!EVALUATION 
AND VALIDATION 
OF SOFT ROBOTIC END EFFECTOR
S FOR PRODUCE 
HARVESTING
  By Zachar
y F.
 Dutcher
                  A THESIS
 Submitted to 
 Michigan State University
 in partial fulfillment of the requirements
 for the degree of
 Mechanical Engin
eering
ÐMaster of Science
 2018ABSTRACT
  EVALUATION 
AND VALIDATION 
OF SOFT ROBOTIC END EFFECTORS FOR PRODUCE 
HARVESTING
  By Zachary F. Dutcher
 Global population is expected to exceed 9 billion people by 2050 which will 
require
 a 70% 
increase in net global food production
. 75% of global farm holding
s are
 considered small at 2.5 
acres or less
. While contemporary industrial farming 
gains efficiency
 through increased 
mechanization, it comes 
with significant
 environment
al costs
. Industrial farming
 practices such 
as frequent 
tillage, monocropping
 and use of 
synthetic fertilizers, 
pesticide
s and herbicide
s are 
unsustainable practices that will continue to degrade the
 surrounding ecosystem.  Small scale 
farms present an opportunity to utilize regenerative farming practices; however, they are 
potentially challenging 
to scale up 
and expensive to automate with conventional automation 
solutions. In this thesis, 
soft roboti
c end effectors
 are explored as a potential means of 
harvesting 
on regenerative 
(as well as conventional) 
farms. Three end effector designs are testing for 
parameters including grasp variability, grasp effectiveness and 
real
-world
 simulation on apple 
orchards at Michigan State University. 
Apple harvesting metrics including detachment force, 
diameter and weight have been collected for one hundred early harvest Spartan
-Macintosh 
variety apples. 
Results of this evaluation show promise for
 the application of these low
-cost 
technologies; however, much work is needed before a 
complete and 
viable soft robotic 
harvesting system is 
available
.!iii This thesis is dedicated to my parents, Kurt and Denise, and my two sisters, Alex and Maddie. 
Thank you for believing in me
.  !iv ACKNOWLEDGEMENTS
     I am 
grateful
 to the many individuals who have help and encouraged me during this project. 
I would especially like
 to thank:
 ¥!Dr.
 Ronald
 Averill
, for his 
guidance
, feedback, 
support
, and
 advice
. Without you 
none of this would have been possible.
!¥!My committee: 
Dr.
 Tamara Reid
 Bush, Dr.
 Ranjan
 Mukherjee
, and Dr. 
Changyong 
Cao.
!¥!Phil Hill, for his generosity in 
allowing me to utilize his workshop
.!¥!William Chase, for 
allowing me to harvest apples in the orchard.
!¥!Mechanical Engineering main
 office staff
!¥!Engineering machine shop staff, 
particularly
 Roy Bailiff and 
Mike K
oschmider
.!!!!!v   LIST OF TABLES
ÉÉÉ...ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.. vii LIST OF FIGURES
ÉÉ...ÉÉÉÉÉÉÉÉÉÉÉÉÉ...ÉÉÉÉÉÉÉÉÉÉÉ. vii
i KEY TO ABBREVIATIONS
ÉÉÉÉÉÉ...ÉÉÉÉÉÉÉÉÉÉÉ...ÉÉÉÉÉ.. .x Chapter 1
: IntroductionÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ
...ÉÉÉÉÉÉÉÉ....1 1.1!Small- holder farms / 
Regenerative farming systems
ÉÉÉÉÉÉÉÉÉÉÉÉ....1 1.1.1!Contemporary
 State of Agriculture in America
ÉÉ.ÉÉÉÉÉÉÉÉÉÉ...1  1.1.2!Farming Structures
ÉÉÉÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉÉÉÉÉÉÉ...3 1.1.3!Labor/harvesting
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.. 6 1.2!Robotic effectors
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ... 7 1.2.1!End Effectors
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ... 7 1.2.2!Contemporary R
obotic H
arvesting
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.. 8 1.2.3!Soft robotic end effectors
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.11 1.2.3.1!PneuNets
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.ÉÉÉ 13 1.2.3.2!FOAMÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉ 14 1.2.3.3!Optimal Gripper Design
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ..ÉÉÉÉ 14 1.3!Proposed Solution
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ...ÉÉÉÉ.. 15  Chapter 2
: Design and ComponentsÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ
ÉÉÉÉ...É... 17 2.1 Optimal Gripper 
Design
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.... 17 2.2 FOAM Magic
-Ball
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ..ÉÉÉÉ.. 24 2.3 PneuNets Design
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ..ÉÉÉÉÉÉ. 26  2.4 Cost and Material Selection
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.. 29 Chapter 3: 
Manufacturing Methods
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉÉ...É... 31 3.1 Optimal Gripper Design
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ 31 3.1.1 Actuator Fabrication
ÉÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉÉÉ...ÉÉÉÉÉ 31 3.1.2 3D 
Print
ing
 Components
ÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉÉÉÉÉÉÉÉ 32 3.1.3 Gripper Finger Casting
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉ...ÉÉ.ÉÉÉÉÉ 33 3.1.4 Assembly
ÉÉÉÉÉÉÉÉÉÉÉÉÉ...ÉÉÉÉÉÉ.ÉÉÉÉÉÉÉ. 35  3.2 PneuNets Design
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ..É.ÉÉÉÉÉÉÉÉ... 37 3.3 Magic
-Ball Design
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ 40 Chapter 4: 
Testing Methods
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ... 43 4.1 Weighted Grasping Variability
ÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉ...ÉÉÉÉÉÉ. 43 4.2 Grasp
 Effectiveness
ÉÉÉÉÉÉÉÉÉÉÉÉ.É........ÉÉÉÉÉÉÉÉÉÉ. 44 4.3 Real World Evaluation
ÉÉÉÉÉÉÉÉÉÉ....ÉÉÉÉÉÉÉÉÉÉÉÉÉ.. 44 Chapter 5: 
Results and Discussion
ÉÉÉÉÉÉÉ.ÉÉÉÉÉÉÉÉ...ÉÉÉÉÉÉ. 46  5.1 Grasping Variability Results
ÉÉÉÉÉÉÉÉÉ.ÉÉÉÉÉÉ...ÉÉÉÉÉÉ. 46 TABLE OF 
CONTENTS
 !vi 5.2 Grasping Effectiveness
ÉÉÉÉÉÉÉÉÉÉÉ...ÉÉÉÉÉÉÉÉÉÉÉÉ.. 48 5.3 Real World Evaluation Results
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉ.ÉÉ...ÉÉÉÉ. 52 Chapter 6: 
Conclusion and Future work
ÉÉÉÉÉÉÉÉÉÉÉÉÉ.ÉÉÉÉÉ...É 58 6.1 Conclusion
ÉÉÉÉÉÉÉÉÉÉÉÉÉ...ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ. 58 6.2 Future work
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ... 58  BIBLOGRAPHYÉÉÉÉÉÉÉÉÉÉÉÉÉ...
ÉÉÉÉÉÉÉÉÉÉÉÉÉÉÉ 60   !vii
 LIST OF TABLES
   Table 1. Cost of materials and manufacturing.
 ............................................................................. 29!Table 2. Results from the weighted grasping variability test performed on the optimal gripper 
and PneuNets designs.
 .......................................................................................................... 47  Table 3. Trial results for grasp effectiveness during drop testing.
 ................................................ 48!Table 4. Results of Apple
 sampling.
 ............................................................................................. 55! !!viii
 LIST OF FIGURES
   Figure 1. Map showing increasing urban immigration within the United States [6].
 ..................... 2 Figure 2. Number of US farms by size in acres between 2002 and 2012 [11].
 .............................. 4 Figure 3. Fragile Crop Harvest
-Aiding Mobile Robots (FRAIL
-bots) [24].
 .................................. 8 Figure 4. Apple harvesting and infield sorting machine designed by Lu et al. [25]
. ...................... 9 Figure 5. End effector and harvesting platform [27].
 ................................................................... 10 Figure 6.
 On the Left FFRobotics End Effector harvesting system and on the right is the vacuum 
harvesting system from Abundant Robotics [28].
 ................................................................ 11  Figure 7. Timeline of nota
ble soft grippers [32].
 .......................................................................... 12 Figure 8. a) showing comparison between elastomer expansion and expansion with elastomer 
and inflexible layer. b) FEM of elastomer with
 inflating channels. c) sectional slices of 
pressurized channels [31]....................................................................................................
... 13 Figure 9. FOAM skeleton designs and their respective motions [34].
 ......................................... 14 Figure 10. Optimally designed two
-finger gripper holding a weighted cylinder [35].
 ................. 15 Figure 11. Pick and place task segmentation [36].
 ....................................................................... 16 Figure 12. A simple zigzag design FOAM actuator is shown on the top left. A four segment 
Òwater-bombÓ bellow "skeleton" is shown on the top right. Contracted zigzag FOAM 
actuator is shown on the bottom.
 .......................................................................................... 18  Figure 13. Two sections of "water
-bomb" base
 folds.
 .................................................................. 19 Figure 14. CAD model of optimal gripper.
 ................................................................................... 19 Figure 15. Fu
nctional prototype of optimal gripper design.
 ......................................................... 20 Figure 16. Hinged plate model shown left with the force balancing model shown right.
 ............ 20 Figure 17. Finite element model for 3D printed components of optimal gripper design.
 ............ 22 Figure 18. Finite element model of optimal finger from the research of Liu et al. Stress contours 
are shown in MPa [35].
 ......................................................................................................... 23  !ix Figure 19. Water
-bomb skeleton fold fr
om Li et al [34].
 ............................................................. 24 Figure 20. Magic ball fold patterned used by Li et al [34].
 .......................................................... 24 Figure 21. Functionality of the "magic
-ball" gripper can be seen as a vacuum is turned on and air 
is removed [34].
 .................................................................................................................... 25  Figure 22. Scaled 
"magic
-ball" gripper design. Skeleton shown left, skeleton with skin shown 
right.
 ...................................................................................................................................... 25  Figure 23. "Starfish
-like" design by Ilievski et al.
 ........................................................................ 26 Figure 24. Original sPN design on the left with the fPN shown on the right [29].
 ...................... 26 Figur
e 25. Four actuator configuration from Soft Robotics Inc. [38].
 .......................................... 27 Figure 26. Pneu
-Nets design unactuated.
 ...................................................................................... 28 Figure 27. Pneu
-Nets design actuated and grasping 3.75" diameter sphere.
 ................................ 28 Figure 28. Strip of w
ater-bomb base folds.
 ................................................................................... 31 Figure 
29. Single side of water
-bomb base strip...........................................................................
 32  Figure 30.
 Two-part gripper finger mold.
 ..................................................................................... 33 Figure 31. Small bubbles can be seen in the silicone material during the casting process.
 .......... 34 Figure 32. Connector mold.
 .......................................................................................................... 34 Figure 33. Three gripper fingers bonded with two triangular rings.
 ............................................. 35 Figure 34. Exposed skeleton after insertion through wafer shown on the left. On the right, the T
-shaped pin inserted through the skeleton.
 ............................................................................. 36  Figure 35. Fully assembled optimal gripper design.
 ..................................................................... 36 Figure 36. 3D printed PneuNets mold.
 ......................................................................................... 37 Figure 37. Eco
-flex 00
-10 spread on cotton sheet.
 ....................................................................... 38 Figure 38. Pneu
Nets top cast curing to cotton layer.
 .................................................................... 39 Figure 39. Fabricated PneuNets design.
 ....................................................................................... 39 Figure 40. Preliminary testing with different laser settings.
 ......................................................... 40 !x Figure 41. Top polyurethane sheet
 for magic ball design shown left, bottom sheet with seam 
trace is shown right
 ............................................................................................................... 41  Figure 42. Magic
-ball skeleton with sphere holding inner skin in place.
 ..................................... 41 Figure 43. FOAM magic
-ball design.
 ........................................................................................... 42 Figure 44. Polypropylene spheres 100, 70, 50 and 25mm diameter with attached monofilament 
tethers.
 ................................................................................................................................... 43  Figure 45. Conventional apple orchard tree shown left, compared to "high
-density" apple orchard 
tree shown right.
 .................................................................................................................... 45  Figure 46. Grasping scheme for the optimal gripper shown left, with caging scheme of the 
PneuNets design shown right.
 ............................................................................................... 46  Figure 47. 
Acceleration graph for 5 cm displacement optimal gripper test.
 ................................. 49 Figure 48. Acceleration graph for 10 cm displacement optimal gripper test.
 ............................... 49 Figure 49. Acceleration graph for 15 cm displacement optimal gripper test.
 ............................... 50 Figure 50
. Acceleration graph for 5 cm displacement PneuNets test.
 .......................................... 50 Figure 51. Acceleration graph for 10 cm displacement PneuNets test.
 ........................................ 51 Figure 52. Acceleration graph for 15 cm displacement PneuNets test.
 ........................................ 51 Figure 53. Apple arrangemen
t on conventionally trimmed orchard trees.
 ................................... 52 Figure 54. End effectors grasping isolated apples from below.
 .................................................... 53 Figure 55. End effectors grasping apples in cluster configuration.
 .............................................. 54 Figure 56. Normal distribution curve for detachment
 force and probability density of Spartan 
apples.
 ................................................................................................................................... 55  Figure 57.Normal distribution curve for diameter and probability density of Spartan apples.
 .... 55 Figure 58. Normal distribution curve for weight and probability density of Spartan apples.
 ...... 56 Figure 59. Detachment force versus bending angle during apple harvesting.
 .............................. 56 Figure 60. Free body diagram of apple detachment method proposed by Li et
 al. [39].
 .............. 57  !xi KEY TO ABBREVIATIONS
   !" - Total contraction function
 #$%&'%&
" - Net force output
 #(" Ð Skeleton
Õs e
lastic force
 L0 - Half void 
opening length
 ks Ð Void wall stiffness
 N - Number of units 
 D - Wall length
 µ - Substitution term, 
 S0 - Half of the arc length from the original parabolic approximation
 h0 - Measured depth of the parabolic approximation before contraction
      1 Chapter 1
: Introduction
 The aim of this research is to evaluate the potential for the use of 
a soft robotic 
effector
 that is
 low
-cost, energy efficient,
 simple
 to manufacture and
 appropriate 
for use by small
-holder 
sustainable 
farms
 for produce harvesting
. This 
chapter discusses
 the motivation for this project
, including
 the 
increasing 
necessity of sustainable farms
 and
 food production
, the issue of food
 security and 
the role of 
labor shortages. 
A review is provided of 
the 
recen
t developments in low
-cost, simple
-to-manufacture soft actuator systems. Soft robotic effectors are then presented as a 
potential 
solution for increased growth of sustainable agriculture systems. Apple harvesting in 
Michigan is the focus for this thesis; h
owever, the technology could easily be applied to a variety 
of produce. 
 1.1
!Small
-holder farms / 
Regenerative
 farming system
 1.1.1!Contemporary State of Agriculture in America
 With global population on the rise and a smaller number of farms, modern agriculture has many 
challenges to face. By 2050 the global population is expec
ted to exceed 9 billion people. Global 
food production would need to increase roughly 70 percent to accommod
ate the influx of 
consumers 
[1]. The United States will not
 see the same population influx
 as developing 
countries
. Nevertheless
, estimated 
U.S. 
population growth
 of 98.1 million
 from 2014 
- 2050 translates to 
a significant 
inc
rease
 in agricultural production 
needed 
to support the American 
consumer
 [3]. The increased strain on the American agricultural complex will have two distinct 
effects. First,
 an increase in population also means an increase in urbanization. Americans are 
moving away from rural areas and into cities and urban regions. As of 2016, 81.78 percent of the 
population in the United States is living in 
urban areas 
with an estimated 87.4 
percent urban 
 2 occupation by 2050 
[4] [5].   Figure
 1. Map showing increasing urban 
immigration within the United States 
[6]
. While 
urbanization has had an effect on the cost and availability of farm land, it has been shown 
that urban sprawl has been on the decline since 2012 
[7]. A declining rate of urban sprawl and a 
conscious effort to design appropriate urban spaces benefits farmers through rural land prices 
and conservation, but 
does nothing to stem
 the flow o
f agricultural
 laborers, both skill
ed and 
unskilled, 
from rural to 
urban areas. 
  3 Second,
 the degradation of natural resources and remaining farmland through unsustainable 
farming practices
 is a growing concern
. Estimations suggest that, with current farming practices, 
top soil erosion is oc
curring at a rate of 
1.73 billion tons per year (measured in 2007), and while 
this is a decrease from the 
estimated 
3.06 billion tons
 per year
 in 1982
, it is still representative of 
practices that cannot sustain the growing population
 [8]. Contemporary monoculture farms, 
while highly 
labor efficient
, degrade
 the ecosystem through the use of agrochemicals, 
unsustainable tillage practices
, minimized 
diversity of 
plant and soil life
, and deforestation. 
Transitioning to more 
regener
ative
 farming practices will allow for more
 harmonious and 
sustainable agricultural production method
s to meet the demands of the growing generations. 
According to a 
FAO, study, 73% of 
the world
Õs farm holdings are small
-holders, consisting of 
2.5 acres
 or less [2]. With such a high number of farming considered Òsmall
-scaleÓ it will be 
important to build a foundation of sustainability through technology that will allow these small
-scale farms to flourish. 
 1.1.2!Farming Structures
  Current agricultural practices can be separated into two categories: industrial and sustainable 
agriculture. Industrial, as the name suggests, represents commercial farming operations. 
Industrial agriculture in general
ly known for 
utilizing technique
s such as 
monocultures, tillage 
fields, simple crop rotations, herbicide
s, pesticides
, synthetic fertilizers
 and genetically modified 
(GMO) crop
 varietie
s [9]. Industrial farms can
 also 
be 
defined as farms 
which generate positive 
net income and require an operator on a full time basis 
[10]. As of 2012, 
61.9% of farms in the 
United States contained over 50 acres o
f land
 and produced more than $54,000 per farm
, with 
larger farms
 over 2,000 acres
 producing 1.5 million dollars
 [11].   4  Figure
 2. Number of
 US farms by size in acres between 2002 and 2012 
[11]
.!The term Òsustainable agricultureÓ was utilized after the World Commission on International 
Environment and Development created the report 
Our Common Future 
in the late 1980s 
[12]. The report utilized 
Sustainable Development 
as a
 key concept and the term 
sustainable 
was adopted by those in agriculture using the term 
alternative
-agriculture 
to define methods of 
farming better suited for the global socio
-environmental enhancement 
[13]. Sustainable 
agricu
lture is used mainly as an umbrella term to encompass a variety of other styles of farming. 
The main methods
 or schools of thought
 that will be discussed include: Organic farming, 
agroecology, permaculture, and regenerative agriculture. 
 Organic Farming, according to the 1995 
USDA National Organic Standards Boards, can 
be defined as the following, Ò
Organic agriculture is an ecological production management 
system that promotes and enhances biodiversity, biological cycles and soil biolog
ical activity. It 
is based on minimal use of off
-farm inputs and on management practices that restore, maintain 
and enhance ecological harmonyÓ 
[14]. Organic production systems can include the use of 
cover 
 5 Òcrops, various
 manures and crop rotations to fertilize the soil, maximize biological activity and
 maintain long
-term soil healthÓ 
[14]. Pests and weeds are managed through the use of biological 
control methods. Organic farming seeks to reduc
e the amount of off
-farm inputs and completely 
eliminate the use of synthetic chemicals such as pesticides, herbicides, hormones and antibiotics.
 Agroecology attempts to connect ecology with agriculture mainly with respect to the 
developing world. ÒAgroeco
logy continues to have ecology as its basis and a focus on farm, 
village level, and bioregional systems. Over time, it has sought to include broader and more 
interdisciplinary concerns, such as analyses of how land tenure, market and trade structures, and 
social inequalitie
s interact with farming systemsÓ 
[15] [16].  Permaculture
 was developed by Australian Bill Mollison
 and his student David 
Holmgren. 
 It 
Òcombines ecosystems
-based models with landscape design processes to develop 
farm
-level systems that integrate household systems with multistory and genetically diverse tree, 
shrub, and ground crops, as well as aquacultural systemsÓ 
[15]. While Agroecology utilizes 
agricultural knowledge from historical systems, Permaculture attempts to integrate small 
production systems with the 
surrounding topography and surrounding resources. 
 Regenerative Farming
 is a broader term that is defi
ned by the desire to understand how to 
regenerate not just local cropping systems, but also the surrounding value chain, including 
families, communities, landscapes and regions that inter
act with the farming community 
[15]. In 2017, the Carbon Underground and 
Regenerative Agriculture Initiative at the California State 
University partnered with Unilever, General Mills, MegaFood, and many other companies to 
develop verification standards for growing food in a regenerative manner. Th
e Carbon 
Underground defines regenerative agriculture as a Ò
holistic land management practice that 
 leverages the power of photosynthesis in plants to close the carbon cycle, and build soil 
  6 health, crop resilience and nutrient densityÓ
 [17]. Regenerative Agriculture has recently replaced 
sustainable agriculture as the umbrella 
term to describe the above
 alternatives to industrial 
agr
iculture. Practices such as contributing to soil health and fertility, increa
sed water retention 
and safe runoff, increasing biodiversity and ecosystem health, and inverting the carbon emissions 
are all concepts deeply rooted in these schools of thought; however, it is the method by which 
these concepts are achieved that separat
es the four schools of thought.
 1.1.3!Labor/harvesting
 Nationally
, the United Sta
tes is dealing with a labor deficit. Between 2002 and 2014 the supply 
of workers available to farmers in the U.S. dropped by over 20 percent. The number of migrant 
labor has also dropp
ed by over 75 percent between 2002 and 2012
 [18]. While the available 
labor force is dropping, the subsequent generation is doing little to replace 
it. From 2002 to 2014, 
US-born farm workers offset only three percent of the de
cline in field and crop workers spurred 
by the lack of immigration 
[18]. A 2
012 study found that 27% decline in available labor cost 3.3 
billion dollars in unmet GDP growth, and 1.3 billion dollars in lost farm income 
[19]. Organic 
farms are also suffering from labor deficits
. In 2006, 32% of organic farms 
surveyed 
in 
California 
reported insufficient access to labor at some point during the 2006 growing season 
[20]. The business structure of organic farms can also limit their access to labor. While organic 
farms often pay higher wages, they do not provide benefits such as healthcare, paid time off, and 
housing options as frequently as Industrial f
arms 
[20]. Employment conditions have a strong 
correlation to retention rate among employees, with return rates rising 19% if health insurance is 
provided, 21% for paid leave, and 9% for no
-fee housing 
[21].   Increasing 
product 
diversity of farms, while an important aspect of regenerative 
agriculture, has show
n to 
have a higher
 demand for physical labor. Organic farms with one to 
 7 five crop varieties required 0.45 workers per acre, whil
e organic farms with five or more crop 
varieties required 0.82 workers per acre 
[20]. Increasing the efficiency of the available labor will 
be critical in order to support the projected population growth, while also meeting the
 defined 
goals for regenerative agriculture. 
 Automated harvesting through the use of robotics is already taking place in industrial 
agriculture; however, the scale at which these systems are developed are not always appropriate 
for small
-scale regenerative agriculture farms. In order to bridge the ga
p in profitability and 
productivity between industrial and regenerative agriculture, appropriately scaled automated 
systems must be developed to fill the void left by an insufficient labor force. The following 
section will define the prerequisite knowledge
 for robotic harvesting systems, as well as, define 
the systems that will be explored as potential solutions. 
 1.2
!Robotic effector
s  1.2.1!End Effectors
  End effectors or End
-of-Arm Tooling are devices that 
attach to the end of a robotic ar
m and 
allow for interaction with the surrounding environment 
[22]. Robotics and end effectors have a 
long history of use in industrial setting
s since the 
mid-20th century 
[23]. In general, robotics and
 end effectors have been a means to alleviate stress caused by simple repetitive tasks such as pick 
and place operations as well as ergonomically challenging situations. Robots are commonplace 
in manufacturing and industrial settings, but they are now bein
g adapted for use in agricultural 
settings. Planting, weeding, harvesting, and packing are all
 potential
 robotic applications that 
allevia
te the ergonomic stresses of agricultural
 labor
, and 
many 
other applications are possible as 
well.    8 1.2.2!Contemporary Robo
tic H
arvesting
  Produce harvesting can be described on a spectrum from a fully manual to a fully automated 
system. A variety of systems are being developed that are meant to aid in production such as the 
FRAIL
-bots, currently being developed by Vougioukas 
et al.
 [24] as a means of maximizing 
harvesting efficiency. FRAIL
-bots do this by transporting full 
loads of strawberries from the 
harvesting station to the unloading station.
   Figure
 3. Fragile Crop Harvest
-Aiding Mobile 
Robots (FRAIL
-bots)
 [24]
. Another example of partial automation or automated assistance
 harvesting
 is in the form of 
harvesting platforms for apple picking. Lu et al. 
[25] evaluated commercially available 
 9 platforms, and present their own development of the self propelled apple harvester and infield 
sorting machine.
  Figure
 4. Apple harvesting and infield sorting machine designed by Lu et al. 
[25]
. To bridge the gap between partially automated harvesting system
s and fully automated systems, 
a means of extracting the fruit or vegetable from the plant is necessary
. This will require
 an end 
effector
. Challenges in designing end effectors for 
produce harvesting include: canopy 
avoidance, collision avoidance, assessing desirability of produce, secure grasping while 
minimizing bruising
 and 
extraction of the fruit from the plant.
 There has been significant 
research into the development of an end
 effector for robotic apple harvesting. Bulanon et al. 
[26] developed an 
end effector and robotic vision system in which the a
pple is grasped at the 
peduncle as opposed to the fruit body. One advantage to this design is the l
ow profile of the end 
effector; however, accuracy require to grasp the peduncle was discussed as an issue in the 
 10 literature. 
Davidson and Silwal developed an under
-actuated end effector used in conjunction 
with a vision and base system 
[27]. !Figure
 5. End effector and harvesting platform 
[27]
. Robotics company, FFRobotics, is developing a fully autonomous harvesting system and is 
competing against Abundant Robotics to be the first to commercialize a robotic apple harvesting 
system 
[28]. The advantages of these system is their fast harvesting speeds, however, they 
require the apple orchards to be arranged in a specific manner to best utilize either platform. 
  11   Figure
 6. On the Left FFRobotics End Effector harvesting system and on the right is the vacuum harvesting system 
from Abundant Robotics 
[28]
. Contemporary end effectors discussed in this section have 
many advantages; however, many 
lack energy efficiency, are difficult to manufacture, and are too expensive for small
-scale regenerative farmers. In the following section, 
soft robotics will be discussed as a cost effective 
replacement for contemporary end
 effectors.
 1.2.3!Soft robotic end effectors
  Soft robotics are a category of 
machines created from compliant materials such as elastomers, 
polymers, hydrogels, and granules. They are driven by either pneumatics, electricity or 
chemically 
[29]. Soft robotic
s and specifically soft end effectors
 have become an active area of 
research due to their safety 
with regards to human robot interaction
, energy efficiency
, low cost
 and 
resilience 
[30]. Soft gripp
ers have gained interest as a means of produce handling for the 
reasons stated previously, as well as their ability to gently grasp the produce without causing 
bruising or 
damage
. Soft pneumatic actuators (SPAs) are a well developed category and have 
seen much progress since their development. 
Pneumatics, especially when using air as the fluid, 
offers many advantages including: rapid inflation due to the low viscosity of air, pressurized air 
is easy to control, readily available, and can be discarded thr
ough venting after use 
[29]. While 
chemically driven actuators do not require the bulky compressors and vacuums that SPAs do, the 
 12 technology needs further development before it 
can produce
 a commercially viable actuator. The 
Silicone
-ethanol elastomer presented by 
Miriyev
 et al. 
[31] utilized the reaction between 
the 
joule heating element and the ethanol voids which vaporize
s and expands the silicone matrix. 
Thi
s novel approach to actuation requires much less supporting equipment when compared to 
SPAs; however, the rapid degradation of the actuator m
akes it incompatible with the 
requirements for an agricultural end effector. The following section will describe th
e styles of 
elastomer soft actuators that will be 
considered 
as potential end effectors for use in soft robotics
  Figure
 7. Timeline of notable soft grippers 
[32]
.     13 1.2.3.1!PneuNets
  PneuNets
 developed by Ilievski et al. 
is a soft robotic design that utilize
d a network of embedded 
pneumatic channels within an elastomer body
 [33]. The repeated channels inflate and deform the 
elastomer 
like
 a balloon. 
Requiring only a single source of pressure the 
channeled elastomer is 
combined with an inflexible bottom layer. When used in conjunction, the inflexible bottom layer 
ÒdirectsÓ the inflation upwards thro
ugh the elastomer and produces the
 curvatu
re, as
 seen in 
Error! Reference source not found.
. Figure
 8. a) showing comparison between elastomer expansion and expansion with elastomer and 
inflexible layer. 
b) FEM of elastomer with inflating channels. c) sectional slices of pressurized channels [31]
. .!  14 1.2.3.2!FOAM Fluid
-driven origami
-inspired artificial muscle (
FOAM
) is a sub category of soft actuator or artificial 
muscles
 developed by
 Robert J. 
Wood
 and
 his coworkers
 [34]
. The actuator is made up of 
three 
components: a folded skeleton, a TPU flex
ible skin, and a means of connecting
 the device 
pneumatically. 
Actuation is driven by the folded skeleton, while the skin provides the means of 
contractile force. 
While the FOAM
 actuator can be driven with positive or negative pressure, 
negative pressure is safer for use in an environment where operators will be present. A variety of 
motions 
such as bending, linear contraction and torsion 
can be achieved based on the folding sche
me 
imposed on the skeleton
 (Figure 9)
.    Figure
 9. FOAM skeleton designs and their respective motions
 [34]
.  1.2.3.3!Optimal Gripper 
Design
  While the grasping strategies for the FOAM actuator are
 elegant, complex folding schemes such 
as the Muri fold require 
laser cutting
 to fold
 rigid materials
. Implementing a passive gripper 
is one solution that would requ
ire less specialized equipment. 
Liu et a
l. [35] designed a compliant 
optimal grippe
r for grasping objects of unknown size or shape. The group utilized a topology 
 15 optimization method to create thei
r two
-finger gripper mechanism
 with 
printed flexible filament. 
The two
-finger gripper use
d linear displacement to engage
 the gr
ipper ends
, as
 shown in 
Figure
.  Figure
 10. Optimally designed 
two
-finger 
gripper holding a weighted cylinder
 [35]
.  1.3
!Proposed S
olution
  As a means of increasing productivity and alleviating stress from labor shortages, it is proposed 
to explore the potential for soft robotics as a 
form 
of appropriate technology for 
regenerative 
farming
. Soft robotic actuat
ors and gripping systems offer many advantages to contemporary 
robotic effectors such as low cost, energy efficiency, operator safety, and reduced produce 
damage. 
The focus of this research will be the grasping phase of the Òpick and placeÓ task 
segmentati
on, as
 seen in 
Figure
.  16  Figure
 11. Pick and place task segmentation
 [36]
. A combination of a FOAM actuator and the two
-finger gripper system will be explored as a 
potential low
-cost end effector for use in regenerative farming produce harvesting. Linear 
contraction will be provided by a FOAM actuator and enable the grasping mechanics of the 
two-finger 
gripper sy
stem. This design will be tested for harvesting efficiency, potential for produce 
damage, manufacturing feasibility, and cost effectiveness. The proposed system will be tested 
against alternative soft actuators such as the PneuNets and an alternative FOAM 
gripper 
for comparison.
    17 Chapter 2
: Design
 and Components
 Three designs were selected for testing as potential soft robotic end effectors. 
A modified version
 of the optimal gripper design
 (Figure 10
), a FOAM gripper design that utilizes the 
magic
-ball
 origami fold pattern, and a four
-"#$%&'
 PneuNets design based on the work of Zhang et al. 
[37]. While 
many other soft robotics designs 
are 
available, these
 were chosen because 
they met the 
design requirements needed for produce harvesting on regenerative farms
. The 
primary 
design 
requirements
 considered here 
for produce harvesting 
on small
-scale regenerative farms are as 
follows:
 simple
, low
-cost, safe, and functional. Functionality refers to manipulative dexterity, 
grasp robustness, and efficiency. 
Chapter 2 will describe the thr
ee selected designs, their cost and
 material requirements
. !2.1
!Optimal gripper Design
  In the original paper by 
Liu et al. 
[35], the 
compliant 
optimal gripper is a 
two-finger 
system
 utilizing a direct current
 motor connected
 to a 
threaded rod via a bevel gear
 to provide 
displacement
. Two 
appendages are
 the minimum required by an end effector to successfully 
contain an object; however, the mate
rial selected for 
the current study
, Dragon S
kin 30, 
is less 
compliant than the thermoplastic elastomer used by Li
u et al. A 
third 
finger
 was added in the 
hopes that it would reduce
 potential
 slippage and ejection during the load phase of grasping
. A three 
finger
 arrangement would also distribute force more evenly across the surface of the 
produce being harvested. 
Displacement 
in the three 
finger 
arrangement 
was produced by a 
FOAM style linear actuator mounted in the center. Early iterations of this design u
tilized a 
simple 
zigzag
 FOAM skeleton pattern
 (Figure 
1) to produce the linear contraction; however, early 
trials 
showed 
the simple 
zigzag
 actuator
 was incapa
ble of producing the necessary force 
before
 buckling. 
  18    Figure 
12. A simple zig
zag design FOAM actuator is shown on the
 top
 left. 
A four segment 
Òwater
-bombÓ
 bellow
 "skeleton" is shown on the
 top
 right.
 Contracted zig
zag FOAM actuator is shown on the bottom.
 As seen 
in 
Figure 
1, the final iteration of the three 
fingered
 optimal gripper design used a
 bellow
ed skeleton
 design 
which was less likely to buckle due to the 3D nature of the fold pattern.
 The
 bellow
ed skeleton is created from
 a single strip of 12 Òwater
-bombÓ
 base origami folds
, Figure
, which is th
en folded in half with the seams bonded
. Material for the skin and pneumatic fittings 
are the same for the 
bellowed skeleton as in the zig
zag design. 
  19  Figure
 13. Two sections of "water
-bomb" base folds
. For the final prototype of the optimal gripper design, the actuator and the 
fingers 
are connected 
by two
 printed PLA rings with the actuator mounted between two ÒTÓ shaped PLA cross 
members.
 A wafer component was added 
to restrict the motion of the ÒTÓ 
shaped cross member 
during contraction.
 Details of the three 
fingered
 optimal gripper design can be seen 
in Figure 
.  Figure
 14. CAD model of optimal gripper
.  20  Figure 
15. Functional prototype of optimal gripper design
. Modeling of the optimal gripper design has been evaluated with FE analysis and 
through
 prior 
analytical modeling
 developed by Li 
et al.
 Actuation by the FOAM linear actuator can be 
modeled as 
a two 
hinged
 ridged beam with an opening angle of 2
!. The hinge itself can be 
shown as two cantilever springs 
[34].   Figure
 16. Hinged plate model shown left with the force balancing model shown right
. Total contraction of a linear zigzag skeleton FOAM actuator can be modeled as 
  21 !")*+,-./"01-./" (1) Where N is the number of units and D is the wall length shown in 
Figure
. Net force output for a 
linear zigzag FOAM actuator can be 
estimated 
as #$%&'%&
")*#"1#(" (2) where F
(!) is the force function and F
e(!) represents the skeleton
Õs elastic force
, and can be 
calculated 
by the equation
 #(")23401,-./" (3) 23)56&7897 (4) where 
L0 is half the void
Õs opening length shown in the force balancing model 
in Figure
. Bending stiffness of the void walls, k
s, includes the tensile modulus of the skeleton 
material, the 
width and the thickness. F(
!) can be 
predicted 
as [34]: #"):1;<6=9>$3
?@A3BC
?@AD@EF?>$3@
AEGF?9?3BC
?@>$3@
>$3@
 (5) H")IJ@KA@LMKAN@K (6) O)P0=1,=QRS=" (7) P0)N=40=JTU0=F?JVK?=WK-./XAN=WKVK (8) 40),-./"0 (9) where 
"(!) is the linear correction term necessary because the previous force function did not 
approach zero when the voids were completely closed (
! = 0). Due to an inaccurate 
approximation of the fluid volume, the correction term 
is needed
. µ is used as a substitution term, 
S0 is one half of the arc length from the original parabolic approximation, h
0 is the measured 
 22 depth of the parabolic appro
ximation before contraction, and L
0 is half the voidÕs opening length 
[34]. A simple finite element model was 
built
 using software package
 NX 10.0 for the 3D printed 
components of the optimal gripper design.
  Figure
 17. Finite element model for 3D printed components of optimal gripper design.
 The model shown in 
Figure
 was generated by fixing three arms of the triangular ring where the 
Dragon Skin 
fingers
 were to be 
mounted
. A distributed load was simulated over the T
-shaped 
pin. The three components were combined under surface
-gluing constraint.
 The resulting 
simula
tion shows that the current design will function within the 
prescribed
 load limits. This 
simulation did not take into account the complexities of layer by layer FDM printing
 which 
could produce a weaker structure.
 Finite element modeling has also been done
 for the compliant 
optimally designed fingers. The simulation was included in the paper by Liu 
et al. 
[35], and 
 23 shows 
equivalent stress contours for 
one finger under displacements of 33 and 50 millimeters 
contacting a sphere o
f 45 millimeter radius. 
Liu 
et al. measured the elastic modulus of their 
flexible filament and it was found to be 11.6 MPa. 
   Figure
 18. Finite element model of optimal finger from the research of Liu 
et al. Stress contours are shown in MPa
 [35]
.                      24 2.2
!FOAM Magic
-Ball
  The Magic
-ball Origami fold is derived from the Òwater bombÓ base folding unit. From the 
original design pr
esented in the paper by Li
u et al. [34], the only alteration made was an increase 
in 
scale to better encapsulate larger objects. Although 
fold layout is dimensionless
, the figures
 in 
the original paper 
show the original design by Li 
et al. as
 approximately 9
 centimeters in 
diameter.
  Figure
 19. Water
-bomb skeleton fold from Li
 et al [34]
.  Figure
 20. Magic bal
l fold patterned used by 
Li et al [34]
.  25  Figure
 21. Functionality of the "magic
-ball" gripper can be seen as a vacuum is
 turned on and air is removed
 [34]
.   Figure
 22. Scaled "magic
-ball" gripper design. Skeleton shown left, skeleton with skin shown right.
    26 2.3
!PneuNets Design
  As described in section 1.2.3.1, PneuNets, designed by 
Ilievski
 et al.
 [33], has been utilized in a 
variety of designs. In the original paper, 
Ilievski
 et al. 
describe
d a six limbed Òstarfish
-likeÓ 
design cast from two parts 
elastomer between one strain
-limiting
 layer. 
  Figure
 23. "Starfish
-like" design by 
Ilievski
 et al.
 This Òstarfish
-likeÓ design is capable of contracting in both directions given the symmetrical 
nature of the design. Further developments by 
Mosadegh
 et al. 
[29] in 2014 created an 
alternative design called 
fast pneu
-net (fPN) with the original design being referred to as slow or 
simple pneu
-net (sPN). 
  Figure
 24. Original sPN design on the left with the fPN shown on the right
 [29]
.  27 fPN actuators have 
various performance benefits over sPN including, significantly faster 
actuation speed (25x sPN) and reduced change in volume. Drawbacks of the fPN design include 
deflection under gravi
ty due to hinge like structure and a non
-linear bending when pressurized
 above their full bending amplitude 
[29]. Companies such as Soft
 Robotics Inc. have taken 
designs similar to fPN and commercialized them
 for pick and place automation
 [38].  Figure
 25. Four actuator configuration from Soft Robotics 
Inc.
 [38]
. For use in a regenerative farm setting, a simple pneu
-nets configuration was selected. Simplicity 
of manu
facturing, cost and compliance were factors 
affecting this decision. Based on the design 
from both 
Ilievski
 et al. 
[33] and 
Zhang
 et al. 
[37], the design
, presented in 
Figure
 and 
Figure
, uses 
a four 
finger
 grasping strategy 
composed of a 
cast of elastomer material 
adhered to a layer
 inelastic material.  An air
 line is inserted and sealed on top of the device. 
  28  Figure
 26. Pneu
-Nets design unactuated.
  Figure
 27. Pneu
-Nets design actuated and grasping 3
.75" diameter sphere.
      29 2.4
!Cost and material selection
  Table 
1. Cost of materials and manufacturing
.   Part
 Material
 quantity
 cost
 Optimal 
Gripper 
design
 Components
 Gripper finger
 Dragon S
kin 30
 3 $5.04 
 Triangular ring
 PLA
 2 $1.00 
 T shaped pin
 PLA
 2 $1.00 
 Wafer support
 PLA
 2 $1.00 
 Actuator skeleton
 polyester sheet
 480 x 40mm
 $0.17 
 Actuator skin
 polyurethane film
 ~ 140 x 120mm 
   Actuator fittings
 Nylon nut
 1 $0.07 
 Rubber washer
 1 $0.43 
 Nylon quick turn 
coupling
 1 $0.55 
   Component Total
  $9.26 
 Manufacturing
 3D printed gripper 
mold
 PLA
 1 $7.00 
 3D printed 
bonding mold
 PLA
 1 $2.00 
 Laser cutting
 - 1 $2.00 
   Manufacturing Total
 $11.00 
       Design Total
 $20.26 
 Pneu
-nets 
design
 Components
 Main body cast
 Eco
-flex 00
-30 1 $7.39 
 Inflexible layer
 Cotton sheet
 200 x 200 mm
 $0.05 
 Air hose
 1/8 inch ID 1/4 inch 
OD PVC tubing
 ~350mm
 $0.15 
     Component Total
  $7.59 
 Manufacturing
 3D printed Main 
body mold
 PLA
 1 $8.00 
     Manufacturing Total
 $8.00 
       Design Total
 $15.59 
 Magic 
Ball 
design
 Components
 Actuator Skeleton
 Polyester sheet
 230 x 460 mm
 $0.92 
 Actuator skin
 polyurethane 
film
 80424mm
2  Actuator fittings
 Nylon nut
 1 $0.07 
 Rubber washer
 1 $0.43 
 Nylon quick turn 
coupling
 1 $0.55 
   Component Total
  $1.97 
 Manufacturing
 Laser Cutting
 - 1 $13.00 
     Manufacturing Total
 $13.00 
       Design Total
 $14.97 
  For the Optimal gripper design, Dragon Skin 30 (from smooth
-on), was selected due to its high 
tensile strength and 100% modulus, as well as its 
safety once fully cured. PLA was selected as 
filament for the 3D printed components because of its low cost, b
iodegradability, and 
 30 availability. Actuator components such as the polyester skeleton material, polyurethane skin 
material, and pneumatic fittings were based on materials utilized in the research by Li
u et al. 
Eco-Flex 00
-10 was chosen due to its use in t
he original development of Pneu
-nets by Ilievs
ki 
et al. and due to is material
 safety. All of these design are not limited to the materials listed in 
Table 
1. For example, in the original paper by 
Ilievski 
et al., paper was first used as the strain
-limiting 
layer of the pneu
-nets design
. Polydimethylsiloxane
 (PDMS) was later utilized due to improved 
performance characteristics 
[33]. Materials selected for this 
study
 have been done so with cost, 
availability, and similarity to prior art in mind. 
      31 Chapter 3
: Manufacturing Methods
 Manufacturing is a critical step in the design process, and often results in 
alterations to original 
designs. 
Processes for manufacturing the three end effector
s were selected based on
 equipment
 availability, cost, and simplicity. 
A multiplicity of process
es exist for manufacturing these 
actuators and end effectors
; however, due to time constraints, 
only the methods discussed were 
evaluated
. This chapter
 details the methods that resulted in successful prototypes
, while
 still adhering to the pro
cess criteria for each of the three designs. 
Equipment such as fused deposition 
modeling 3D printers and laser cutters were available through university facilities; however, 
many 3D printing and laser cutting services exist online. 
 3.1
!Optimal Gripper D
esign
 3.1.1!Actuator Fabrication
  Actuator fabrication can be broken up into three steps: skeleton folding, initial skin sealing, and 
fitting installation and final sealing. Fitting installation a
nd final sealing is the last ste
p in the 
optimal 
gripper design
 assembly process. To start this process, a sheet of polyester film was sent 
to the laser cutter with the programmed pattern seen in 
Figure
. Cutting was done on the 
Full 
Spectrum
 hobby series CO
2 laser cutter. Vector cutting was used on power setting 20% with 
speed at 85%.
  Figure
 28. Strip of water
-bomb base folds.
 The strip is folded in half connecting the two free ends
, pre-folding on 
the patterned cuts is 
recommended before connecting free sides
. Polypropylene tape (packaging tape) is applied to all 
 32 three of the free edges
. Fold by pushing on the top and bottom of the vertical cut lines then 
compress horizontally to finish the folds.
  Figure
 29. Single side of water
-bomb base strip.
 The resulting six segment bellowed skeleton should look similar to the four segment bellowed 
skeleton in 
Figure 
1. Cut an approximately 140mm x 120mm rectangle of the 0.0015
-gauge 
polyurethane film material and seal length wise around the bellowed skeleton using an impulse 
sealer. An AIE
-305 table top im
pulse sealer was used on the number three dial setting for 
0.0015- gauge polyurethane material. 
 3.1.2!3D 
Printing Components
  Components used in this 
work 
were 3D printed by the 
MSU College 
of Engineering computing 
services using 
standard fused deposition modeling (FDM) printers with PLA filament. Printer 
model use was based on availability and included models such as the Monoprice MK11 and 
Makerbot replicator. 
Infill percentages for the molds was 15% using the honey comb pattern. 
The triangular ring, wafer and
 T-shaped pin were printed at 50% and 70% respectively; however, 
the T
-shaped pin 
and wafer 
was small
/thin enough they
 did not require any infill. 
   33 3.1.3!Gripper Finger casting
  Casting of the gripper fingers required the following
 equipment: digital scale, disposable mixing 
cup and stirring utensil, Dragon Skin 30 (part A and B), painters tape, and the two printed molds. 
Molds are inspected to ensure they are clear of extraneous filament and debris. The two
-part 
gripper mold is con
nected using the locator pins then sealed along the edge using the painters 
tape. 
    Figure
 30. Two
-part gripper finger mold
. 25 grams of both part A and B of the Dragon Skin 30 
platinum cure 
silicone is measured and 
combined in the disposable cup. After thoroughly mixing both casting parts, the mixture is 
slowly poured into one corner of the mold. Holding the disposable cup at a higher distance from 
the mold forms a thinner stream of silicone 
and can reduce the number of large bubbles in the 
cast. 
  34  Figure
 31. Small
 bubbles can be
 seen in the silicone material during the casting process.
 Alternatively, vacuum degassing in
 a vacuum chamber eliminates
 bubbles from the silicone 
mixture prior to c
asting. 
Casts of the Gripper finger are fully cured in 12 hours.
 No mold release 
was used during the casting process.
 This process is repeated 
three times. After the three grippers 
have been cast, they are inserted into the connector mold along with a 3D p
rinted triangular ring 
with the outer curved side of the gripper finger facing away from the center of the mold. 
  Figure
 32. Connector mold.
  35 30g of Dragon skin 30 (
15g of each part) is mixed and poured into the mold with the gripper and 
triangular 
component. The notches where the triangular ring is inserted should be taped with tape 
so the silicone can completely fill the mold. Both the top and bottom of the gripper finger are 
combined with the triangular ring in this manner.
  Figure
 33. Three grip
per fingers bonded with two triangular rings.
  3.1.4!Assembly
  Assembly begins by first inserting the unfinished actuator into one of the two wafer components. 
After the actuator is inserted into the wafer, the skin material is folded over the sides of the 
wafer, exposing the bellowed skeleton. Inserting the T
-shaped 
pin into the skeleton ensures that 
the skeleton will not pass through the wafer during actuation. Unfold the skin material and seal 
with the AIE
-305 impulse sealer on setting number 3. Before this process is repeated on the top 
 36 end of the actuator, the ass
embly must be inserted through both triangular rings in the gripper 
ring assembly. 
   Figure
 34. Exposed skeleton after insertion through wafer shown on the left. On the right, the T
-shaped pin inserted 
through the skeleton
. Before final sealing of the 
actuator, the pneumatic fittings should be attached to the skin material 
via a small incision. Once assembly is completed
, the three raised edges of the wafers should 
hang over the edges of the triangular ring and constrain the actuator to the center of th
e assembly.
  Figure
 35. Fully assembled optimal gripper design.
  37 3.2
! PneuNets D
esign
  PneuNets fabrication is the simplest of the three designs. Howev
er, it requires approximately 24
 hours of casting time for this specific design. Required materials include: Eco
-flex 00
-10 platinum cure silicone, digital scale, disposable cup and mixing utensil, 
3D printed top mold, 
cotton or equivalent, a 1/8
th inch diameter nail or equivalent, and P
VC tubing. 
First, insert the 
nail into the hole in the center of the 3D printed mold, this allows the tube to be inserted without 
cutting material after casting. 
  Figure
 36. 3D printed PneuNets mold.
 Mix one hundred grams of both part A and part B of Eco
-Flex 00
-10 thoroughly in 
a disposable 
cup
 and pour into the mold
. Pouring the mixture from an elevated height above the mold will 
limit the amount of large bubbles or defect in the casting. As in th
e optimal gripper finger 
casting, vacuum degassing can also be used to extract air trapped in the mixture. 
Allow for
 a minimum of
 4 hours of curing time before removing
 (a 12 hour overnight cure was used in this 
study
). After the mold has been cast, 
use any remaining mixture or prepare another small batch 
 38 and cast a cylinder of approximately 30mm diameter and 25 mm height. Insert another nail or 
equivalent in the center of the cast. Molding the cylinder can be done with spare polyester film 
material r
olled into a cylinder and taped at the seams. After both casts have had time to cure 
remove the cast from the mold. Prepare a sheet of cotton that is the approximate size of the top 
cast. Prepare a small batch of Eco
-Flex 00
-10 and spread a thin layer in
 an ÒXÓ pattern on
to the 
cotton sheet.
  Figure
 37. Eco
-flex 00
-10 spread on cotton sheet.
 Place the top mold in the center of the ÒXÓ. Ensure that the layer of Eco
-flex on the cotton layer 
is not too thick or it will block the pneumatic channels on the top 
layer and can lead to 
asymmetric gripper contractions.
  39  Figure
 38. PneuNets top cast curing to cotton layer.
 While the top and bottom PneuNet layers are curing, the cylinder can be adhered to the top of the 
mold. 
Brush a thin layer of mixed Eco
-flex 00
-10 on to both surfaces and align tube holes. Once 
fully cured, the excess cotton layer can be trimmed and the PVC hose can be inserted. A rubber 
band or sil
-poxy can be added to the hose inlet if air escapes during inflation of the device. 
  Figure
 39. Fabricated PneuNets design.
    40 3.3
! Magic
-Ball Design
 !While requiring few steps, folding of the Magic
-Ball pattern can be a tedious process. Processing 
on the Full Spectrum CO
2 laser takes 13 minutes. A power level of 20% and speed of 85% were 
used to cut the
 pattern, 
Figure
, into the polyester material. The pattern file was generated in 
Adobe Illustrator and issues arose when the dotted lines were created. It should be n
oted that the 
Full Spectrum software will read dotted lines as solid, unless they are separated in the original 
Illustrator file. There are many ways to accomplish this, but outlining
/expanding
 the ÒstrokeÓ of 
the dotted line will create small rectangular 
shapes
 composed of four cuts instead of a single cut
. The laser cutter will produce a dotted line, but this will significant
ly increase the cutting time. 
  Figure
 40. Preliminary testing with different laser settings
. Once the patterned polyester sheet 
has been folded, the two free ends are taped together with 
packing tape. Then the Ò
magic
-ballÓ can be formed by re
-folding the connected ends and 
compressing the top and bottom together. 
A polyurethane f
ilm of 0.0015
-gauge thickness was 
used as the skin material. Two circular sheets of approximately 
200mm diameter should be cut. 
  41   Figure
 41. Top polyurethane sheet for magic ball design shown left, bottom sheet with seam trace is shown right
 Place one sheet over the top of the Òmagic
-ballÓ skeleton and the second sheet on into the inner 
cavity of the skeleton. A 100mm diameter polypropylene sphere 
was used in this research 
to 
keep
 the inner skin in place while marking the seam locations
.   Figure
 42. Magic
-ball skeleton with sphere holding inner skin in place.
 After seams have been marked on both the top and bottom skin, the sphere and skeleton should 
be removed. Line up the seam lines on both the top and bottom skin and seal on heat setting
  42 number three on the AIE
-305 impulse sealer. Leave an opening unsealed and before the skeleton 
is inserted, install the pneumatic fittings in the center of the top skin. Next, insert the skeleton 
into the partially sealed skins, and 
seal the remaining edge
s. 
  Figure
 43. FOAM magic
-ball design.
    43 Chapter 4
: Testing Methods
  Testing method described in 
this 
chapter 
are a means of evaluating the desirable parameters of an 
end effector including: grasping strength and variability, grip effectiveness, and performance in a 
realistic scenario. Very few standardized testing methods currently exist for the end effector 
desi
gns presented in this work; however, some of the tests are loosely based on similar ASTM 
testing methods. 
 4.1
!Weighted 
Grasping 
Variability
  Grasping variability was tested for each end effector design on its ability to securely grasp 
spherical objects of varying size and varying weights. Four polypropylene spheres of 100mm, 
70mm, 50mm, and 25mm diameter
 were measured at 44, 13, 8, and 2 grams
, respectively
. Each 
sphere was incrementally 
weighted with
 50, 100, 200 and 300 grams in addit
ion to the weight of 
the sphere. 
  Figure
 44. Polypropylene spheres
 100, 70, 50 and 25mm diameter
 with attached monofilament tethers.
 A grasping test was considered successful if the end effector could maintain its grip on the 
sphere for five seconds with out any visible slippage or ejection. If the end effector was able to 
 44 successfully grasp each weight increment for a sphere, a maximum
 weight was attempted by 
incrementally loading and checking for slippage or ejection until the device failed. 
All end 
effectors were positioned in a vertical orientation suspended by monofilament wire. Each sphere 
was fixed with a length of monofilament wi
re that could be attached to a paper weigh
t boat, 
weighing four grams, that was used to hold the weight. Two
-inch wood screws, weighing three 
grams, were used as the mass in this experiment. 
A 4.7
-ounce syringe with two check valves was 
used as a pump and 
a vacuum to actuate the designs. 
 4.2
!Grasp E
ffectiveness
  Grasp effectiveness was test
ed through a series of 
tests in which
 an impulse or mechanical shock
 was induced
 to the end effector while grasping an object. 
Each end effector was suspended in a 
vertic
al orientation by monofilament wire. A gala apple weighing 140 grams was grasped by the 
end effector which was then subjected to a series of vertical drops at increments of 5, 10 and 15 
centimeters above its starting position. 
Five trials were performed at
 each drop interval for every 
end effector tested.
 A test was successful if the apple did not slip or eject from the gripper. 
 A secondary test was conducted in which a Bosch BMA280 accelerometer was used to collect 
acceleration data during test. The devic
e was suspended in a Plexiglas fixture and the end 
effector was 
secured
 below. The testing procedure was identical to the initial impulse test, 
however, only one trial was completed for each height increment. 
 4.3
!Real W
orld 
Evaluation
  All three end effector designs were tested at an apple orchard at the horticulture farm on 
Michigan State UniversityÕs campus. 
Tradition
al apple trees
 that had been trimmed the previous 
season were selected due to their similarity to 
possible 
scenarios fou
nd in regenerative farming 
situations. 
  45   Figure
 45. Conventional apple
 orchard
 tree shown 
left
, compared to "high
-density" apple
 orchard
 tree shown 
right
. Each end effector 
was placed in various scenarios with apples in clusters as well as single apples 
of varying orientation. The accessibility of the apples was evaluated in this qualitative test for 
each end effector. An attempt was also made to harvest an apple from the tre
e with each end 
effector being manually manipulated by an operator. 
  Harvesting metrics for Spartan apples, a variety of Macintosh, were tested for one 
hundred samples
 on Michigan State UniversityÕs horticulture farm
. Detachment force, apple 
diameter and 
weight were recorded during the first week in August 2018. A Òfish hookÓ uniaxial 
force gauge was attached to the apple during harvesting to record force data.
 The PneuNets 
design was used to harvest the first 20, with the remaining 80 samples using the Òf
ish hookÓ and 
monofilament sling.
 Weight wa
s recorded using the same Òfish 
hookÓ gauge. Diameter was 
determined using a set of digital calipers.  
   46 Chapter 5
: Results and Discussion
  Chapter five presents the results of the test
s performed and t
hen provides a discussion on 
these 
results. Only two designs were successfully tested
, with the Magic
-ball design failing to grasp 
any of the testing materials without the use
 of a vacuum pump, which was unavailable at the time 
of test
ing
.  5.1
!Grasping Variability Results
 Grasping variability 
test results
 in Table 2
 show that 
the optimal grippers design 
had success 
with a greater siz
e range, but at a lower weight. Contrary to the optimal gripper results, the 
PneuNets design successfully grasped a maximum weight of approximately 1.2 kilograms for 
both the 100 and 70 millimeter spheres. Smaller objects proved to be a challenge for the 
PneuNets 
design which was unable to grasp any of the weighted scenarios for the 25 millimeter 
sphere.
 Grasping 
for the optimal gripper was 
more reliant
 upon side wall contact between the 
fingers
 and object being grasped, while the PneuNets design applied a 
caging style of grasping. 
The difference between these two methods can be seen in 
Figure
.   Figure
 46. Grasping scheme for the optimal gripper shown 
left
, with caging scheme of the PneuNets design shown 
right
.    47 Table 
2. Results from the weighted grasping variability test performed on the optimal gripper and PneuNets 
designs.
 Optimal Gripper
 Sphere (mm)
 Added weight (g)
 Total weight (g)
 Result
 100 50 95 pass
 100 147 pass
 200 245 pass
 300 345 fail
 70 50 64 pass
 100 113 pass
 200 213 pass
 300 313 fail
 50 50 59 pass
 100 108 pass
 200 208 pass
 300 308 pass
 25 50 52 pass
 100 103 pass
 200 202 fail
 300 302 fail
 PneuNets 
 100 50 94 pass
 100 144 pass
 200 244 pass
 300 344 pass
 1,199
 1243 pass
 70 50 63 pass
 100 113 pass
 200 213 pass
 300 313 pass
 1,199
 1212 pass
 50 50 58 pass
 100 108 pass
 200 208 pass
 300 308 fail
 25 50 52 fail
 100 102 fail
 200 202 fail
 300 302 fail
  PneuNets and optimal gripper designs were successfully 
tested; however
, a loss of actuation 
occurred over time for all designs.
 This is due to the permeability of the materials used in each 
design. Applying a constant positive or negative pressure would prove beneficial and may have 
led to improved results.
   48 5.2
!Grasping Effectiveness
  As shown in Table 3, w
hile the optimal gripper d
esign passed the drop test for the five 
centimeter displacement, failures occurred during the 10 and 15 cm displacements.  Even though 
the apple was not 
constrained
, the PneuNets design succeeded at all displacement
 drop increments
. During 
each trial, the caging method used by the PneuNets design allowed the apple 
to shift within the grasp of the device, however, it did not allow for ejection due to the rigidity of 
the limbs when inflated.
 Table 
3. Trial results for 
grasp effectiveness during drop testing.
 Optimal Gripper
 Displacement (cm)
 Trial
 Result
 5 1 pass
 2 pass
 3 pass
 4 pass
 5 pass
 10 1 fail
 2 pass
 3 pass
 4 fail
 5 pass
 15 1 fail
 2 fail
 3 fail
 4 pass
 5 fail
 PneuNets
 5 1 pass
 2 pass
 3 pass
 4 pass
 5 pass
 10 1 pass
 2 pass
 3 pass
 4 pass
 5 pass
 15 1 pass
 2 pass
 3 pass
 4 pass
 5 pass
   49 !Figure
 47. Acceleration graph for 
5 cm 
displacement 
optimal gripper test.
 !Figure
 48. Acceleration graph for 10 
cm 
displacement optimal gripper test.
  50 !Figure
 49. Acceleration graph for 15 
cm displacement optimal gripper test
.  Figure
 50. Acceleration graph for 
5 cm displacement PneuNets 
test
.  51  Figure
 51. Acceleration graph for 10 
cm 
displacement PneuNets test
.  Figure
 52. Acceleration graph for 15 
cm 
displacement PneuNets test
.    52 5.3
!Real World Evaluation Results
  Sampling from the apple orchard evaluation gave insight 
into 
the variability in harvesting 
scenarios that potential end effectors
 might encounter
. While majority of the desirable apples 
grow on the perimeter of the tree, some 
desirable apples are covered 
by branches and foliage 
creating potential hazards and clearance issues for end effector designs. Apples can be found in 
both clusters and as individuals in a variety of orientation
s along the branch. Each growing 
pattern presents its own unique challenge 
for approach planning and end effector orientation with 
relation to the branch. 
 [A]
 [B] Figure
 53. Apple arrangement on conventionally trimmed orchard trees.
 Each end effector
 was tested on both isolated
 apples and apples arranged in clusters. 
When 
harvesting the isolated apple, the end effector approached the apple from below to avoid 
accidentally grasping foliage or branches
. For the isolated harvesting, the PneuNets design was 
the only end effector to 
successfully remove an apple from the branch. During the clustered 
 53 harvesting, an angled approach
 as shown
 in 
Figure
 was taken to avoid interference from the 
surrounding app
les, branches and other foliage. The PneuNets design and the Magic
-Ball design 
both had success removing apples from the branch in the
 clustered arrangement. 
  [A]
 [B] [C] Figure
 54. End effectors grasping isolated apples from below.
  54      Figure
 55. End effectors grasping apples in cluster configuration.
 Although apples were successfully harvested by the PneuNets g
ripper and the Magic
-Ball 
design, the testing done in the orchard was purely for qualitative purposes. No conclusions can 
be drawn from these samples taken due to a lack of standardization and 
sample size. With that 
being said, simulating a realistic grasp
ing scenario gave significant feed
back on various elements 
of each design
, including the end effector compliance, dexterity, maneuverability and actuation 
speed.
  Results of the Spartan
-Macintosh apple sampling can be seen in 
Table 
4.  55 Table 
4. Results of Apple sampling.
  Average
 Standard Deviation
 Force (lbf.)
 3.15 ±1.32 Diameter (mm)
 65.23 ±4.46 Weight (lb.)
 0.24 ±0.045   Figure
 56. Normal 
distribution curve for detachment force and probability density
 of Spartan apples
.  Figure
 57. Normal distribution curve for diameter and probability density
 of Spartan apples
. (()(*
()+
()+*
(),
(),*
()-
()-*
(,./01'2343#5#67!8&$9#67
:$#4;#45!8&646<=>&$6!"2'<&!?53"@
A2'<&!$2'>45!8#96'#3B6#2$
(()(+
()(,
()(-
()(.
()(*
()(/
()(C
()(0
()(D
()+
*(**/(/*C(C*0(0*1'2343#5#67!8&$9#67
E#4>&6&'!?>>@
E#4>&6&'!$2'>45!8#96'#3B6#2$
 56  Figure
 58. Normal distribution curve for weight and probability density of 
Spartan apples.
 Because these apples were harvested 
prematurely
 (not fully ripened)
, their detachment force 
might not be exact representations of the forces seen during 
conventional harvesting periods
. The
 uniaxial pulling
 harvesting method was the simples
t mechanically, but also produce
d the 
highest
 detachment force.
 Li et al.
 [39] demonstrated the mechanics of a rotated pulling technique that 
produces a moment on the stem and lessens the grasping force required to detach the apple. 
 !Figure
 59. Detachment force versus bending angle during apple harvesting.
 (+,-.*/C0D+((()(*
()+
()+*
(),
(),*
()-
()-*
().
1'2343#5#67!8&$9#67
F&#%=6!?539@
F&#%=6!$2'>45!8#96'#3B6#2$
 57  Figure
 60. Free body d
iagram of apple detachment method proposed by Li et al.
 [39]
. Despite difference in detachment technique, the early harvest sample data collect
ed for this 
study
 shows similar detachment force values for low angle detachments seen in the work by Li et al. 
[39].     58 Chapter 6
: Conclusion and Future Work
 6.1
!Conclusion
  Experts are increasingly concerned about the potential for 
a crisis involving the industrial 
agricultural complex. Non
-regenerative farming practices continue to degrade soil quality and 
the surrounding environment. 
Automation through soft robotic
 presents a novel, low cost and 
energy efficient method of harvesting produce grown in nonconventional arrangements. The 
under
-actuated end effectors tested in this thesis have shown promising results
, and further 
evaluation should be conducted. 
Although 
the 
majority of the test
s conducted were done so in 
a non-standard qualitative manner, some conclusions can still be drawn about the selection of end 
effectors. 
With the 
particular 
designs, manufacturing approaches and testing protocols in the 
current study
, the PneuNets and the Optimal gripper designs were the two most successful end 
effectors, with the PneuNets performing the 
best overall
. At a low cost of $15.59
, the simplest 
manufacturing procedure, 
a high grasping strength and effectiveness
, the
 PneuNets design 
shows 
promise
 as a 
viable option
 for future implementation
.  6.2
!Future work
  A significant amount of future work is required before soft robotics 
can out-perform
 human 
operators 
in 
harvesting produce. 
 Although minimal success was found with the Magic
-Ball design, the base technology, 
FOAM, presents a novel and low cost actuator that can be arranged in a variety of end effector 
configurations. 
The potential of
 utilizing a purely FOAM design should by no
 means be 
disc
ounted
  based purely on the
 results of this research. When grasping non
-spherical objects 
such as bottles or objects with non
-smooth edges, the Magic
-Ball design is able to lift significant 
 59 weight. Combining the FOAM method with a 
softer skeleton or adding soft material to the 
grasping surface of a FOAM design could potentially improve 
slippage and ejection
 issues.
 Improvements to the optimal gripper design should focus on the
 stiffness
 and the shape of 
the 
gripper 
fingers
. The la
ck of rigidity in the finger tips
 led to slippage and ejection with
 objects 
of larger mass. Using a material with a larger modulus of elasticity such as flexible filament 
would also benefit the manufacturing process by removing the need for silicone castin
g. Having 
the bulk of the design 3D printed would improve manufacturing time and reduce cost.
 For the PneuNets design, changes should be explored that would enhance actuation 
speed, reduce weight, and decrease pre
-actuated foot print. Groups such as Soft R
obotics 
Inc. 
have been working towards this goal with their pick and place system 
as seen in 
Figure
. Shifting 
to the Fast PneuNets design would possibly provide benefits in actuation speed, but the device 
would be sacrificing some compliance.
  This introductory evaluation has shown the poss
ible application for soft robotics in small 
scale agriculture. While these designs are not optimized, they show potential while remaining 
significantly less expensive than their conventional counterparts. Future end effector design
s should have compliance,
 speed and produce security as high priority design parameters.
 Produce 
damage is also an important factor, which was not directly considered here. 
Next steps should 
also include the development of a low
-cost robotic limb and vision system to further devel
op a 
fully automated harvesting solution for regenerative farms. 
  ! 60 BIBLOGRAPH
Y 61  ![1]  FAO, "High Level Expert Fourm 
- How to Feed the World in 2050," in 
Global 
Agriculture Towards 2050
, Rome, 2009. 
  [2]  HLPE, "Investing in smallholder agriculture for food security. A report by the High Level 
Panel of Experts on Food Security and Nutrition of the Committee on World Food 
Security," Rome, 2013.
  [3]  J. M. O. Sandra L. Colby, "Projections of the Size and C
omposition of the U.S. 
Population: 2014 to 2060," U.S. CENSUS BUREAU, 2015.
  [4]  The World Bank, "Urban Population (% of total)".
  [5]  The Department of Economic and Social Affairs of the United Nations, "World 
Urbanization Prospects The 2014 
Revision," New York, 2014.
  [6]  USDA, "ers.usda.gov," 2016. [Online]. Available: 
https://www.ers.usda.gov/webdocs/charts/84911/sept17_datafeature_cromartie_fig05.png
?v=42976. [Accessed 2018].
  [7]  A. M.
-B. Christopher Barrington
-Leigh, "A century of sp
rawl in the United States," 2015. 
 [8]  Natural Resource Conservation Service, "2007 National Resources Inventory," 2010.
 [9]  J. Rockstrom, P. Kaumbutho, J. Mwalley, A. Nzabi, M. Temesgen, L. Mawenya, J. 
Barron, J. Mutua and s. Damgaard
-Larsen, "Conservation farming strategies in East and 
Southern Africa: Yields and rain water productivity from on
-farm action research," 
Soil 
& Ti
llage Research, 
vol. 103, no. 2009, pp. 23
-32, 2009.   [10]  D. A. Sumner, "American Farms Keep Growing: Size, Productivity, and Policy," 
Journal 
of Economic Perspectives, 
vol. 28, no. 1, pp. 147
-166, 2014.   [11]  United States Department of Agriculture
, "United States Summary and State Data," 2012.
 [12]  World Commission on Environment and Development, "Our Common Future".
 [13]  K. A. Dahlberg, "Sustainable Agriculture: Fad or Harbinger?," 
BioScience, 
vol. 41, no. 5, 
pp. 337-340, May 1991. 
 BIBLIOGRAPHY
  62 [14]  M. 
V. Gold, "Organic Production/Organic Food: Information Access Tools," April 2016. 
[Online]. Available: https://www.nal.usda.gov/afsic/organic
-productionorganic
-food-information
-access-tools. [Accessed July 2018].
  [15]  K. A. Dahlberg, "REGENERATIVE FOOD 
SYSTEMS," in 
MANAGEMENT OF 
AGRICULTURAL, FORESTRY AND FISHERIES ENTERPRISES 
- Vol. II
, MI: 
Encyclopedia of Life Support Systems. 
  [16]  M. A. Altieri and S. B. Hecht, "Small
-Scale Agriculture in SouthEast Asia," in 
Agroecology and Small Farm Development
, CRC, 1990.   [17]  The Carbon Underground, "What is Regenerative Agriculture?," 24 Feburary 2017. 
[Online]. Available: https://thecarbonunderground.org/wp
-content/uploads/2017/02/Regen
-Ag-Definition
-7.27.17-1.pdf. [Accessed 2018].
  [18]  P. Stephen G. 
Bronars, "A Vanishing Breed How the Decline in U.S. Farm Laborers Over 
the Last Decade Has Hurt the U.S. Economy and Slowed Production on American 
Farms," 2015.
  [19]  S. Bronars, "NO LONGER HOME GROWN: How Labor Shortages are Increasing 
AmericaÕs Relianc
e on Imported Fresh Produce and Slowing U.S. Economic Growth," 
Partnership for a New American Economy, 2014.
  [20]  R. STROCHLIC, C. WIRTH, A. FERNANDEZ BESADA and C. GETZ, "FARM 
LABOR CONDITIONS ON ORGANIC FARMS IN CALIFORNIA," 2008.
  [21]  S. M. GABBAR
D and J. M. PERLOFF, "The Effects of Pay and Work Conditions on 
Farmworker Retention," 
INDUSTRIAL RELATIONS, 
vol. 36, no. 4, 1997. 
  [22]  Destaco, "End Effectors," [Online]. Available: https://www.destaco.com/end
-effectors.html.
  [23]  A. Robinson, "The
 History of Robotics in Manufacturing," cerasis, 2014.
 [24]  UC Davis , "Biological and Agricultural Engineering," [Online]. Available: 
https://bae.engineering.ucdavis.edu/blog/stavros
-vougioukas
-awarded
-funding
-develop
-robots
-harvest/. [Accessed 2018].
  [25]  R. Lu, Z. Zhang and A. K. Pothula, "Innovative Technology for Apple Harvest and In
- eld Sorting," 
NEW YORK STATE HORTICULTURAL SOCIETY, 
vol. 25, no. 2, 2017. 
  [26]  D. Bulanon and T. Kataoka, "A Fruit Detection System and an End Effector for Robotic 
Harvesting of Fuji Apples," 
Agricultural Engineering International, 
vol. XII, no. 2010, 
2010.   63 [27]  J. Davidson and A. Silwal, "Robotic Apple Harvesting in Washington St
ate," in 
IEEE 
Agricultural Robotics & Automation Webinar
.   [28]  J. H. TIBBETTS, "Not Too Far From the Tree," 2018. [Online]. Available: 
https://www.asme.org/wwwasmeorg/media/ResourceFiles/EngineeringTopics/Robotics/0
218_ME_CoverStory.pdf. [Accessed 2018
].  [29]  B. Mosadegh, P. Polygerinos, C. Keplinger, S. Wennstedt, R. F. Shepherd, U. Gupta, J. 
Shim, K. Bertoldi, C. J. Walsh and a. G. M. Whitesides, "Pneumatic Networks for Soft 
Robotics That Actuate Rapidly," 
Advanced Functional Materials, 
vol. 24, no
. 15, p. 2163Ð2170, 2014.   [30]  J. Rossiter and H. Hauser, "Soft Robotics 
- The Next Industrial Revolution? [Industrial 
Activities]," 
IEEE Robotics & Automation Magazine, 
vol. 23, no. 3, pp. 17
-20, September 2016. 
  [31]  A. Miriyev, G. Caires and H. Lipson, "Functional properties of silicone/ethanol soft
-actuator composites," 
Materials & Design, 
vol. 145, pp. 232
-242, May 2018. 
  [32]  Super Releaser, "superreleaser.com/," [Online]. Available: 
https://superreleaser.com/blog/2017/5/11/a
-short
-history
-of-soft
-grippers.
  [33]  F. Ilievski, A. D. Mazzeo, R. F. Shepherd, X. Chen and G. M. Whitesides, "Soft Robotics 
for Chemists," 
Angewandte Chemie, 
January 2011. 
  [34]  S. Li, D. M. Vogt, D. Rus and R. J. Wood, "Fluid
-driven, origami
-inspired artificial 
muscles," 
Proceedings of the National Academy of Sciences of the United of America, 
vol. 114, no. 50, pp. 13132 
- 13137, 27 November 2017. 
  [35]  T.-L. C. C.
-H. C. M.
-C. H. Chih
-Hsing Liu, Y. Chen, T.
-Y. Pai, W.
-G. Peng and Y.
-P. 
Chiang, "Optimal Design of a Soft Robotic Gripper for Grasping Unknown Objects," 
SOFT ROBOTICS, 
vol. 00, no. 00, 2018. 
  [36]  J. Falco, K. Van Wyk, S. Liu and S. Carpin, "Grasping the Performan
ce Facilitating 
Replicable Performance Measures via Benchmarking and Standardized Methodologies," 
IEEE ROBOTICS & AUTOMATION MAGAZINE, 
2015.   [37]  J. Zhang, A. Jackson, N. Mentzer and R. Kramer, "A Modular, Reconfigurable Mold for 
a Soft Robotic Gripper
 Design Activity," 
Front. Robot. AI,, 
september 2017. 
  [38]  Soft Robotics Inc, "Core Technology," Soft Robotics INC, 2018. [Online]. Available: 
https://www.softroboticsinc.com/core
-technology. [Accessed 2018].
   64 [39]  J. Li, M. Karkee, Q. Zhang, K. Xiao and T. Feng, "Characterizing apple picking patterns 
for robotic harvesting," 
Computers and Electronics in Agriculture, 
vol. 127, pp. 633
-640, september 2016.