12-2019-0255.R3
Ro

Manuscript Type: Original Manuscript

Keywords: Cartesian robot, articulated robot, workspace, kiwifruits, harvesting

bo
t
Page 1 of 17 Industrial Robot

1
2
3 Workspace analysis of Cartesian robot system for kiwifruit harvesting (Revised version)
4
5
6 Abstract
7
8 Purpose – This article aims to investigate if a Cartesian robot system for kiwifruit harvesting
9 works more effectively and efficiently than an articulated robot system. Robot is a key
10 component in agricultural automation. For instance, multiple robot arm system has been
11
developed for kiwifruit harvesting recently because of the significant labour shortage issue.
12
13 The industrial robots for factory automation usually have articulated configuration which is
14 suitable for the tasks in manufacturing and production environment. However, this articulated
15 configuration may not fit for agricultural application due to the large outdoor environment.
16
17 Design/methodology/approach – The kiwifruit harvesting tasks are completed step by step so
18 that the robot workspace covers the canopy completely. A two–arm, Cartesian kiwifruit
19
20
harvesting robot system and several field experiments are developed for the investigation.
21 The harvest cycle time of the Cartesian robot system is compared to that of an articulated
Ind

22 robot system. The difference is analysed based on the workspace geometries of these two
23 robot configurations.
24
25 Findings – It is found that the kiwifruit harvesting productivity is increased by using a
us

26
multiple robot system with Cartesian configuration owing to its regular workspace geometry.
27
28
Originality/value – Articulated robot is a common configuration for manufacturing due to its
tri

29
30 simple structure and the relatively static factory environment. Most of the agricultural
31 robotics researches employ single articulated robot for their implementation. This article
al

32 pinpoints how the workspace of a multiple robot system affects the harvest cycle time for
33 kiwifruit harvesting in a pergola style kiwifruit orchard.
34
Ro

35
36 Keywords: Cartesian robot, articulated robot, workspace, kiwifruits, harvesting
37
38 1. Introduction
bo

39
40 An industrial manipulator is usually an articulated robotic arm with rotational axes. However,
41
it is not necessarily optimal for every task. Without considering the end-effector orientation, a
t

42
43 three-axis arm provides enough degrees of freedom necessary for 3D positioning. This
44 approach is used in developing a three–axis planar articulated robot arm with an autonomous
45 mobile platform for picking kiwifruit (Williams et al 2019). Nevertheless, there are
46 speculative avenues for improving this configuration, such as the workspace efficiency of the
47 robot arms and harvesting efficiency of end–effector.
48
49
50
If the workspace geometry of a robot conforms to the canopy geometry, the robot arm
51 reachability to complete the task will be high. Hence, a kiwifruit is reachable if its position in
52 the canopy is inside the workspace of a robot arm. The harvesting will then be more efficient
53 as it does not need to move the robot base frequently to include the fruit into its workspace.
54 However, different configurations (such as articulated robot and Cartesian robot) yield
55 different workspace geometry.
56
57
58 This article investigates the effects of robot arm workspace geometry on a multiple robot arm
59 system for kiwifruit harvesting. It is hoped that the investigation of robot configuration will
60 promote the kiwifruit harvesting automation for alleviating current labour shortage.

1
 Industrial Robot Page 2 of 17

1
2
3 2. Related work
4
5
6 Many agricultural tasks such as harvesting, pollination, crop-maintenance etc are typically
7 repetitive. In order to be economically competitive with equivalent human labour, it is
8 beneficial for an agricultural machine to complete these tasks effectively, efficiently and
9 reliably within environmentally challenging conditions.
10
11
A harvesting robot typically consists of three main components (Bachche 2015):
12
13
14 1. A mobile platform for movement.
15 2. A recognition system for fruit identification and location.
16 3. A manipulator and an end-effector for fruit harvesting.
17
18 A comprehensive review on 50 robotic harvesting systems finds that the average harvesting
19
20
robot will locate 85% of the fruit, detach 75%, harvest 66% and damage 5% (Bac et al 2014).
21
Ind

22 Harvest cycle time is a metric for measuring the robot performance. This is the time taken to
23 locate, detach and harvest the fruit. The average harvest cycle time for published harvesting
24 robot systems in 2014 is 33 seconds (Bac et al 2014). Much of the technical challenges
25 involved in robotic harvesting are well known (Sarig 1993, Li, Lee and Hsu 2011, Kapach et
us

26
al 2012), and mostly arise from task complexity in a dynamic environment.
27
28
One approach to improve the harvest cycle times is to optimize the hardware configuration. A
tri

29
30 cucumber harvesting robot has been developed (Henten et al. 2009). However, it is
31 acknowledged that the original manipulator kinematic structure is far from ideal. One study
al

32 proposes the design optimization and simulation of the kinematic structure of an eggplant
33 picking robot (Han et al. 2007). Another study (Baur 2014) covers several topics essential for
34
the design of a modular pepper harvesting and precision spraying robot manipulator.
Ro

35
36 Investigation on the kinematic design for this system is shown (Baur et al. 2012) by analysing
37 various high degree of freedom workspace combinations in relation to the task space.
38 Another methodology (Bloch et al. 2017) for optimizing the kinematics and location of an
bo

39 agricultural robotic arm is presented by considering the characteristic environment that

40 approximates the actual orchard environment.
41
t

42
43 In spite of the aforementioned contributions, the majority of robotic harvesting systems focus
44 their research on identification and manipulator control instead of optimizing the hardware
45 configuration. There are many published robotic harvesting systems across different crop
46 types since 2018 such as cucumber (van Henten et al. 2003), sweet pepper (Bac et al. 2017),
47 radicchio (Foglia and Reina2006), tomato (Li et al 2017), strawberry (Hayashi et al 2014),
48 apples (Silwal et al 2017) and kiwifruit (Mu et al 2020).
49
50
51 However, explicit justifications for adopting certain kinematic structures are found only on
52 single robotic arm harvesting systems and most of the robot arms employed are articulated
53 robots. In order to raise the efficiency and effectiveness of employing robots for harvesting,
54 multiple robots should be considered. For instance, a robot with four articulated picking arms
55 (Scaefe 2009) is developed for harvesting kiwifruits. Nevertheless, if the robot configuration
56 for fruit harvesting is not efficient and effective, then this insufficiency is just magnified for a
57
58
multiple robot system. As a result, this arrives at the research question that how the robot arm
59 workspace affects the kiwifruit harvesting, especially, when multiple robot system is used.
60

2
Page 3 of 17 Industrial Robot

1
2
3 3. Harvest cycle time and workspace geometry
4
5
6 Most of the kiwifruit in an orchard are grown in a rectangular array with the vines trained
7 onto a pergola structure so that the fruit hang across the canopy as shown in Figure 1. The
8 canopy is described as a Cartesian system with x axis along the length and y axis along the
9 width of the array of fruit trees.
10
11
12
13
14
15
16
17
18
19
20
21
Ind

22
23
24
25
us

26
27
28
tri

29
30
31
al

32
33
34
Ro

35
36
37
38 Figure 1. Kiwifruit in the canopy.
bo

39
40 The harvesting process is assumed to be performed by a robot system with 𝑁𝑎𝑟𝑚𝑠 arms. It
41 also has a machine vision system for locating the kiwifruit. Due to the large size of kiwifruit
t

42
orchard, the harvesting process performed by the robot system is split into rows of steps.
43
44 Each discrete harvesting step is executed longitudinally along a row. The process is
45 completed row by row laterally.
46
47 A step consists of the following processes:
48
49 i. The mobile platform starts moving from one location to another location
50
51
ii. The kiwifruit are detected and located by the vision system when the platform
52 arrives at a specific location
53 iii. The located kiwifruit are all harvested by the robot arms
54
55 A fruit scheduler (Barnett 2018) is implemented in the robot system to read the fruit locations,
56 allocate the fruit to the robot arms and determine the picking order.
57
58
59 Because of the non-uniform fruit distribution, cluster growing style and fruit location relative
60 to the robot arm position (as these factors affect the picking order and robot arm availability),

3
 Industrial Robot Page 4 of 17

1
2
3 the work distributed to each robot arm by the fruit scheduler may not balance. A parameter
4
5
termed work distribution 𝑊𝐷 is introduced to reflect the imbalance of work distribution
6 among the robot arms. It is defined as
7
8 𝑊𝑝𝑎𝑟𝑖𝑡𝑦
9
𝑊𝐷 = 𝑊𝑏𝑢𝑠𝑖𝑒𝑠𝑡 (1)
10
11 where 𝑊𝑝𝑎𝑟𝑖𝑡𝑦 is the average number of fruits which should be harvested by a robot arm and
12
13
14 𝑊𝑏𝑢𝑠𝑖𝑒𝑠𝑡 is the maximum number of fruits harvested by one robot arm
15
16 The average harvest cycle time 𝑡𝐻𝐶𝑇 is the average time required to harvest a kiwifruit, which
17 is
18
19 𝑁𝑠𝑡𝑒𝑝𝑠(𝑡𝑚𝑜𝑣𝑒 + 𝑡𝑙𝑜𝑐𝑎𝑡𝑒) 𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝
20 𝑡𝐻𝐶𝑇 = + 𝑁𝑎𝑟𝑚𝑠 ∙ 𝑊𝐷. (2)
𝑁𝑓𝑟𝑢𝑖𝑡𝑠
21
Ind

22
23 where 𝑡𝑚𝑜𝑣𝑒 and 𝑡𝑙𝑜𝑐𝑎𝑡𝑒 are the average times for the autonomous robotic system to move
24 from its current position to the position of next step, detect and locate the fruit in each step
25 respectively. 𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝 is the sub-step time taken for a robot arm to detach a fruit, and then
us

26
27
move to the next fruit. 𝑁𝑠𝑡𝑒𝑝𝑠 is the number of steps to completely harvest 𝑁𝑓𝑟𝑢𝑖𝑡𝑠 kiwifruits.
28
At each step, a space is obtained by intersecting the robot workspace with the canopy, which
tri

29
30 contains the fruit at the reachable positions. These fruit will be harvested. For each row, the
31 volume of the space in the first step and the subsequent steps are termed harvestable volume
al

32
𝑉ℎ and step advance volume 𝑉𝑎 respectively such that
33
34
(3)
Ro

35 𝑉𝑎 ≤ 𝑉ℎ
36
37 Consider the average volumetric density of fruit 𝜌 (number of fruit per unit volume) within
38 𝑁𝑓𝑟𝑢𝑖𝑡𝑠
bo

39 an orchard, the number of fruits per step ratio 𝑁𝑠𝑡𝑒𝑝𝑠

is approximated as
40
41 𝑁𝑓𝑟𝑢𝑖𝑡𝑠
= 𝜌 ∙ 𝑉𝑎 (4)
t

42 𝑁𝑠𝑡𝑒𝑝𝑠
43
44
45 Thus, equation (2) can be rewritten as
46
𝑡𝑚𝑜𝑣𝑒 + 𝑡𝑙𝑜𝑐𝑎𝑡𝑒 𝑡𝑠𝑢𝑏 - 𝑠𝑡𝑒𝑝
47 𝑡𝐻𝐶𝑇 = +𝑁 . (5)
48 𝜌 ∙ 𝑉𝑎 𝑎𝑟𝑚𝑠 ∙ 𝑊𝐷
49
50 As a result, the harvest cycle time can be estimated based on the average volumetric density
51 of fruit in an orchard. In order to shorten the harvest cycle time, the step advance volume
52
should be increased by having large step advance distance.
53
54
55 Figure 2 depicts the difference in step advance volume covered by a Cartesian robot and an
56 articulated robot. Since the workspace geometry of a Cartesian robot is rectangular shape, the
57 step advance volume equals to its harvestable volume. On the other hand, the workspace
58 geometry of an articulated robot is irregular shape; the step advance volume is less than its
59 harvestable volume. As a result, a Cartesian robot has a larger step advance volume than that
60

4
Page 5 of 17 Industrial Robot

1
2
3 of an articulated robot if both possess the same harvestable volume or a Cartesian robot has
4
5
larger harvestable volume than that of an articulated robot. Since the union of all the
6 harvestable volume of each step must cover the complete canopy (task space), employing a
7 Cartesian robot for fruit harvesting should have less number of steps. Of course, the step
8 advance volume of an articulated robot can be larger than that of a Cartesian robot if its
9 harvestable volume is large enough. However, this implies a comparatively large workspace
10 or robot dimensions.
11
12
13
14 z 𝑉ℎ Step advance
15 distance
16
17
18
19
20
x
21 𝑉𝑎
Workspace
Ind

22
23
24 Canopy (task space)
25
us

26 Workspace of robot
27
arm
28
Harvestable volume 𝑉ℎ
tri

29
30
31
Step advance volume 𝑉𝑎
al

32
33
34 (a) A Cartesian robot
Ro

35
36
37 z 𝑉ℎ
38 Step advance distance
bo

39
40
41
t

42
43
44 𝑉𝑎
45
46
Canopy (task space)
47
48 Workspace of robot
49 arm
50 Workspace Harvestable volume 𝑉ℎ
51
52
53 Step advance volume 𝑉𝑎
54
55 (b) An articulated robot
56
57
58
59 Figure 2. Workspace geometry of robot arm.
60

5
 Industrial Robot Page 6 of 17

1
2
3 4. Field tests
4
5
6 The field tests aim to investigate the kiwifruit harvesting performance of a Cartesian robot.
7 Figure 3 shows a two-arm, Cartesian kiwifruit harvesting robot, which is developed as a
8 research platform to perform the workspace analysis and investigate how it affects the
9 kiwifruit harvesting comparing to an articulated robotic system.
10
11
Each Cartesian robot-arm i has 𝑥𝑖, 𝑦𝑖 and 𝑧𝑖-axis (i =1, 2). Axis 𝑥1 and 𝑥2 for both robot arms
12
13 share a common axis such that they can move synchronously throughout a resultant
14 workspace.
15
16
17
18
Resultant
19
20 workspace
21
Ind

22
23
24
25 Arm 2
us

26 Arm 1
27
28
tri

29
30
31
al

32
33
34
Ro

35
36
37
38
bo

39
40
41
t

42
43
44
45
46
47
48
49
50
51 Figure 3. A Cartesian robot system with two arms and its resultant workspace.
52
53 The limits for axis x, y and z are 1500mm, 650mm and 450mm respectively, which yields a
54
55
resultant rectangular workspace volume as shown. Figure 4 shows the multiple robot systems
56 in the orchard.
57
58
59
60

6
Page 7 of 17 Industrial Robot

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Ind

22
23
24
25
us

26
27
28
tri

29
30
31
Figure 4. A Cartesian robot system with two arms in the orchard.
al

32
33
34 The robot system is positioned such that the maximum arm reachable height is just above the
canopy top. The end-effector consists of a plastic sheath with a slightly conical upper edge
Ro

35
36 and a beak-like shutter. Once the kiwifruit is enveloped by the sheath, the beak shutter
37 actuates closed. The end-effector assembly rotates with fruit inside and then the manipulator
38 retracts the end-effector downwards. These three steps are illustrated in Figure 5.
bo

39
40
41
t

42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60

7
 Industrial Robot Page 8 of 17

1
2
3
4
5
6
7
8
9
10
11
12
13 End-effector
14
moves
15
16 upwards along
17 the z axis of
18 the robot
19 system.
20
21
Ind

22
23
24 (a) The kiwifruit (green) is (b) The shutter is closed. (c) The end-effector assembly
25 about to be enveloped. rotates with fruit inside.
us

26
27 Figure 5. The operations of the end-effector.
28
tri

29
As kiwifruit are grown on pergola frames and hang down vertically, the end-effector attached
30
31
to the Cartesian robot approaches the fruit from the bottom vertically in the z direction as
shown in Figure 6. Furthermore, kiwifruit from the early-maturing orchards has been picked
al

32
33 for early shipments, hence, all the fruit will be harvested and there is no concern for unripe
34 fruit avoidance in the end-effector path. The grasping success rate of this end-effector is high
Ro

35 as the inference of fruit poses (which is almost vertically oriented) is ignored.

36
37
38
bo

39
40
41
t

42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60 Figure 6. The end-effectors are harvesting the kiwifruit.

8
Page 9 of 17 Industrial Robot

1
2
3 The kiwifruit harvesting performance is evaluated in ten steps, five from the Batemans
4
5
orchard and five from the Newnham orchard. Both orchards grow Hayward strain kiwifruit
6 with a pergola style located in Tauranga, New Zealand.
7
8 The performance of the robot is measured in two perspectives, which are the success rate and
9 the harvest cycle time. Success rate is the ratio of number of fruits harvested to all fruit in the
10 canopy during a specific step. This also dictates how many fruit can be physically reached
11
and successfully harvested by the end-effector with the required dexterity.
12
13
14 It is assumed that all harvested fruit are non-damaged as an optimal case scenario and instead
15 focus will be made toward the knocking or dropping of fruit as part of the end-effector
16 detachment process. All kiwifruit, which are knocked out of the canopy or dropped by the
17 end-effector, are considered rejects in industry; thus, they are considered damaged fruit in
18 this evaluation.
19
20
21 The harvest cycle time is obtained by equation (2) with average measured values of time for
Ind

22 moving the robot between steps (𝑡𝑚𝑜𝑣𝑒), time for the vision system to locate the fruit (𝑡𝑙𝑜𝑐𝑎𝑡𝑒),
23 time for the robot to move between fruit harvesting (𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝) and fruit density per step (𝜌).
24
25
5. Results
us

26
27
28 The robot system for kiwifruit harvesting is a cooperative application where the robot arms
work together to perform the harvesting task step by step. All the fruit in the canopy are
tri

29
30 reachable and will be harvested if they are within the resultant robot arms workspaces in the
31 current step. Those outside the workspace are unreachable. However, these unreachable fruit
al

32
are also located by the vision system as the portion of canopy captured by the camera is
33
34 larger than the workspace of robot. Hence, they must be eliminated from the located fruit list
before the fruit are scheduled to be harvested. These fruit will be harvested in the other steps.
Ro

35
36
37 Table 1 shows the success rate across the 10 steps. Of the fruit that are attempted, on average
38 77% are successfully harvested and 12% are dropped. These fruit are mainly knocked off by
bo

39 the end-effector. Some fruit are missed due to the shifting of fruit position when their
40
41
neighbouring fruit are harvested within a cluster.
t

42
43 Table 1. The success rate of kiwifruit harvesting across 10 steps
44
45 Step Total
46
47 1 2 3 4 5 6 7 8 9 10 All steps
48
49 Number of 62 51 28 35 47 26 31 34 52 35 401
50 fruit registered
51
52 Success rate 81 80 93 57 72 54 87 76 81 86 77
53
(%)
54
55 Dropped rate 16 0 7 14 15 19 13 18 8 11 12
56
57 (%)
58
59
60 The average volumetric density of fruit within these ten steps is 137 fruit/m3.

9
 Industrial Robot Page 10 of 17

1
2
3 The average of recorded time to advance between steps and locate kiwifruit (𝑡𝑚𝑜𝑣𝑒 + 𝑡𝑙𝑜𝑐𝑎𝑡𝑒)
4
5 is 3 minutes and the common sub-step time (𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝 ) is 2s.
6
7 Combining these variables and assuming that both robot arms share the same amount of work
8 (that is 𝑊𝐷 = 1), the average harvest cycle time for the two-arm Cartesian robot (𝑡𝐻𝐶𝑇) in
9
10
these ten steps is approximately 6.8s.
11
12 6. Discussion
13
14 The harvest cycle time is reduced if the robot has a large step advance volume. Hence, the
15 workspace geometry is one of the key factors to determine if a Cartesian robot is more
16
effective than an articulated robot for kiwifruit harvesting.
17
18
19 As a case study, the workspace of an articulated robot system with two arms is analysed by
20 comparing to the Cartesian robot system. The articulated robot used is RA605 series from
21 Hiwin Corporation (Hiwin 2020). This is a commercially available industrial robot for
Ind

22 generic application. Although it is not specifically for outdoor kiwifruit harvesting

23 application, its workspace geometry is irregular due to its articulated configuration. The
24
25
comparison is to investigate the effect of workspace geometry on the number of steps
required to cover the canopy.
us

26
27
28 The robot system comprises two articulated robot arms which are evenly spaced 325mm
tri

29 across the width. The harvester is approximately 838mm from the ground and is able to reach
30 a maximum height of 1869.75mm for picking the kiwifruit. The configuration and workspace
31
of the articulated robot system are plotted in Figure 7. The dimensions can be obtained from
al

32
33 the website of Hiwin Corporation.
34
Ro

35
36
Resultant The angular limits for axis 1, 2 and
37 workspace 3 are (-165o, +165o), (-125o, +85o)
38 and (-55o, +185o) respectively.
bo

39
40
41
t

42
43
44
45
46
47
48
49
50 325mm
Axis 3 Axis 2
51
52
53 Axis 1
54
55
56 Figure 7. An articulated robot system with two arms and its resultant workspace.
57
58
59 The harvestable volume and step advance volume of both robot systems are illustrated in
60 Figure 8. The canopy volume is assumed to span 300mm in the vertical z dimension. The x, y

10
Page 11 of 17 Industrial Robot

1
2
3 canopy dimensions span the array row width and row length respectively. The harvestable
4
5
volume and step advance volume of a Cartesian robot system are regular since it has three
6 prismatic joints along its x, y and z axis. Conversely, the geometry of harvestable volume and
7 step advance volume of an articulated robot system are irregular due to the rotational axes.
8
9
10 Harvestable
11 z volume
Step advance
12
13 volume
14
15
16 y
17 Task
18 space
19 300mm
20
21 x
Ind

22
23
24 Step advance
25 distance
us

26
27
28
tri

29
30
31
al

32
33
34 (a) Cartesian robot system.
Ro

35
36 Harvestable
37
z
volume Step advance
38 volume
bo

39
40
41 y
t

42
43
Task
44 300mm
45 space
46 x
47
48
49 Step advance
50 distance
51
52
53
54
55
56 (b) Articulated robot system.
57
58 Figure 8. The harvestable and step advance volume of the robot systems.
59
60

11
 Industrial Robot Page 12 of 17

1
2
3 The kiwifruit locations of the first five steps (tabulated in table 1) are plotted in Figure 9 with
4
5
both robot systems to simulate the harvesting tasks.
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Ind

22
23
24
25
us

26
27 (a) Cartesian robot system
28
tri

29
30
31
al

32
33
34
Ro

35
36
37
38
bo

39
40
41
t

42
43
44
45
46
47
48
49
50 (b) Articulated robot system
51
52
53 Figure 9. Harvesting simulation.
54
55 The fruit in the canopy are partitioned into steps based on these two robot system
56
57
configurations. Figure 10 depicts the number of steps required to complete the harvesting.
58
59
60

12
Page 13 of 17 Industrial Robot

1
2
3
4 3250mm
5
6 Step 1 Step 2 Step 3 Step 4 Step 5
7 (62 fruit) (51 fruit) (28 fruit) (35 fruit) (47 fruit)
8
9 300mm
10
11
12 729.75mm
13
14
15
16
17
18 840mm
19
20
21
Ind

22
23 (a) Cartesian robot system
24
25
us

26 3250mm
27
28 Step Step Step Step Step
1 3 5 7 9
tri

29
30
300mm
31
al

32
33
34 731.75mm
Ro

35
36
37
38 Mobile
bo

39 838mm platform
40
41
t

42
43
44
45 (b) Articulated robot system
46
47 Figure 10. The canopy covered by the harvesting steps.
48
49
50 The workspace volume analysis is tabulated in table 2.
51
52
53
54
55
56
57
58
59
60

13
 Industrial Robot Page 14 of 17

1
2
3 Table 2. Workspace volume analysis between the articulated and Cartesian robot system.
4
5
6 Cartesian robot system Articulated robot system
7
8 No of steps 5 9
9
10 Workspace volume (m3) 0.439 2.15
11
12
13 Harvestable volume (m3) 0.293 0.348
14
15 Step advance volume (m3) 0.293 0.148
16
17
18 ( )
Workspace efficiency = 𝑉ℎ
𝑉𝑎
100% 42.529%
19
20
21
Ind

22 The articulated robot system requires 9 steps to complete the harvesting while the Cartesian
23 robot system only needs 5 steps. Despite the workspace volume of the articulated robot
24 system is much larger than that of the Cartesian robot system (2.15m3 vs 0.439m3), the
25 harvestable volumes are close to each other (0.348m3 vs 0.293m3). This is because the task
us

26 space locates right above the robot systems. However, the difference in step advance volume
27
28
is large (0.148m3 vs 0.293m3).
tri

29
30 The irregular workspace geometry of an articulated robot means that the actual step advance
31 distance is shorter. This implies that the articulated robot system moves, detects and locates
al

32 fruit more frequently. Thus, operational downtime (such as time for the robot system to move
33 and vision system to locate the fruit) increases.
34
Ro

35
36 Non-reachable
37 fruit
38
bo

39
40
41 Task space
t

42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57 Figure 11. A non-reachable fruit in the canopy.
58
59
60

14
Page 15 of 17 Industrial Robot

1
2
3 In fact, nine steps of harvesting tasks using the articulated robots do not completely cover the
4
5
task space as shown in Figure 11. One fruit is not harvested in this simulation as its location
6 (which is the centre of fruit) is outside the step advance volume. This implies that the
7 articulated robots needs to be positioned closer to each other due to its irregular shape
8 harvestable and step-advance volume.
9
10 Under the assumption that both robot systems have the same mobile platform, same vision
11
system for locating fruit, same sub-step time for harvesting and the work distribution among
12
13 the arms are equally balanced (𝑊𝐷 = 1), the estimated harvest cycle time for the robot
14 systems are tabulated in Table 3.
15
16 Table 3. The estimated harvest cycle time for both Cartesian and articulated robot systems.
17
18 𝑡𝑚𝑜𝑣𝑒 + 𝑡𝑙𝑜𝑐𝑎𝑡𝑒 𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝
19
Estimated 𝑡𝐻𝐶𝑇 (s)
20 𝜌 ∙ 𝑉𝑎 𝑁𝑎𝑟𝑚𝑠 ∙ 𝑊𝐷
21 Cartesian robot 180 2
5.48
Ind

22 137 × 0.293
= 4.48 2×1
=1
system
23
24 180 2
25 Articulated robot = 8.88 =1 9.88
137 × 0.148 2×1
system
us

26
27
28
tri

29 The estimated harvest cycle time for articulated robot system is approximately 4s more than
30 that of the Cartesian robot system. However, the vertical rotation axis of the articulated robot
31 system should yield a more efficient lateral movement of the end-effector than that of the
al

32
33
Cartesian robot system, which implies a decrement in the sub-step time 𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝. This
34 efficient lateral movement of end-effector is significant if the time for moving between steps
Ro

𝑡𝑠𝑢𝑏 ― 𝑠𝑡𝑒𝑝
35 and locating fruit are reduced to a comparable value of the term 𝑁𝑎𝑟𝑚𝑠 ∙ 𝑊𝐷. These are the future
36
37 improvements for the harvesting task in orchard.
38
bo

39 These harvest cycle times are obtained under the assumption that the work distribution is
40 uniform (𝑊𝐷 = 1). In fact, the work distribution relates to the kiwifruit distribution and the
41
fruit allocation to the robot arm by the fruit scheduler. The actual average work distribution
t

42
43 for this Cartesian robot system is higher than that of the articulated robot system due to
44 differences in configuration. The work distribution is discussed in another article (Josh et al
45 2020).
46
47
7. Conclusion
48
49
50 Multiple robot arm system should be employed for fruit harvesting in an orchard in order to
51 raise the productivity. The contribution of this article is to pinpoint the fact that the
52 workspace geometry of a robot system is an important factor to consider for determining the
53 robot system configuration.
54
55 When the canopy is much larger than the resultant workspace of a multiple robot arm system,
56 the harvesting task has to be performed step by step. The union of workspaces in each step
57
58 should cover the entire canopy. If the workspace geometry of the robot system conforms to
59 the geometry of canopy volume, the step advanced distance can be large. This implies a
60

15
 Industrial Robot Page 16 of 17

1
2
3 decrease in the number of steps and harvest cycle time. As a result, the harvesting task is
4
5 more efficient.
6
7 Most of the kiwifruit orchards are of pergola style so that the kiwifruit locations can be
8 described by a Cartesian coordinate system. A Cartesian robot system yields a workspace of
9 rectangular shape, which conforms to the canopy volume. This implies no overlapping of
10 harvestable volume between steps. Hence, the step advance volume can be as large as the
11 harvestable volume. Conversely, the step advance volume of an articulated robot is less than
12 its harvestable volume due to its irregular workspace geometry. As a result, using a Cartesian
13
14
robot system to harvest the kiwifruits in a canopy yields less steps and reduces the
15 operational down time. Consequently, the economic benefit is raised.
16
17 Reference
18
19 Bac, C. W., Hemming, J., van Tuijl, B., Barth, R., Wais, E. and van Henten, E. J. (2017),
20
21
“Performance evaluation of a harvesting robot for sweet pepper”, Journal of Field
Ind

22 Robotics, Vol. 34, pp. 123-1139.

23
24 Bac, C. W., van Henten, E. J., Hemming, J., and Edan, Y. (2014), “Harvesting robots for
25 high-value crops: State-of-the-art review and challenges ahead”, Journal of Field Robotics,
us

26 Vol. 31, pp. 888-911.

27
28 Bachche, S. (2015), “Deliberation on design strategies of automatic harvesting systems: A
tri

29 survey”, Robotics, Vol. 4 No. 2, pp. 194-222.

30
31 Barnett, J. (2018), “Prismatic axis, differential-drive robotic kiwifruit harvester for reduced
al

32
33
cycle time”, Master of Engineering Thesis. University of Waikato.
34 Barnett, J., Duke, M., Au, C. and Lim, S. (2020), "Work distribution of multiple Cartesian
Ro

35
36 robot arms for kiwifruit harvesting", Computers and Electronics in Agriculture, Vol. 169,
37 doi.org/10.1016/j.compag.2019.105202
38
bo

39 Baur, J. T. (2014), “Agricultural manipulators simulation, design and motion planning”, PhD
40 thesis, Technischen Universität München.
41
Baur, J. T., Pfaff, J., Ulbrich, H. and Villgrattner, T. (2012), “Design and development of a
t

42
43 redundant modular multipurpose agricultural manipulator”, in 2012 IEEE/ASME
44
International Conference on Advanced Intelligent Mechatronics (AIM), pp. 823-830.
45
46 Bloch, V., Bechar, A. and Degani, A. (2017), "Development of an environment
47
48 characterization methodology for optimal design of an agricultural robot", Industrial
49 Robot, Vol. 44 No. 1, pp. 94-103.
50
51 Foglia, M. M. and Reina, G. (2006), “Agricultural robot for radicchio harvesting”, Journal of
52 Field Robotics, Vol. 23 No. 67, pp. 363-377.
53
54 Han, S., Xueyan, S., Tiezhong, Z., Bin, Z. and Liming, X. (2007), “Design optimisation and
55 simulation of structure parameters of an eggplant picking robot”, New Zealand Journal of
56 Agricultural Research, Vol. 50. pp. 959-964.
57
58
59
60

16
Page 17 of 17 Industrial Robot

1
2
3 Hayashi, S., Yamamoto, S., Saito, S., Ochiai, Y., Kamata, J., Kurita, M. and Yamamoto K.
4
5 (2014), “Field operation of a movable strawberry-harvesting robot using a travel platform”,
6 Japan Agricultural Research Quarterly, Vol. 48 No. 3, pp. 307-316.
7
8 Henten, E. J. V., Van’t Slot, D. A. , Hol, C. W. J. and Van Willigenburg, L. G. (2009),
9 “Optimal manipulator design for a cucumber harvesting robot” Computers and Electronics
10 in Agriculture, Vol. 65 No. 2, pp. 247-257.
11
12
13 Hiwin Corporation accessed in February 2020 https://www.hiwin.com/articulated-
14 robots.html).
15
16 Kapach, K., Barnea, E., Mairon, R., Edan, Y. and Shahar, O. B. (2012), “Computer vision for
17 fruit harvesting robots-state of the art and challenges ahead”, International Journal of
18
19 Computational Vision and Robotics, Vol. 3 No.1-2, pp. 4-34.
20
21
Li, P., Lee, S. H. and Hsu, H. Y. (2011), “Review on fruit harvesting method for potential use
Ind

22 of automatic fruit harvesting systems”, Procedia Engineering, Vol. 23 No. , pp. 351-366.
23
24 Mu L., Cui, G.,Liu, Y.,Cui, Y., Fu, L. and Gejima, Y. (2020), “Design and simulation of an
25 integrated end-effector for picking kiwifruit by robot”, Information Processing in
us

26 Agriculture, Vol. 7 No. 58-71.

27
28 Sarig, Y. (1993), “Robotics of fruit harvesting: A state-of-the-art review”, Journal of
tri

29 Agricultural Engineering Research, Vol. 54, pp. 265-280.

30
31 Scarfe, A. J., Flemmer, R. C., Bakker, H. H. and Flemmer, C. L. (2009) "Development of an
al

32
autonomous kiwifruit picking robot" 4th International Conference on Autonomous Robots
33
34 and Agents, Wellington, 2009, pp. 380-384.
Ro

35
36
Silwal, A., Davidson, J., Karkee, M., Mo, C., Davidson, J. R., Zhang, Q. and Lewis, K.
37 (2017), “Design, integration, and field evaluation of a robotic apple harvester”, Journal of
38 Field Robotics, Vol. 34 No. 6, pp. 1140-1159.
bo

39
40 van Henten, E. J., Hemming, J., van Tuijl, B. A. J., Kornet, J., Meuleman, J., Bontsema, J.
41 and van Os, E. A. (2002), “An autonomous robot for harvesting cucumbers in green-
t

42
houses”, Autonomous Robots, Vol. 13 No. 3, pp. 241-258.
43
44 Wang, L., Zhao, B., Fan, J., Hu, X., Wei, S., Li, Y., Zhou, Q. and Wei, C. (2017),
45
46 “Development of a tomato harvesting robot used in green-house”, International Journal of
47 Agricultural and Biological Engineering, Vol. 10, pp.140-149.
48
49 Williams, H., Jones, M. H., Nejati, M., Bell, J., Penhall, N., Seok, H. A., Lim, J., MacDonald,
50 B., Seabright, M., Barnett, J., Duke, M., and Scarfe, A. (2019), “Robotic kiwifruit
51 harvesting using machine vision, convolution neural networks, and robotic arms”,
52
Biosystems Engineering, Vol. 181 No. pp. 140-156.
53
54
55
56
57
58
59
60

17