Top Navigation Bar

Article Request Page ASABE Journal Article

Design and Testing of a Kiwifruit Harvester End-Effector

S. S. Graham, W. Zong, J. Feng, S. Tang

Published in Transactions of the ASABE 61(1): 45-51 (doi: 10.13031/trans.12361). Copyright 2018 American Society of Agricultural and Biological Engineers.

Submitted for review in March 2017 as manuscript number MS 12361; approved for publication by the Machinery Systems Community of ASABE in October 2017.

The authors are Sean S. Graham, Graduate Student, Wangyuan Zong, Professor, Jun Feng, Graduate Student, and Shengping Tang, Graduate Student, College of Engineering, Huazhong Agricultural University, Wuhan, Hubei, China. Corresponding author: Wangyuan Zong, No. 1 Shizishan Street, Hongshan District, Wuhan, Hubei 430070, China; phone: +86-131-006-26908; e-mail:

Abstract.á Mechanisms to aid fruit harvesting are undergoing constant development with increasing available technologies. However, fruits grown on vines, such as kiwifruit, have complex tree architectures and present difficulties in confirming design parameters. The objective of this research was to develop an end-effector for a kiwifruit harvester based on integrating the physical characteristics of the fruit, such as stem length, the space between mature fruits, and the growing environment provided by a trellised system into the design. These properties contribute to developing a mechanism that is lightweight, battery operated, and requires only one translational joint for positioning. Scissor cutting actuated by a linear solenoid is used to provide the required torque of 1.38 Nm to completely sever Hayward variety kiwifruit at the stem using a curved blade with a 20░ relief angle. The cutting of the stem is actuated by a force sensor located on the device that enables cutting at less than 10 N, preventing premature detachment of the fruit and damage to the vine. The cutting time was measured to be 0.1 s ▒0.03 s per cut. This end-effector design adds to the body of research aimed at developing a fully mechanized kiwifruit harvester.

Keywords.Detachment force, End-effector, Fruit harvester, Kiwifruit, Linear solenoid.

Kiwifruit is commercially produced in many countries around the world, ranging from Italy, which is the largest producer at 385,000 metric tons per year, to Japan, which produces 28,000 metric tons (FAO, 2014). Despite the relatively high production rates, there are no commercially available machines specialized for kiwifruit harvesting. Kiwifruit harvesting appears to be done exclusively by hand. A robust mechanical kiwifruit harvester would assist in reducing the labor requirement associated with the manual procedure, making the fruit an even more viable agricultural consideration for farmers.

Kiwifruit is an edible berry that grows on a woody vine (Actinidia deliciosa). The vine grows vigorously in subtropical climates. The fruit is normally harvested by snapping the stem at the abscission layer at the base of the fruit (Mainland and Fisk, 2006). Because kiwifruit grows on a vine, inertial shakers, such as the one suggested by Polat et al. (2007), are not feasible. Mechanisms to aid harvesting must try to optimize the balance between robustness, speed, and cost, as noted by Silwal et al. (2016). To this end, any fruit characteristics or environmental factors that can aid mechanical harvesting should be explored. This could lead to a reduction in the number of actuators and control mechanisms needed to achieve the design objective.

Literature Review

One kiwifruit harvester design used linear positioners in conjunction with a four degree-of-freedom (DOF) robot having two rotational joints and two translational joints (Gao et al., 2013). However, challenges with providing adequate space close to the vines was foreseen, as the bulk of the robot was mounted close to the vines. Supporting the weight of the suspended robot in the orchard environment was another challenge. While no other studies were found that specifically targeted the mechanization of kiwifruit harvesting, studies on other fruit and vegetable mechanisms can provide substantial and relevant information toward developing mechanisms for kiwifruit harvesting. Research by Wang et al. (2016) to design an end-effector for tomato harvesting categorized end-effectors based on function, including sucker type, shear type, spiral twisting type, and multi-fingered type. Combination types have also been applied to fruit and vegetable harvesting, such as the vacuum/suction type with multi-fingered grippers developed by Ling et al. (2004). This method initially positioned the sucker using a vacuum pump and then held the object in place using a suction device powered by a vacuum pump. To improve the grip, under-actuated fingers were used to hold the object in place. Some challenges noted by the authors included managing the overall weight of the fingers, which significantly affected the magnitude of torsion necessary to prevent rotation about the joints. In addition, field testing indicated that although the under-actuated fingers were inconsistent in their operation, increasing the actuation would lead to an increased payload. For a cucumber harvesting robot developed by Tang et al. (2009), positioning the end-effector required seven actuators and another within the end-effector for cutting. Silwal et al. (2016) used a seven DOF robot programmed to mimic the twisting motion of a human when picking an apple. The end-effector could be classified as a gripper based on function and was under-actuated to provide shape adaptation. In both studies, every DOF was assigned an actuator, resulting in an increase in the power requirement of the overall machine. To test the mechanism, the plants were grown on a custom-built trellised structure in a laboratory environment so that the produce would grow in the desired location. This is unlikely to be the case on commercial farms. A cucumber de-leafing robot was designed and developed by Van Henten et al. (2006) to locate the stalk of the cucumber vine and from that location move up the stalk and de-leaf the cucumber vine automatically. The problem with this robot was that field tests indicated it was actually 35 times slower than a human.

Scissor cutting employs a method of dissection that is the result of applying concentrated energy to a small area. Two blades rotate around a common pivot, and the point at which they cross paths is where the energy from the input source is concentrated, thereby effecting the destruction of bonds in the target material. This concentration of energy into a small spot provides a controlled cutting area, in contrast to sawing (Callaghan et al., 2011). Knife blade cutting, which is essentially a single-blade scissors, also concentrates the energy between the blade and the target material in a small area, and this method is used in biological precision industries because the nature of biological materials varies greatly. Research has shown that the blade inclination angle, steel grade, and sharpening angle have significant effects on cutting performance. Curved blades or blades inclined with respect to the handle are recommended (Marsot et al., 2005). Methods for calculating the cutting force of fibrous materials are subject to the materialsĺ rheological properties. Research on litchi stalks (Liu et al., 2012) showed that the relief angle (blade angle) affected the cutting force; however, the cutting angle showed no effect on cutting force. Wilhoit et al. (1989) investigated the effect of high-speed cutting on broccoli stalks and concluded that the smallest force required for cutting was at the highest speed, and the cross-section of the stalk significantly affected the cutting force. In both studies concerning cutting force, the force was applied directly above the stalks.

Mahvash et al. (2008) modeled the cutting forces of scissors using a fracture mechanics approach. Their approach quantified the change in elastic potential energy stored in a material (dWa) before and after cutting and the irreversible work of fracture done (dU). This irreversible work of fracture was calculated based on work by Atkins et al. (2005). Mahvash et al. (2008) assumed that energy was conserved before and after the material was cut; hence, the total work done by the scissors was the summation of both prior mentioned quantities. They developed an equation that related the torque required to sever the material with its fracture toughness and the change in angle between the handles of the scissors from the start of the cut to the completion of the cut. In the case of kiwifruit stems, the fracture toughness of the stem is unknown.

Badadal Raghavendra et al. (2015) investigated the forces required to completely sever different types of plant stalks, such as custard apples, mangoes, and guavas, but not kiwifruit. Their method applied varying forces to the handle of a pruning shear, much like when the handles of scissors are squeezed together, trapping the subject material between the blades. In their research, the force was applied manually using measured weights. The highest torque measured was 1.653 Nm, which was used as a design parameter to test whether this torque was sufficient to completely cut the kiwifruit stem.


The objective of this research was to develop a kiwifruit end-effector requiring a low number of actuators that can completely sever mature Hayward variety kiwifruit at the stem. Specific objectives were:


The end-effector design proposed in this research (fig. 1a) will be a part of an array, as shown in figure 1b. The hanging position and stem length of mature kiwifruits present an opportunity for a device to be able to reach the stem if initially positioned well. The height from the ground to the fruit and the distance between vertical supports provide an area under the orchard for a mobile unit to move freely. The designed mechanism must fit within the space between the top of the kiwifruit and the vine, grasping the stem of the fruit. The general design concept relies on the following specifications:

To collect information such as stem diameter and stem length, a Vernier caliper with an accuracy of 0.02 mm was used. To measure the stem length, the fruit was held stationary while still connected to the vine, and the caliper jaws were opened wide enough so that one end was touching the fruit and the other end was touching the vine. The measurement was then recorded. To measure the stem diameter, the same Vernier caliper was used to grasp the stem, and the measurement was recorded. There was no observable difference when measurements were taken at different points on the same stem.

Figure 1. (a) Single end-effector with kiwifruit, and (b) array of four end-effectors.

End-Effector Design and Operation

The end-effector uses a scissor cutting technique. The stem is trapped in a slot, as described in the previous section, and then clipped by a blade of the type commonly used in pruning shears with a relief angle of 20░. One handle of the shear is held stationary while the other handle is moved by a linear solenoid, which applies a force that rotates the handle about the pivot. To completely sever the stem, the end-effector must accomplish two objectives:

Positioning Stem for Cutting

After the stem falls into the slot, the fruit presses up against the underside of the device. The machine continues to move forward, resulting in the machine tugging on the vine because the fruit is still attached to the vine. To limit this tug and enable the actuator to make the cut, a particular resistive target value was set; this is called the activation force. The pressure between the top of the kiwifruit and the underside of the device is measured using two force-sensitive resistors (FSR, range 0.1 to 1000 N, Interlink Electronics) that are located directly where contact occurs between the top of the fruit and the device. When the target value set for the FSR is achieved on both sensors on either side of the mechanism, the stem is at the right location for cutting. The sensors are connected to a microcontroller (Mega 2560, Arduino) through two 10 kO pulldown resistors. When the fruit presses against an FSR, the resistance of the FSR decreases, which leads to an increase in current in the circuit and increases the voltage across the 10 kO resistor. This voltage is measured at the analog pins of the Arduino microcontroller. The microcontroller analog pins are read at 100 Ás, which results in 10,000 samples per second. When the activation force target is reached, the microcontroller activates the solenoid to make the cut.

FSR Calibration Apparatus

A calibration bench was set up to relate the resistivity change in the FSR sensor to the corresponding force value. A known weight of 500 g was verified on a scale, the end-effector was placed on top of this weight, and force was applied straight down through the end-effector. The weight had a flat top that properly seated the FSR sensing surfaces. The device was connected to an Arduino microcontroller with the related data acquisition software. The scale displayed the sum of the 500 g weight and the force applied via the end-effector. The values collected from the FSRs were treated in the software such that the force applied and the resulting value displayed on the scale corresponded to the values recorded by the software. Figure 2 illustrates the calibration setup.

Figure 2. FSR calibration setup: 1 = computer, 2 = microcontroller, 3 = scale, 4 = end-effector, 5 = direction of applied force, 6 = weight.

Detachment Force

Because the machine traps the fruit and tugs on the vine, it is important to know the maximum force that the machine can exert on the vine without detaching the fruit prematurely. Therefore, information was needed on the detachment force due to tugging on the vine. The detachment force was determined experimentally because there are many non-linearities related to the characteristics of kiwifruit and the transfer of energy through the fruit stem and vine. To measure the detachment force, the same two FSR sensors, as mentioned earlier, were used. An apparatus was set up as shown in figure 3. The two sensors were placed on the underside of the end-effector, a test kiwifruit still connected to the vine was identified, and the device was placed on top of the fruit. The device was pulled down vertically while the vine was sup-ported, causing an extension of the kiwifruit stem while still attached to the vine. The vine was supported to ensure that the tensile force was confined to the stem and not transferred to the vine.

Figure 3. Kiwifruit detachment force measurement: 1 = kiwifruit vine, 2 = end-effector, 3 = FSR sensor, 4 = kiwifruit, 5 = kiwifruit stem.

Cutting the Stem

To execute the stem cut, the blades are positioned within the end-effector such that when a fruit stem falls into the slot the blades will be able to cut it. Because the blade motion is controlled by a linear solenoid, the relationship between the stroke length of the solenoid and the motion of the blade was established. When the linear solenoid was energized, the actuator was pushed out in a straight line, moving the mobile blade, which rotated about the pivot. The stroke length of the solenoid should be large enough so that the induced blade motion completely cuts the fruit stem.

In figure 4a, x1 is the stroke length of the linear solenoid, Fa is the force exerted by the actuator on the mobile blade, x2 is the perpendicular distance from the actuator to the pivot, x3 is the distance from the pivot to the desired cutting location, and Fr is the reaction force created at the cutting location by the force exerted by the solenoid (Fa). The side of the pivot where the solenoid is attached is called the action side, and the side where the cutting of the stem is done is called the reaction side. The hollow outline in figure 4b shows the position of the mobile blade after a cut, where ▀ is the angle that the mobile blade moves through. While the solenoid moves linearly, the blade rotates about the pivot, sweeping an arc. Two arcs are swept, one on the action side and one on the reaction side. To ensure that the arc on the reaction side is sufficient to cut the stem, the chord related to the arc must be greater than the stem of the fruit. The equation that relates the arcs and chords is:


where L is the length of the chord swept by an arc with radius R and angle subtended ▀. Because angle ▀ will be the same on both sides of the pivot after the blade completes a cut, the following relationship can be derived:


where lr is the chord of arc swept on the reaction side. This relationship yields the distance that the blade will move to cut the stem.

Figure 4. (a) Cutting mechanism showing linear solenoid and blade and (b) blade position before and after cut.

Cutting Location and Torque

The cutting torque required for completely severing other plant stalks, such as mangoes and custard apples, was found by Badadal Raghavendra et al. (2015). In our research, the highest cutting torque (1.635 Nm) determined in that previous study was used as a design parameter to size the linear solenoid. This cutting torque was tested to see if was sufficient to cut a kiwifruit stem. This approach prevented having to re-analyze the forces acting at the blade and stalk interface, as performed in other studies; a summary of the various approaches was outlined in the Literature Review section. In the design process, some parameter values were selected based on the desired properties of the mechanism, such as the overall size and weight. Design decisions in this research included the material for the end-effector body, the force value of the linear solenoid (Fa), and the location for the cut (x3). After these specific parameter values were chosen, other related parameter values were then calculated. Acrylonitrile butadiene styrene (ABS) was chosen as the material for the body of the end-effector because it has the advantages of low weight and resistance to environmental factors such as high humidity and significantly high or low temperatures.

Linear solenoids have a predefined holding force, usually in discrete force levels. By applying the standard moment formula, which yields the resulting torque when a force is applied at a perpendicular distance, certain parameters can be set and others can be solved for. A design decision was made to select a linear solenoid capable of exerting a force of 50 N, and the perpendicular distance was found that would generate a torque of 1.635 Nm. The perpendicular distance on the action side that generated the design torque is labeled x2 in figure 4a. Based on the moment formula, x2 = 0.033 m.

Figure 5. Solenoid force test bench: 1 = scale and 2 = solenoid.

On the reaction side, another design decision was made to set the cutting location (x3). This location was chosen somewhat arbitrarily but was closer to the pivot to create a force multiplier effect. With x3 chosen to be 0.015 m, the moment formula was applied with the desired torque of 1.653 Nm, and the resulting force created on the reaction side was Fr = 110.2 N. The perpendicular force on the reaction side is more than twice that of the action side because the cutting point is closer to the pivot. Thus, the total design length of the blade from the actuation point to the cutting point was x2 + x3 = 48 mm.

Solenoid Force Measurement

To ensure that the solenoid generated the force required to produce the target torque, an experimental determination of the force was done. To measure the force exerted by the solenoid, one end of a wire of negligible mass compared to the weight of the solenoid actuator was attached to the linear solenoid in its de-energized position. The body of the solenoid was fixed in place, but the actuator was allowed unimpeded motion. The other end of the wire was connected to a scale that was also fixed in place. Figure 5 illustrates the setup. The distance between the solenoid and the scale was such that the wire was held in tension and was not sagging. The scale was then tared. When the solenoid was energized, the actuator pulled on the scale via the connecting wire. The value was read from the scale, and the force was calculated.

Results and Discussion

In a review of the literature, measurements specifically for kiwifruit, such as stem length and stem diameter, were not found. These parameters were important for the design and were therefore collected from ready-to-harvest Hayward variety kiwifruits in the Huazhong Agricultural University orchard. The tree architecture and growing environment were also reviewed, as they played a significant role in confirming the design parameters. In a trellised kiwifruit orchard, the kiwifruit trees are planted in a specific pattern. One such pattern is shown in figure 6. Columns are placed in a specific layout to support the vines while providing an area for the vines to spread. Wire mesh between adjacent columns supports the vines. Most mature fruits hang down between trees, as shown in figure 6. The spacing between trees is approximately 1.8 m, and the average height of the fruit above the ground in a level kiwifruit orchard is approximately 1.2 m.

The distance from the ground to the base of mature fruits hanging from the vine varies between fruits; however, according to orchard caretakers, this height variance can be kept to less than 100 mm by pruning and repositioning the vines as they gradually spread throughout the orchard. The stem length of Hayward variety kiwifruit ranges from 40 to 70 mm, with a median of 54 mm. This length provides enough space for an appropriately sized mechanism to comfortably grasp the fruit by the stem. This factor was used to size the part of the end-effector that was intended to interface with the stem.

Figure 6. Kiwifruit tree with fruits.

Table 1 shows data collected from 18 harvest-ready kiwifruits. The median is irrelevant as a design parameter because the device has to handle a variety of fruit sizes. Therefore, the maximum and minimum values were important for determining the design parameters. The stem length limits the maximum height of the part of the machine that operates between the top of the fruit and the vine; this is the location for trapping the fruits. Based on the results, this distance can be no greater than 40 mm or else the machine will not be able to comfortably grasp the fruit and cut the stem. For stem length, 78% of the data fell within one standard deviation of the mean, and 100% of the data fell within two standard deviations of the mean, likely indicating a normal distribution. For stem diameter, all the data were within ▒0.02 mm of 2.12ámm, except for an outlier of 4 mm. This outlier cannot be ignored; in fact, it changed the design value of the slot opening. If the design is made for a lower value and an outlier such as this is encountered during normal operation, it would create a major problem. Because there is no impediment to increasing the slot opening, the prudent decision was to design for the outlier rather than ignore it.

Table 2 shows detachment force values measured for nine kiwifruits. An analysis of variance (ANOVA) was performed on the univariate dataset. The range was 5.8 N, with two modes of 14.7 N and 15.8 N and a mean of 13.4 N with a standard deviation of 2.5 N. However, the key parameter necessary for the design was the minimum value of 10 N. This parameter limits the maximum target value for the activation force. Setting the activation force to less than 10 N ensured that the stem was cut before it could break due to the end-effector tugging on the vine.

The solenoid holding force was measured 15 times with the same solenoid to determine the force that the solenoid was capable of providing. The mean value was 15.9 N with a standard deviation of 3.1 N. A device to generate a target torque of 1.653 Nm was designed. The distance of the actuation force from the pivot and the location for the cut were initially designed to be 33 and 15 mm, respectively, for the activation force generated using a 50 N linear solenoid. This would generate a 110.2 N cutting force at the cutting location. However, a 50 N linear solenoid was not obtained for the prototype, so the actuation point and cutting location were altered to ensure the same approximate target torque of 1.653 Nm.

A single end-effector was constructed based on the data collected, as shown in figure 7, with the following specifications:

Figure 7. Prototype of end-effector.

The difference between this method and other methods for fruit harvesting is the low number of actuators used. In addition, the activation force was applied at a distance about a pivot, thereby creating a force multiplication effect. The functionality of the end-effector depends on two actuators, one for vertical positioning and the other to cut the stem of the fruit. By not having numerous actuators, the overall weight of the mechanism is reduced. Another advantage of a low number of actuators is a reduction in the power requirement. For the prototype, a 24 DC, 2600 mAh lithium ion battery was used. This battery was able to perform the cutting process more than 100 times without recharging. The open-loop control system provides some disturbance rejection. This disturbance can arise from leaves or other parts of the vine touching the force sensors, but false activation was deterred because the two force sensors are located opposite to each other, and the control system was programmed to expect the same value from both sensors. Both force values are compared to the target force and must exceed the dead-band value programmed into the microcontroller before actuation.

Using a 29 fps camera, a sample cut was recorded. Inspection of the images showed that the cut took approximately 0.1 s ▒0.03 s. The location of the actuating force when the stem was completely severed was 87 mm from the pivot. The cutting location on the reaction side was 31 mm from the pivot. Based on the actuation point used in the prototype, the torque required to cut the stem was 1.38 Nm, which indicates that the cutting torque for kiwifruit stems is similar to the cutting torque for other plant stalks, as measured by Badadal Raghavendra et al. (2015). From the initial calculations, the total length from the actuation point to the cutting point could have been 48 mm. However, because a linear solenoid generating a force of 50 N was not obtained, a solenoid with a lower force was used in the prototype, resulting in the total length increase to 124 mm. This indicates that the size of the end-effector could be reduced by using a solenoid with a higher force rating.

According to the measurements of stem diameter, most of the kiwifruit stems were within a measurement error of ▒0.02 mm, so the variance in detachment force may be due to the various sizes of the vines to which the fruit stems were attached. Further studies need to be conducted to test this hypothesis. The detachment force results indicate that the activation force must be less than 10 N; otherwise, the fruit may separate from the vine before the end-effector can cut the stem. Using the locations chosen in this research for actuation and cutting, a stem of not more than 8.5 mm diameter could be completely severed. This is a favorable result because the maximum stem diameter observed in the measurement data was 4 mm. Fruit stems are living material, and therefore their hardness and other physical characteristics may change with growth, so further studies are needed to determine if these results for Hayward variety kiwifruit can be applied to other varieties of kiwifruit and possibly other fruits.


A lightweight electromechanical end-effector that provides sufficient cutting force is an important step toward creating fruit harvesting machines with improved mobility in orchard environments. Data were collected relevant to designing an end-effector, such as the stem length and stem diameter. A novel design approach was taken that imple-mented fruit stem cutting using a linear solenoid-actuated scissor motion. Each stem cutting device uses two actuators, one for vertical positioning and the other for cutting the fruit stem. The stem cut is triggered by force-sensitive resistors that also determine when a stem is in the right location to be severed, preventing false actuation of the device due to leaves, vines, etc. The mechanism speed was 0.1 s per cut with a cutting torque of approximately 1.38 Nm. This torque is sufficient to sever the stems of ready-to-harvest Hayward variety kiwifruits. Based on measurements, a fruit may detach from the vine prematurely if the device tugs on it with a force greater than 10 N; therefore, the activation force must be below this value. The prototype was powered by a 2600ámAh lithium ion battery and was able to perform more than 100 cuts without recharging.


Atkins, A. G., & Xu, X. (2005). Slicing of soft flexible solids with industrial applications. Intl. J. Mech. Sci., 47(4-5), 479-492.

Badadal Raghavendra, R., & Naik S., B. (2015). Experimental determination of cutting force required for severing fruit stalks. Intl. J. Innov. Res. Sci. Tech., 1(8), 154-157.

Callaghan, D., Rajan, G., McGrath, M., Coyle, E., & Semenova, Y. (2011). Investigation and experimental measurement of scissor blade cutting forces using fiber Bragg grating sensors. Smart Material Struct., 20(10), article 105004.

FAO. (2014). FAOSTAT. Rome, Italy: United Nations FAO. Retrieved from

Gao, H., Wang, H., & Chen, J. (2013). Research and design of kiwi fruit harvesting robot. J. Agric. Mech. Res., 2013(2), 74-76.

Ling, P. P., Ehsani, R., Ting, K. C., Chi, Y.-T., Ramalingam, N., Klingman, M. H., & Draper, C. (2004). Sensing and end-effector for a robotic tomato harvester. ASAE Paper No. 043088. St. Joseph, MI: ASAE.

Liu, T. H., Liang, Z. H., & Guo, J. D. (2012). Experimental comparison of litchi fruit stalk cutting force. Appl. Eng. Agric., 28(2), 297-302.

Mahvash, M. M., Voo, L., Kim, D., Jeung, K., Wainer, J., & Okamura, A. M. (2008). Modeling the forces of cutting with scissors. IEEE Trans. Biomed. Eng., 55(3), 848-856.

Mainland, C. M., & Fisk, C. (2006). Kiwifruit. Horticultural Information Leaflet 208. Raleigh, NC: North Carolina State University Extension.

Marsot, J., Claudon, L., & Jacqmin, M. (2005). Assessment of knife sharpness by means of a cutting force measuring system. Appl. Ergon., 38(1), 83-39.

Polat, R., Ibrahim, G., Guner, M. D., Erdogan, D., & Bilim, H. C. (2007). Mechanical harvesting of pistachio nuts. J. Food Eng., 79(4), 1131-1135.

Silwal, A., Davidson, J., Karkee, M., Mo, C., Zhang, Q., & Lews, K. (2016). Effort toward robotic apple harvesting in Washington State. ASABE Paper No. 162460869. St. Joseph, MI: ASABE.

Tang, X., Zhang, T., Ling, L., Dan, X., & Chen, Y. (2009). A new robot system for harvesting cucumber. ASABE Paper No. 096463. St. Joseph, MI: ASABE.

Van Henten, E. J., Van Tuijl, B. A., Hoogakker, G.-J., Van Der Weerd, M. J., Hemming, J., Kornet, G., & Bontsema, J. (2006). An autonomous robot for de-leafing cucumber plants grown in a high-wire cultivation system. Biosyst. Eng., 94(3), 317-323.

Wang, G., Yu, Y., & Feng, Q. (2016). Design of end-effector for tomato robotic harvesting. IFAC-PapersOnLine, 49(16), 190-193.

Wilhoit, J. H., Schertz, L. M., Vaughan, D. H., & Byler, R. K. (1989). Measuring broccoli stalk cutting forces. Appl. Eng. Agric., 5(3), 324-328.