Design and Fabrication of Soft Locomotion Robots

based on Spatial Compliant Mechanisms

Andrija Milojevic Kyrre Glette
Department of Mechanical Engineering RITMO, Department of Informatics
LUT University University of Oslo
Lappeenranta, Finland (former) Oslo, Norway
[email protected] [email protected]
arXiv:2408.05207v1 [cs.RO] 9 Aug 2024

Abstract—Soft robotics has emerged as a promising technology

that holds great potential for various application areas. This
is due to soft materials unique properties, including flexibil-
ity, safety, and shock absorption, among others. Despite many
advancement in the field, the development of effective design
methodologies and production techniques for soft robots remains
a challenge. Although numerous robot prototypes have been
proposed in recent years, their designs are often complex and
difficult to produce. As such, there is a need for more efficient
and unified design approaches that can facilitate the production
of soft robots with desirable properties. In this paper, we propose
a method for designing soft robots using elastic beams and spatial
compliant mechanisms. The method is based on an evolutionary
approach that enables the creation of designs with both high
motion and force transmission ratios. Specifically, we focus on
the development of locomotion mechanisms using a central linear
actuator. Our approach involves the use of commonly available
plastic materials and a 3D printer to manufacture the designs. Fig. 1. Concept of soft locomotion robots based on a net of thin elastic
We demonstrate the feasibility of our approach by presenting beams (spatial compliant mechanisms): (a) illustration of one robot design,
experimental results that show successful production and real- (b) working principle - active state, c) working principle – passive state, (d)
world operation. Overall, our findings suggest that the use of examples of different robot designs and actuator placements.
elastic beams and an evolutionary approach can facilitate the
creation of soft robots with desirable locomotion properties,
including fast locomotion up to 3.7 body lengths per second,
locomotion with a payload, and underwater locomotion. This Different solutions of soft robots are introduced over time:
method has the potential to enable the development of more continuous robots in form of manipulators [9–11], robotic
efficient and practical soft robots for various applications. grippers for object manipulation [12–15], for realizing ter-
restrial locomotion like walking [16–19], crawling [7, 20],
I. I NTRODUCTION running/galloping [21], jumping [22], multimodal locomotion
gaits [23–27], swimming [28, 29], or underwater exploration
Expanding research interests in the soft robotic field have [30]. Most of these soft robotic concepts are realized by
brought new understanding and knowledge of how soft and utilizing pneumatically driven soft elastomer actuators [7, 22,
elastic materials can be utilized to realize simple yet highly 27, 31] and combination of these actuators with other elements
functional soft robots [1, 2]. Unlike rigid-body-based robots [21, 32–35]. Active smart materials [18, 24–26, 36–41], tendon
composed of rigid links and joints, soft robots are usually built driven actuation [17, 42, 43], origami-based structures [19, 44–
from soft, elastic, or hyperelastic materials [3]. Traditional 47], and tensegrity structures [48, 49] are applied as well for
rigid body-based systems and robots are often limited in the actuation and development of soft robots.
degrees of freedom, maneuvers, and complexity of motions Robots based on pneumatically driven soft elastomer actu-
they can achieve, while not being safe to operate around ators are largely realized by utilizing hyperelastic materials
humans and often complicated and expensive to realize. Being [1, 7, 21, 50]. Smart material-based robots use other types
inherently soft and continuously deformable, soft robots can of active and responsive materials (when exposed to external
overcome these limitations and drawbacks of classical robotic influences, e.g. heat, light, current) [18, 24–26, 36–41], where
systems. This led to several innovations and applications of the material properties often imply slow actuation speed.
soft robots in medicine [4], wearable devices [5, 6], remote Origami structures are built from rigid or soft, thin material
exploration and inspection [3, 7], and food industry and sheets that can be easily folded [19, 44–47], where actuation
agriculture [8]. speeds remain relatively slow. Tensegrity soft robots are re-
alized either via a set of rigid rods and springs [49] or from mechanism robots). As one example of Combots, we present
completely soft hyperelastic materials [48]. Other soft robotic in this paper robots that can realize locomotion. The basic
concepts mostly utilize a combination of hyperelastic materials concept is illustrated in Fig. 1. The operating principle is based
and rigid elements [3, 35]. Hyperelastic materials exhibit poor on the transmission of displacement/forces from the input port
capabilities of rapidly storing and releasing relatively large to the appendage end-effector via elastic deformation of the
deformation energy, limiting the possibility to achieve robots individual beam-like thin segments (Fig. 1a). Having contact
with high locomotion speeds. with the ground, the appendages will push the robot body
Although these concepts lead to the realization of highly forward, realizing the locomotion of the whole soft robot
functional soft robotic systems, still there are great limitations (Fig. 1b). With the aim to realize high locomotion speeds
in actuation speed and payload capacity. while having relatively large available actuation forces in a
This is mainly due to the mentioned limitations of hy- compact design, we use electromagnetic actuation, in form of
perelastic materials and slow activation time of active smart linear solenoids, as actuators for the proposed soft robots. By
materials. supplying the appropriate electrical power to the solenoid, the
Some attempts to increase the soft robot locomotion speed plunge will contract and provide the input displacement and
include leveraging the mechanical instability inspired by spine force to the robot (Fig. 1b). After the supplied power is turned
flexion and extension in quadrupedal mammals, such as in off, the stored elastic deformation energy in the structure of
the galloping cheetah [21]. This concept remains dependent the appendages together with energy stored in the solenoid
on external air supply making onboard actuation hard, while spring will be released rapidly, providing in some portion an
the capability of such robots to realize transport of certain additional push forward momentum, contributing to the overall
payloads was never proven or tested. robot locomotion, Fig. 1c. We leverage these effects to realize
The design of soft robots in general represents a challenging soft robotic systems that can achieve high locomotion speeds
task. Past work mostly includes developing soft robots based compared to other existing soft robots [21].
on designer experiences [1–3] or seeking inspiration from In general, the structure of the Combots resembles spatial
nature, such as from animals [21, 25] or insects [17]. Methods compliant mechanisms with distributed compliance. The re-
for automated synthesis of soft locomotion robots remain search of 3D compliant mechanism design methods remains
limited and are mostly focused on pneumatically actuated soft very limited [57–61]. Most soft locomotion robots based on
robots [50–53], and only a few of these approaches are verified compliant mechanisms are realized by utilizing a planar (2D)
with real-world experiments [54]. compliant mechanism with concentrated compliance [62],
Material sets like common plastic in form of a net of thin where compliance is localized across the mechanism structure,
elastic beams have rarely been explored for the realization of in combination with classical motors to actuate the robot
soft robots [1, 28], however these have the capability to both locomotion [55, 56]. This is mainly due to the relatively simple
rapidly store and release deformation energy while providing synthesis process, where compliant mechanisms are designed
more strength compared to other mentioned material sets. based on existing rigid body mechanisms [62, 63]. Spatial (3D)
Thus, in this paper, we present a different approach to the compliant mechanisms with distributed compliance, where
realization of soft robots formed as a set of spatially distributed compliance is uniformly distributed across the structure, are
thin elastic beams. not utilized before for the realization of soft locomotion robots,
The structure of the introduced soft robots leverage compli- and their potential remains unexplored. This is mainly due to
ant mechanisms that are rarely utilized and explored for the lack of design synthesis approaches and a way to effectively
realization of soft robots [28, 55, 56]. actuate them.
Unlike rigid-body based mechanisms comprising of joints The many possible design solutions of the proposed soft
and linkages, compliant mechanisms represent monolithic compliant mechanism robots (Fig. 1d) motivate us to develop
structures that utilize elastic deformation of its individual an evolutionary algorithm-based synthesis framework for the
segments to realize motion and force transmission. Being automated design of Combots. This gives a convenient ap-
inherently monolithic, compliant mechanisms offer several proach to synthesis as no prior design experience is needed
benefits over classical ones: reduced complexity, ease of to yield an optimal solution, while allowing to explore design
manufacturing, no need for assembly, zero backlash, friction- solutions beyond those imagined by a human designer. The
free motion, no wear, no need for lubrication, better scalability, focus is on the design of one robotic appendage, as based on
better accuracy, lightweight design, built-in restoring force, this two-leg or multiple-leg robots could be developed.
low cost. In addition to proposing a synthesis framework with focus
Due to their many advantages, compliant mechanisms offer on easy transfer to real-world solutions, we demonstrate that
a promising solution for innovating in the soft robotics field. the Combots can achieve high locomotion speeds not reported
Thus, the main goal of this paper is to introduce a different before (up to 3.7 BL/s), with modest strength, while also
approach to the realization of soft robots, formed as net of being capable of transporting a certain payload. Moreover, we
spatially connected thin elastic beams and built by using show that the robots can realize both terrestrial and underwater
common plastic material. Inspired by compliant mechanism locomotion. This could open up for new application areas, like
structures, we name this class of robots Combots (compliant exploration and inspection in different environments, where
Fig. 2. Synthesis framework overview for soft locomotion robots: (a) problem formulation, (b) discretization, (c) evaluation of one solution for the optimization
process, (d) best found solution, (e) nonlinear Finite Element Method verification, (f) two-leg robot prototype.

such robots are lightweight, easy to transport, require little for compliant mechanism synthesis are introduced over time
energy to operate, could be made disposable, and are simple [61, 65, 69–71]. But most of these methods focus only on the
to realize. design of planar 2D compliant structures. Methods for optimal
The remainder of the paper is organized as follows: Section synthesis of spatial 3D compliant mechanisms are rarely
III describes the methodology of the synthesis framework and introduced as they impose a new level of complexity into the
presents results from the synthesis process. Then, section IV design process, due to spatial distribution of elements and thus
presents methods and results from a verification stage apply- relatively large design space [61]. Recent work [57, 58, 60]
ing nonlinear FEM simulations to selected designs. Further, propose a load flow method for the design of spatial compliant
section V describes the prototype realization of the selected mechanisms with distributed compliance, utilizing predefined
designs, and presents results from deformation behavior and building block designs to yield a mechanism topology. But
locomotion tests. Finally, sections VI and VII discuss the such a method is not suitable for the realization of complex
results and conclude the paper. appendages, as many of the possible design solutions are
disregarded at the beginning and remain unexplored. Other
II. R ELATED WORK IN DESIGN METHODS
works in the field of spatial compliant mechanism design
A structural topology optimization approach is adopted as include using continuum synthesis approaches [72–74]. These
the basis for the synthesis, mainly derived from the structural methods are not suitable for the design of discrete beam-like
mechanics [64] and compliant mechanisms research fields Combot appendages, as they usually lead to structures with
[65]. Topology optimization methods are rarely applied for concentrated compliance, or solutions that are too stiff for
the design of soft locomotion robots [66–68]. In topology the level of functionality that is required from soft robots.
optimization, the designer only needs to define the desired in- Moreover, existing research of spatial compliant mechanisms
puts (Fig. 2a), while through an evolutionary search algorithm, rarely leads to the fabrication of physical prototypes, where
solutions for the given problem are automatically obtained the suitable material solutions remain largely unexplored.
(Fig. 2f). The typical structural optimization process includes
Outside of the compliant mechanisms field, in our past
the following steps: topology optimization (finding the optimal
work [75], we developed a synthesis method for the design
material distribution within the given design space), dimen-
of MEMS micro-robot appendages. But this method is still
sional synthesis (dimensions of individual segments are opti-
limited to 2D planar-looking topologies and cannot be applied
mized), and shape optimization (shape of individual structural
in the case of soft locomotion robots. Thus, a new approach
segments or overall structure is synthesized). For the design of
to synthesis needs to be developed.
Combot appendages, we focused on the topology optimization
process as a more creative part of the synthesis steps including In general, none of the existing synthesis methods can
shape optimization, to broaden the possible design search. address the complex design requirements imposed by the
The Combot appendages resemble a spatial 3D compliant Combot appendages. To broaden the possible design search,
mechanism structure. Different topology optimization methods we incorporated both topology and shape optimization in one
synthesis process, which is not presented before for spatial methods for parameterizing include block-based methods [76]
compliant structures. Further, none of the introduced design or continuum discretization [69], but these are not applicable
methods have addressed the problem of synthesis of functional for the case of spatial Combot appendages synthesis, due to
compliant robotic appendages with stroke and payload require- limited exploration of the design space and different problem
ments. Thus, novel objective functions need to be investigated formulation. The discretized design domain i.e. network of
and formulated. elements shown in Fig. 2b represents the initial solution
from which the Combot appendage design is searched. In
III. E VOLUTIONARY ALGORITHM - BASED LEG MECHANISM the proposed synthesis process, individual elements can be
SYNTHESIS removed from or returned to the ground structure (Fig. 2c).
The synthesis goal for the Combots is to obtain an ap- Additionally, the overall shape of the appendages is optimized
pendage design that can realize a controllable motion of its by allowing the individual nodes to wander within a predefined
end-effector (tip of the robotic leg) in the desired directions region (Fig. 2b), changing the position and length of the
when input displacement is applied. Additionally, maximizing elements.
the motion transmission (ratio of realizing output to input After the discretization, a search method needs to be ap-
displacement) and force transmission (ratio of realized output plied to find the optimal solution within the given design
force to applied input force) of the appendages. This is done space. The discrete nature of the problem motivates us to
to realize soft compliant locomotion robots that can utilize utilize discrete optimization methods, of which Evolutionary
actuators with smaller input strokes, while also being energy algorithms (EAs) are especially suitable, as they have proven
efficient and capable of carrying a certain payload. The syn- to be efficient when searching large design spaces and have
thesis goal formulation is motivated by the assumption that an already been successfully applied in the topology optimization
appendage end-effector having more contact with the ground, of compliant-based structures [75, 77]. An objective function
due to higher motion transmission, can lead to Combots has to be provided for the EA, and in our case two alternative
that realize high locomotion speeds. In other words, it is a functions are proposed, where one focuses on the calculated
hypothesis that robots with high values of motion transmission geometric advantage, and the other also includes the mechan-
ratio can lead to higher locomotion speeds of the overall ical advantage of the appendage design. For the evaluation
robot structure while having a modest force transmission ratio, of the objective function, a linear finite element method
leading to energy-efficient robots. analysis is applied. After the optimization has converged the
remaining elements in the structure together with the chosen
A. Synthesis framework overview node locations will form the design of the Combot appendages
The proposed synthesis method is illustrated in Fig. 2. The that can potentially lead to Combot locomotion (Fig. 2d).
goal is to optimize the design of one robotic appendage, as Based on the obtained solution a solid 3D model of the
two- or multiple-leg soft robots could be developed based on appendage is designed (Fig. 2e). To verify the appendage
the same appendage design. The Combot appendages consist deformation behavior and functionality a nonlinear Finite
of a net of spatially connected thin beam-like segments, Element Method simulation is performed (Fig. 2e). As a final
where connections between elements are modeled as point step, the appendage structure is fabricated via a fuse deposition
connections. This simplifies the analysis and is an advantage modeling process by using a 3D printer, after which the linear
of the synthesis method. The main synthesis steps include solenoid actuator is attached (Fig. 2f).
problem formulation (Fig. 2a), discretization (parameteriza-
tion) (Fig. 2b), and optimization (Fig. 2c). B. Problem formulation
The problem formulation consists of defining the design The main goal is to design a Combot appendage that can
domain shape and boundaries, input displacement/force of realize the output motion of its end-effector in the desired
the actuator, location of the input port where the actuator direction when input displacement is applied, leading to over-
is attached, location of the end-effector, desired end-effector all robot locomotion. Here for the problem formulation, the
displacement direction, location of supports (where the ap- cuboid design domain is defined (Fig. 3a). Fixed supports are
pendage is attached to robot body), external loads that act placed at the left edge areas of the design domain, representing
on the appendage (to simulate the resistance of the ground), places where the appendage will be attached to the robot body.
material characteristics, and other constraints (like the desired Input port (where the actuator will be attached) is located at
density of the solution, or minimum required output displace- the middle of the left upper edge of the design domain. As
ment). an input, a displacement is applied (actuator available stroke),
As a next step, the discretization of the design domain is where the direction of input motion is defined in a Y direction
realized (Fig. 2b). Due to the discrete nature of the beam- (it is assumed that the used actuator has enough available input
like design of the appendages, the Ground Structure Approach force to actuate the appendages). The direction of the input
(GSA) is utilized for the parameterization [75], here expanded displacement ultimately depends on how the actuator will be
for the spatial problems. The design domain is divided by positioned, allowing different design options to be explored.
a number of nodes and a set of beam elements spatially The end-effector point is located at the design domain right
connecting the nodes representing the ground structure. Other side, placed at the down edge lower point. The desired
Fig. 3. (a-c) Problem formulation and discretization of the design domain for the soft compliant locomotion robot synthesis. Different investigated cases,
where input and output displacement direction are varied together with location of the appendage end-effector point. (d) Fully connected ground structure
with overlaying elements, (e) partially connected ground structure containing overlapping intersecting elements, (f) ground structure with identified crossing
- intersecting points.

direction of output motion is set in opposite from the direction ment are varied, together with location of the appendage end-
of the input displacement, and in the -Z direction (Fig. 3a- effector point.
c). This is motivated by the aim to realize robotic appendage
C. Parameterization of the design domain
that can both move the robot body in the desired direction of
motion (here Y direction) while also pushing the end-effector To realize automated synthesis the problem formulation and
against the ground to lift the robot body from the ground and design domain needs to be represented by a set of variables
realize overall robot locomotion (motion in -Z direction). The that can be optimized. The design domain is discretized by
external forces are applied at the end-effector point (in all three using the ground structure approach (GSA) [75, 77], but here
directions opposite the desired output displacement direction) developed for spatial problems (Fig. 3a-c). The cuboid design
to simulate the resisting force of the ground. Additionally, to domain is divided by nx × ny × nz number of nodes and a
this, a material characteristic (Young modulus) from which network of spatial beam elements connecting these nodes. For
the appendages will be produced is defined as well (here the the element formulation, spatial beam elements with 6 degrees
whole robotic appending is produced from a single material, of freedom at each end are used. The ground structure in
but multiple materials can be used). The specific parameter Fig. 3a-c represents the initial solution or design space within
values that are used for the problem formulation are given in the optimal solution is searched for. The design variables are:
i
Table A.1. • vel - defines if the selected element is present or absent
from the ground structure. The variable can take discrete
i
To demonstrate different design possibilities with the pro- values vel ∈ {0, 1}; denoting absence and presence in the
i
posed synthesis method, three different cases are investigated structure, respectively. The total number of vel variables
(Fig. 3a-c), where direction of the input and output displace- is equal to the total number of elements, i = numelem .
 • vxj , vyj , vzj – defines the position of the nodes. The robotic while requiring larger input forces to actuate the robot, where
appendage shape is optimized by allowing the individual Combot design when both GA and MA criterias are optimized,
nodes to wander within the predefined spatial region, would be more energy efficient but at the cost of realizing
changing the lengths and position of the elements. All the smaller output displacement. These fundamental trade-offs
variables can take discrete values within the predefined leave space for optimization, with a goal to realize robotic
range (Table A.1). The number of position variables is appendages that can achieve both high GA and MA values,
equal to the total number of the nodes in the structure, and potentially higher locomotion speeds.
j = numnodes . In addition, three more objectives are introduced to the
Initially, all the nodes in the structure are interconnected optimization process:
with one beam element, leading to a fully connected ground • To maximize the appendages payload capacity or yield a
structure (Fig. 3d). Such a structure contains overlaying solution that is stiff enough to realize locomotion and
elements that are difficult to produce (Fig. 3d). To avoid carry a payload, the end-effector displacement due to
obtaining such solutions a filtering algorithm is applied that external forces, dext
out , is minimized.
eliminates overlaying elements from the structure prior to the • To prevent obtaining solutions with non-existing or very
optimization. small densities, the desired density of the structure of
The ground structure can also contain overlapping ele- the appendages is optimized. The optimization goal is
ments, intersecting each other at one or multiple positions implemented by minimizing
Pnrel the difference between the
(Fig. 3e). The appendages with such elements are difficult realized ( Lrel
tot =
tot
i=1 Li ) and the desired (Ldes ) sum
to manufacture, and lead to stiffer robotic solutions. Thus, of all element lengths in the structure.
an algorithm is developed that identifies the total number of • The number of overlapping, or crossing, elements are
crossing points in the structure (Fig. 3f), where this number eliminated by minimizing the total number of identified
is then minimized during the optimization. The problem of overlapping points, noverlap , in the given solution.
eliminating such elements has not been solved before in the The automated synthesis of the Combot appendages rep-
topology optimization of spatial compliant structures. resents a multi-objective optimization problem. We follow a
Depending on the grid resolution, the total number of weighted sum approach to reduce the objectives into a single
variables can vary (e.g. for the design case in Fig. 3a-c, the objective function for the evolutionary algorithm.
overall number of variables is 232). Table A.1 contains the As two synthesis cases are investigated, different formula-
rest of the design domain parameterization parameters. tions of the objective function are defined.
The objective function when only GA is considered, in the
D. Objective formulation for optimization optimization, is given as follows:
The initial ground structure represents a large design space
max[GA − w1 · dext tot tot
out − w2 · |Lrel − Ldes | − w3 · noverlap ] (1)
containing a possible solution to the set problem. Search
method in form of optimization is applied to find the optimal The objective function when both GA and MA are consid-
appendage design. The main goal of the optimization is to ered is given in the following form:
obtain Combot appendages that can realize output displace-
ment of the end-effector in desired directions when input 1
displacement is applied. This is motivated by the end aim max[GA−w1 ·dext tot tot
out −w2 ·|Lrel −Ldes |−w3 ·noverlap −w4 · ]
MA
to realize Combot designs that can achieve locomotion in (2)
the desired direction while selecting the direction of input The weights used for the objective functions can be found in
actuation. Two different synthesis cases are investigated here, Table I, and are varied as described in Sec. III-F. The terms in
thus two main objectives are defined: the objective functions are calculated by using linear analysis.
• Maximizing the motion transmission ratio of the ap- For this synthesis case the initial problem setup is modified
pendage, that is, the ratio of realized output displacement by adding a spring element at the end-effector tip to calculate
of the end-effector to the applied input displacement. This the work done by the agent on the external environment. The
is defined as the Geometric Advantage, GA = dout /din . realized output force is calculated as force required to deform
• Maximizing the force transmission ratio of the ap- the spring element attached at the end-effector tip.
pendage, that is, the ratio of realized output force at the
E. Evolutionary algorithm
end-effector point to the needed input force to deform the
appendage. This is defined as the Mechanical Advantage, The design of Combot appendages represents a discrete
M A = Fout /Fin . optimization problem, as elements are either present or absent
from the initial structure, while also other variables have
These are done to explore the trade-offs between Combot
discrete values. This motivates us to use an Evolutionary
performances, designed when only GA is optimized and when
Algorithm (EA), in particular the Genetic Algorithm1 variant,
both GA and MA are taken into consideration. Here the
hypothesis is that appendages, where only GA is considered, 1 We will refer to the genetic algorithm as the EA to avoid confusion with
would realize higher output displacement of the end-effector the geometric advantage (GA) to be optimized.
 TABLE I Z direction of the end-effector point (Fig. 4), as these have the
E VOLUTIONARY A LGORITHM PARAMETERS . greatest effect on the robot locomotion. Taking performance
Parameter Value parameters into account, Structure 7 and 8 are of special
population size 200 - 400 interest. As can be seen, these designs can realize relatively
number of generations 1000 high GA values and realize large displacements in Y and Z
selection function roulette
crossover probability 95% directions, but at the cost of having low MA values, thus
mutation probability 9% requiring a relatively large input force to actuate them. The
elite count 2 optimization also managed to find some solutions where there
w1 0.3 : 0.1 : 0.7 * 102
w2 0.3 : 0.1 : 0.5 * 10 is a trade-off between GA and MA, Structures 5 and 10, while
w3 1 realizing modest values of end-effector displacement in both
w4 0.1 : 0.1 : 0.6 Y and Z directions. Solutions in this region represent the area
of interest with highest probability to realize locomotion.
2) Optimizing the mechanical and geometric advantage:
as these algorithms are considered to be robust in discrete, The squares in Fig. 4 show the performance of the solutions
complex, and relatively large search spaces. Genetic algo- when both mechanical and geometric advantage criteria are
rithms have previously shown good performance in structural taken into consideration, following Eq. 2. In all cases opti-
optimization [78–81], and have also proven to be efficient mization led to appendage designs that can realize the desired
for solving different optimization problems in the compliant direction of the end-effector motion. Here the special focus is
mechanisms field [75, 77]. on solutions Structure 13, 19 - 21. These appendage designs
The EA works on a population of candidate solutions, which realize both relatively high MA values while having modest
are coded as bit strings. The length of a single solution varies GA values, with a modest range of end-effector displacement
with the specific parameterization of the optimization case to in desired directions. Not surprisingly Structure 14 realizes a
be carried out, such as the total number of nodes and elements, high MA but at the cost of lower GA values, and Structures 18
outlined in Table A.1. Selection is performed using the roulette and 23 realize high GA but at the cost of lower MA values. In
wheel method, one-point crossover and bit flip mutations are general, the EA manages to find designs where there are good
used, and a generational replacement method with elitism is trade-offs between GA and MA, such as Structures 19-21. It
employed. If variation operators lead to solutions which do not is hypothesized that these solutions have a higher probability
satisfy the constraints, these are discarded and the operations to realize locomotion while also being capable of carrying a
are applied again until valid solutions are found. payload. In general, it could be concluded that these designs
Details on the EA parameters can be found in Table I. are more energy-efficient, while it is left to verify if they can
lead to better locomotion performance.
F. Synthesis results
To further study locmotion performances, appendage de-
To obtain Combot appendages solutions multiple Genetic signs 5 (only GA optimized) and 21 (GA and MA optimized)
Algorithms optimizations are run for each of the set synthesis are selected, as different representative solutions of each
cases (varying weighting constants and initial problem set- investigated case.
tings, Table I, Table A.1). The optimization simultaneously
eliminates elements from the initial structure (or allows them IV. N ONLINEAR FEM DEFORMATION BEHAVIOR
to reappear), while moving the structure nodes within the INVESTIGATIONS
predefined region. The remaining elements with selected po- To investigate the deformation behavior of the obtained
sitions of the nodes will form the optimal solution of the soft Combot appendages solutions, a nonlinear finite element
compliant mechanism robot appendage. Synthesis results for method analysis is performed by using commercially available
both design cases are summarized in a combined plot in Fig. 4, FEM software (Abaqus). The investigations are done for the
together with example structures. The following subsections appendage design solutions that were selected as representative
discuss the results for the investigated design cases. designs of the studied optimization cases (Structure 5 and
1) Optimizing geometric advantage only: The circles in 21, Fig. 4). The solutions were chosen to represent different
Fig. 4 show the performance of the solutions when motion optimization cases and different objective functions (GA vs.
transmission ratio criteria are optimized as described with GA and MA).
Eq. 1. Different appendage designs are obtained for various
weighting constants and initial problem parameters definition, A. Method
as shown in Fig. 3a-c (to explore different design trends, see Based on the obtained solutions, a solid 3D model of the
Table A.1). Examples of these designs can be seen in the appendages is realized (Fig. 5a, d). For the FEM simulations,
bottom part of Fig. 4. The results demonstrate that in all cases a fixed boundary condition is applied at the surfaces where
the soft robotic appendages realize motion of the end-effector appendages should be attached to the robot body. To simulate
in the desired direction when input displacement is applied. the actuator stroke, displacement of din = 5 mm (same as in
The appendage’s design performances are compared consid- the optimization) is applied at the input port surface. For the
ering GA, MA, and realized displacement values in the Y and solver, a large deformation analyst setup is defined. Fig. 5
Fig. 4. Results of the soft compliant appendage robot synthesis. Solutions where only motion transmission ratio is optimized (GA values) are marked with
circle. Solutions where both motion and force transmission ratio are optimized (MA values) are marked with rectangle. Results are shown for the various
problem setups defined in Fig. 3a-c. Solutions 1-4 correspond to problem setup at Fig. 3a, Solutions 5-8 correspond to problem setup at Fig. 3b, Solutions
9-12 correspond to problem setup at Fig. 3c. This is repeated for solutions 13-24, where both GA and MA were optimized for the same setups. Circles signify
structures optimized for GA only, while squares signify optimization for both GA and MA.

shows the complete initial simulation setup (a, d) as well as case of Str. 5 there is a better agreement between the results
the mesh model of one appendage (b, e). As different synthesis (Fig. 5g). Similar trends could be observed when investigating
cases are investigated (Fig. 3 and 4), FEM simulations are run the realized GA (Fig. 5h) and MA values (Fig. 5i). When
for each of the investigated cases. in the linear regime of deformation behavior there is a good
agreement between results, while for a higher input displace-
B. Results
ment, there is a difference in results (for GA, Fig. 5h, better
Fig. 5c shows the appendage’s deformation behavior for agreement in case of Str. 21, while for MA, Fig. 5i, better
the case when only GA was optimized, while Fig. 5f when agreement in case of Str. 5i).
both GA and MA were optimized. In all cases, simulation Considering the performance parameters (displacement,
results show that similar deformation behavior is realized as GA, and MA values), Str. 5 realizes a larger output displace-
in the solutions obtained with optimization (Structures 5 and ment compared to Str. 21 (Fig. 5g). This is also reflected in
21 in Fig. 4). The end-effector realizes motion in the desired the GA of the appendages, where Str. 5 archives higher GA
direction when input displacement is applied. values compared to Str. 21 (Fig. 5h). Not surprisingly, Str.
To investigate the appendage motion path, the realized end- 21 realizes higher MA values compared to Str. 5, leading
effector displacement in the Y and Z directions is plotted to a more energy-efficient solution (smaller input forces are
(Fig. 5g). The achieved values of GA (Fig. 5h) and MA required to deform the appendage structure). In general, Str.
(Fig. 5i) with respect to applied input displacement, are 21 realizes a better trade-off between GA and MA values.
analyzed as well. All the results are compared with values
obtained from optimization. For smaller values of end-effector V. P ROTOTYPE REALIZATION AND LOCOMOTION
displacement, there is a good agreement between optimization EXPERIMENTS
and FEM results (Fig. 5g), structures behave linearly. Two Combot designs are selected for realization as rep-
As the output displacement increases, nonlinear deformation resentative results from each of the investigated synthesis
effects are more expressed, thus there is a difference between cases (structures 5 and 21). In this paper, the focus is on
the results (linear analysis is used in the optimization). This realizing two-legged robots, but three, four, or multiple leg
is more evident in the case of Str. 21, while in general in the robots could be achieved as well. Based on the obtained
Fig. 5. Nonlinear finite element method (FEM) analyses of appendage deformation behavior: FEM simulation setup (a, d), mesh model of the appendages (b,
e), deformation behavior (c, f). Results of FEM investigations: g) end-effector output displacement, h) realized Geometric Advantage, i) realized Mechanical
Advantage.

solutions (Fig. 4) and FEM investigations (Fig. 5a, d), 3D while the PVA (polyvinyl alcohol) soluble material filament
solid models of the two-leg soft compliant mechanism robots is used for the support fabrication (Fig. 6c). Thus, the multi-
are designed (Fig. 6a). The following subsections describe a extrusion 3D printing is realized with two materials, one for
fabrication process and prototyping of the two-leg Combots. the Combot structure and the other for the supports. Depending
Further, the experimental investigations of robots’ deformation on the complexity and density of the Combot structure, the
behavior, characteristics, and locomotion capabilities under 3D printing process time took on average 20 hours. After
various conditions are presented. the production, the soft robots are submerged in a water-
based solution, to release the main Combot structure from the
A. Prototyping and fabrication support material (Fig. 6c).
The robot fabrication is realized via the fuse deposition Based on multiple trial-and-error approach experimentation
modeling process by utilizing a commercially available 3D with various materials, fabrication of different robot designs,
printer Ultimaker (Fig. 6b). Substantial research efforts are and considering the needed actuation forces to deform the
made to determine the suitable material and printing process Combot structure, it was determined that Ultimaker PP fil-
parameters for the Combot prototypes. Several filament mate- ament led to the best results.
rials are explored: ABS, Ultimaker PP, BASF PP, and different This takes into consideration robot functionality and print-
settings for the printing process (values of the material Young ing quality: small actuation forces are needed to achieve
modulus are given in Tab. A.2). As Combot designs represent a deformation, while having a structure firm enough to realize
complex structure comprised of spatially connected beam-like robot locomotion. Thus, PP filament was adopted as the final
elements, the supports need to be used to yield the production material for the soft compliant mechanism robot’s fabrication.
of a firm robot structure. The supporting structure geome- Fig. 6 shows the 3D printing process (Fig. 6b), produced
try is automatically generated via Ultimaker Cura software, structure with supports, the release of the support material
Fig. 6. Prototyping and fabrication of the Combot soft compliant mechanism robots: a) two-legged soft robot designs, b) fused deposition modeling process –
3D prating, c) produced structure with supports and realize of the supports, d) final produced robot prototype based on solution Str.5 – Robot 1, e) produced
prototype based on the appendage solution Str. 21 – Robot 2, f) baseline design and produced robot prototype.

(Fig. 6c), and final prototypes of the two selected Combot lution with solenoid actuation. This is also done to demonstrate
designs (Fig. 6d, e). that Combot locomotion does not come from the solenoid ac-
To realize autonomous actuation of the Combots, an appro- tuation/vibrations only and that appendages mechanism design
priate actuator is needed. Motivated by the aim to achieve high has a large influence on the overall robot motion capabilities.
locomotion speeds of the Combots, while having relatively The baseline design is realized in form of two-beam elements
large available actuation forces (in compact design) with and a robot body where the solenoid is attached, Fig. 6f,
modes power consumption, we adopted the electromagnetic the same filament material is used as in the case of Combot
actuation principle in form of a common solenoid. The actua- fabrication. In such a design, a motion transmission ratio is GA
tion in form of solenoids is rarely applied or used in the soft = 1, where the MA≈1, meaning that there is no amplification
robotics field, thus offering new insights into how these actu- of the applied input displacement. The hypothesis is that low
ators can be utilized to drive soft robots. The solenoids offer values of GA will lead to slow locomotion speeds of the
advantages of realizing relatively large actuation forces and overall robot proving the benefits of realizing soft compliant
high actuation speeds (by controlling the solenoid frequency) mechanism robots through the proposed synthesis approach.
while requiring reality low actuation power. Some drawbacks
are a poor ratio of actuator overall size to possible actuator B. Deformation behavior
stroke while being relatively heavy compared to Combot The measurement setup for experimental investigation of
structure. The solenoid with 5 mm available actuation stroke the Combot deformation behavior (output motion of the ap-
is attached to the soft compliant mechanism robot structure pendage end-effector) is shown in Fig. 7a, d. The robot body
(at predefined dedicate space), where the solenoid plunger is is fixed above the ground to enable the free movement of the
connected to the robot input port. The actuator position is appendages, while two rulers are placed to visually represent
secured by fixing (gluing) the solenoid body to the Combot the values of applied input and realized output displacement
body structure. Fig. 6d, e shows the final prototypes of soft of the appendage end-effector Fig. 7a (in case of measuring
compliant mechanism robots with integrated actuation. output displacement in the Z direction, one ruler is used
Additionally, to the optimized designs in Fig. 6a, a simple Fig. 7d). Additionally, the value of applied input voltage for
intuitively developed solution (baseline design) is produced the solenoid is recorded. The input voltage is supplied to
as well (Fig. 6f). This is motivated by the aim to prove the the solenoid in increments of 1 V (in range of 0 – 25 V
advantages of the proposed synthesis approach and Combot for Robot 1, and 0 – 15 V for Robot 2), while in each
locomotion capabilities when compared to a simple design so- step the solenoid stroke (input displacement) and end-effector
Fig. 7. Experimental investigation of soft compliant mechanisms robot deformation behavior. Setup for measurement of end-effector displacement in the Y
direction (a). Robot 1 (b) and Robot 2 (c) are shown in initial and deformed positions (X-Y plane). Setup for measuring output displacement in the Z direction
(d). Robot 1 (e) and Robot 2 (f) are shown in initial and deformed positions (Z-Y plane). Measurement results for different investigated cases (g-k).

displacement is measured. Computer Vision software is used to archives a larger value of displacement in both Y and Z direc-
track the trajectory of the appendage end-effector point while tions. For intermediate values of actuator stroke, higher values
also allowing to realize accurate displacement measurements. of GA are realized in the case of Robot 2 (Fig. 7h). When
The same measurement procedure is applied for all analyzed full actuator stroke is reached, a similar range of GA values is
Combot designs (Fig. 7b, c, e, and f). Fig. 7g shows results of realized in the case of both Robot 1 and Robot 2 (Fig. 7h). This
end-effector output displacement in Y and Z direction (similar is contrary to FEM simulation results (Fig. 5h). In the case of
to Fig. 5g), and Fig. 7h shows the realized GA values com- Robot 1, results are not in good agreement with simulation,
pared to applied input displacement, for both Combot designs. due to the manufacturing imperfections of the robot design
Further, the correlation between solenoid input stroke and and connection problems between the solenoid and robot input
required input voltage (Fig. 7i), and the correlation between port. Fig. 7k shows the comparison between results obtained
the input voltage and realize output displacement in the Y with optimization, FEM simulation, and measurements, for
direction (Fig. 7j) are plotted. As could be seen (Fig. 7h), Robot 2. As could be seen there is a good agreement between
in general, GA values are dropping with input displacement FEM and measured displacement of the end-effector point.
increment, where both Robots 1 and 2 realize a similar range Based on the results in Fig. 7i, j, it could be concluded
of end-effector displacement (Fig. 7g). Surprisingly Robot 2 that a larger input voltage is required to actuate Robot 1
Fig. 8. Experimental investigation of soft compliant mechanism robots locomotion capabilities and speed: (a) Robot design 1, (b) Robot design 2, (c) baseline
design (robot position is shown in different time intervals and at different actuation frequencies), (d) comparison of locomotion principles of robot 1 and
robot 2. See video 1.

(based on Str. 5) compared to Robot 2 (based on Str. 21).

These results agree with trends observed in Fig. 5i, where Str.
21 achieve better MA values compared to Str. 5 (meaning it
requires smaller input forces to actuate/deform the structure).
This proves that the synthesis case when both GA and MA
are considered in the optimization leads to more energy-
efficient Combot solutions, that also can realize relatively
high GA values. Although there is a difference when results
are compared to measurements, overall, the optimization can
lead to appendages solutions that realize similar deformation
behavior when the solutions are experimentally tested (both
Combot designs realize end-effector displacement in the de-
sired direction as the set goal of the optimization).
C. Terrestrial locomotion
The Combot designs (Fig. 6d, e) locomotion characteristics
Fig. 9. Comparison of locomotion speeds at different control frequencies for
and capabilities are investigated by realizing motion on the the different robots shown in Fig. 8.
flat smooth surface. The ruler is used to visually represent the
robot’s locomotion speed (Fig. 8). Here, several experiments
are done:
adding anti-slip pads to the end-effectors to increase the
• actuating the solenoid manually by pressing a switch,
friction between the ground and robots.
where a constant voltage is supplied to the actuator,
• actuating the solenoid via a controller where a constant The same experiments are done for all Combot designs
voltage is supplied, and the actuation frequency is varied (Fig. 6d, e). Fig. A.1, Fig. 8d shows the robot operation princi-
(in range of 2.5 to 33 Hz), ple i.e. locomotion principle captured in different consecutive
• for both cases (manual actuation and using the controller), time steps. Two motion phases could be observed, which we
the Combot locomotion is investigated without and with will refer to as the active and the passive phases:
 In the first active phase, when the voltage is supplied i.e. there is a good trade-off between GA and MA values.
to the actuator the solenoid input stroke is transmitted to Compared to baseline design (Fig. 8c) Robot designs 1 and
the end-effector via appendages mechanism structure. The 2 (Fig. 8a, b) realize higher locomotion speeds (Fig. 9).
appendages push the robot (with the solenoid) agents to the
ground (Fig. A.1), realizing the lift and forward motion of
the robot body (due to designed end-effector free motion D. Locomotion with payload
in Y and -Z direction, Fig. 7c, f). In the second passive
To prove that introduced concept of soft compliant mecha-
phase, when the solenoid is turned off, the stored deformation
nism robots can also carry a payload while realizing locomo-
energy of the appendage’s mechanism, together with the elastic
tion (Fig. 10), similar investigations are done like in the case
energy of the deformed solenoid spring, is released. This
of (Fig. 8). A container in the shape of a box is 3D printed and
provides, in some portion, additional push forward momentum
filled with small items (bolts), the payload values are shown
(work) enabling the whole Combot structure to detach from
in Fig. 10d. The experiments are realized by placing/attaching
the ground and realize forward jump-like motion (Fig. A.1,
the container on top of the Combot solenoid, where the robot
Fig. 8d) (considering that returning path of the end-effector
locomotion capabilities are then tested at different actuation
is revers of the graph in Fig. 7g, thus not having contact
frequencies (Fig. 10a, b). These investigations are done for
with the ground). By leveraging these two effects/phases, high
all Combot designs. Fig. 10a, b shows soft robots realizing
locomotion soft robot speeds can be achieved.
locomotion, captured in different time intervals, while the
Realized Combot locomotion (all robot designs) for differ-
realized locomotion speeds are plotted in Fig. 10c. Based on
ent investigated cases and at different actuation frequencies,
the results it could be concluded that in all cases robots can
is shown in Fig. 8 (captured in different time intervals).
realize stable locomotion and carry a payload. Not surprisingly
Fig. 8 shows results when pads are added to the robot end-
smaller values of locomotion speeds are achieved (Fig. 10c),
effectors to increase the friction with the ground, but the same
with similar trends like in the case of Combots without the
investigations are realized for robots without pads (Fig. A.1).
additional payload (Fig. 9).
Achieved robot locomotion speeds are plotted in Fig. 9.
Based on the results, several trends could be observed. In the case of Robot design 2, high locomotion speeds up
Not surprisingly, higher locomotion speeds are realized at to 3.03 BL/s could be realized (Fig. 10b).
higher actuation/solenoid frequencies, for most Combot de-
signs (Fig. 9). In the case of Robot 2 (Fig. 8b), it is noted that E. Underwater environment locomotion
at frequencies above 30Hz, the locomotion becomes unstable,
as the solenoid induces high vibrations into the system, having The inherent monolithic structure of the soft compliant
phases where appendage end-effectors do not realize contact mechanism robots and type of used actuation, allow Combots
with the ground, thus leading to drops in locomotion speed. to operate in other environments besides terrestrial. To experi-
Still, relatively high velocities are achieved. mentally investigate this, robots are fully submerged underwa-
Comparing results with and without an anti-slip pad, in ter (Fig. 11). Similar investigations are realized like in the case
general robots achieve higher speeds when the pads are added of terrestrial locomotion (Fig. 8, 10). Combots are actuated via
to the robot, due to increase friction with the ground (Fig. 9). the controller by varying the actuation frequency (Fig. 11). The
Interestingly, Combot design 2 can achieve high locomotion same investigations are done for both robot designs. Fig. 11
speeds up to 3.76 BL / s (body length per second, where shows Combot locomotion capabilities realized underwater
body length is measured in direction of robot motion), at (captured in different time intervals). The achieved locomotion
an actuation frequency of 20Hz, outperforming some of the speeds are plotted in Fig. 11c. Similar deformation behavior
existing soft locomotion robots (even at the lower solenoid is exhibited like in the case of terrestrial locomotion, but with
frequencies of 10Hz) [21]. Additionally considering that robots several additional effects. Due to the different environments,
are carrying relativity heavy solenoids. Combot design 2 robots realize more jumping-like motion, with several phases
(Fig. 8b), realizes higher locomotion speeds compared to of the robot body being detached fully from the ground when
Combot design 1 (Fig. 8a). Even when comparing Robot 2 robots are actuated. Still, the robot body is heavy enough to
with no pads and Robot 1 with pads (Fig. 9). This is because bring the Combot to the ground. In general, results show
in Robot design 2 both GA and MA are optimized (based that the introduced concept of soft compliant mechanism
on Str. 21 in Fig. 4, 5d), leading to a more energy-efficient robots can realize stable locomotion even underwater. Not
solution, as more of maximal available solenoid force is at surprisingly, the locomotion speeds are lower than in the
Combot disposal (in the case of Combot design 1, a maximal case of terrestrial operation (Fig. 9). Robot design 2 (on
possible solenoid actuation force is needed to drive the robot). average) realizes higher locomotion speeds compared to Robot
Additionally, to this, due to higher MA values in the case design 1, reflecting similar trends like in (Fig. 9, 10c).
of Robot 2, the transmitter output force at the appendages At actuation frequencies of 20Hz, both robot designs realize
end-effector is larger, leading to pushing the robot body with slower locomotion, due to solenoid inducing higher vibrations
larger forces. Moreover, results show that even relatively lower into the system thus Combots lose contact with the ground
values of GA can lead to faster soft robots if MA is optimized and realize unstable motion.
Fig. 10. Experimental investigation of robot locomotion when carrying a payload: (a) Robot design 1, (b) Robot design 2, (c) comparing robot locomotion
speed, (d) weight of the payload and overall robot. See video 2

VI. D ISCUSSION means that the synthesis framework can be executed at

practical speeds, avoiding the computationally expensive
Overall, from our experiments with the synthesis frame- nonlinear FEM simulations for every candidate solution
work, the FEM validation, the prototyping, and the locomotion evaluation.
experiments, we observe the following: • Prototyping. Finding the right material for fabrication
• Synthesis approach. Overall, the EA design approach al- required substantial trial and error. It was necessary to
lowed us to explore a range of different designs according find good a tradeoff bewtween the required input force
to our different specified constraints, and solutions had and the structural stiffness, which both affect the robot
good performance. The EA managed to find solutions locomotion. For our type of actuator, we ended up with
that were good tradeoffs when both GA and MA were Ultimaker PP as a good tradeoff. For other types of
considered in the objective function, whereas when con- actuation, other materials may have better properties and
sidering only on GA, the resulting MA performance was should be explored.
relatively low. Potentially, a native multi-objective EA • Real-World Locomotion. The results from real-world lo-
such as NSGA-II [82] could produce a better pareto front comotion experiments demonstrated successful transfer of
of design tradeoffs, and should be investigated in future the synthesized solutions, and shoved that the proposed
studies. Likewise, while the discovered solutions were synthesis approach can lead to an optimized soft com-
relatively diverse, potentially Quality-Diversity EA vari- pliant mechanism solution that realizes better locomotion
ants [83] could facilitate a more systematic exploration capabilities than a simple intuitive design. Moreover, it is
of various design alternatives. also clear that the combined optimization of motion (GA)
• FEM Validation. We can see that for the solution where and force (MA) transmission ratios, can lead to new soft
GA was optimized to a high value, a higher discrepancy robot solutions that realize fast locomotion, outperform-
was observed. This is due to the large displacements that ing the design considering only GA. When considering
led to a large deformation behavior of the appendages. other optimization methods, taking into account both GA
These effects could not be captured by the linear analysis and MA is likely to lead to better performing solutions.
during the optimization process. A similar effect could be The proposed concept of soft compliant mechanism lo-
seen for the MA, for the solution with the highest MA comotion robots is also capable of carrying a certain
value. Although there is a difference between the results payload, which is rarely explored for other soft robots
predicted by the optimization process and the nonlinear [7, 18, 21, 50]. Moreover, although several concepts of
FEM simulations, similar trends could be observed in soft robots that can realize locomotion both on the ground
both cases, meaning that the optimization could still and underwater have been introduced [3, 30, 50], the main
capture the desired behavior of the appendages. This limitation is slow locomotion. The locomotion speed
Fig. 11. Experimental investigation of robot locomotion underwater (robot location is shown in a different time interval and at various actuation frequencies):
(a) Robot design 1, (b) Robot design 2, (c) comparison between robot achieved locomotion speed, (d) plain view of Robot design 2 in the water container.
See video 3

with payload of 3.03 BL/s even outperformed some of and fabricated via fused deposition modeling by using a
the existing soft locomotion robots [21], and underwater commonly available 3D printer.
speed is also relatively high. This can further broaden the Various experimental investigations are done, testing the
application scope of the Combots. robot locomotion capabilities. The results show that both
For different robot structures, payloads, and environ- robots could successfully locomote, however the design con-
ments, different actuation control strategies may be opti- sidering both GA and MA had the best performance for
mal. Tuning the shape and frequency of these could be both terrestrial locomotion, locomotion with a payload, and
subject of future studies. underwater locomotion.
Future work will include developing a systems framework
VII. C ONCLUSION
where soft compliant mechanism robots are designed based
This paper introduces a different approach to realizing soft on interaction with the environment and their locomotion
locomotion robots –Combots– by utilizing a net of thin- capabilities, realized in a simulated environment. We will also
elastic beams, distributed in form of a spatial compliant consider more advanced optimization methods and different
mechanism, and fabricated from common PP plastic material appendage designs that can realize directional locomotion.
on a commercial 3D printer. The inherent material properties
as rapid store/release of deformation energy, coupled with ACKNOWLEDGEMENTS
electromagnetic actuation in form of a common solenoid,
allow us to achieve a higher locomotion speed of soft robots, This work was supported by the Academy of Finland
compared to some existing concepts. The spatially connected Research Council for Natural Sciences and Engineering, grant
beam-like elements with the possibility to form different robot number 318390, and partially supported by the Research
topologies allow exploration of a much broader design space Council of Norway through its Centres of Excellence scheme,
compared to other types of soft locomotion robots. project number 262762.
A synthesis framework for obtaining automated design of
R EFERENCES
soft locomotion robots is presented. By utilizing the developed
systems, different solution designs are obtained, investigated [1] B. Jumet, M. D. Bell, V. Sanchez, and D. J. Preston, “A
under various problem setups. data-driven review of soft robotics,” Advanced Intelligent
Based on the obtained results, two Combot designs were Systems, vol. 4, no. 4, p. 2100163, 2022.
selected for experimental testing of robot locomotion capabil- [2] C. Laschi, B. Mazzolai, and M. Cianchetti, “Soft
ities. A two-leg soft complaint mechanism robot was designed robotics: Technologies and systems pushing the bound-
 aries of robot abilities,” Science Robotics, vol. 1, no. 1, vol. 3, feb 2017.
p. eaah3690, 2016. [17] M. Jiang, Z. Zhou, and N. Gravish, “Flexoskeleton print-
[3] S. I. Rich, R. J. Wood, and C. Majidi, “Untethered soft ing enables versatile fabrication of hybrid soft and rigid
robotics,” Nature Electronics, vol. 1, no. 2, pp. 102–112, robots,” Soft Robotics, vol. 7, pp. 770–778, dec 2020.
2018. [18] Q. He, Z. Wang, Y. Wang, A. Minori, M. T. Tol-
[4] Y. Kim, G. A. Parada, S. Liu, and X. Zhao, “Ferromag- ley, and S. Cai, “Electrically controlled liquid crystal
netic soft continuum robots,” Science Robotics, vol. 4, elastomer–based soft tubular actuator with multimodal
no. 33, p. eaax7329, 2019. actuation,” Science Advances, vol. 5, p. eaax5746, oct
[5] D. H. Kim, Y. Lee, and H.-S. Park, “Bioinspired high- 2019.
degrees of freedom soft robotic glove for restoring versa- [19] S. Felton, M. Tolley, E. Demaine, D. Rus, and R. Wood,
tile and comfortable manipulation,” Soft Robotics, vol. 9, “A method for building self-folding machines,” Science,
no. 4, pp. 734–744, 2022. PMID: 34388039. vol. 345, pp. 644–646, aug 2014.
[6] B. W. K. Ang and C.-H. Yeow, “Design and modeling [20] C. du Pasquier, T. Chen, S. Tibbits, and K. Shea, “Design
of a high force soft actuator for assisted elbow flexion,” and computational modeling of a 3d printed pneumatic
IEEE Robotics and Automation Letters, vol. 5, no. 2, toolkit for soft robotics,” Soft Robotics, vol. 6, no. 5,
pp. 3731–3736, 2020. pp. 657–663, 2019. PMID: 31173562.
[7] R. F. Shepherd, F. Ilievski, W. Choi, S. A. Morin, A. A. [21] Y. Tang, Y. Chi, J. Sun, T.-H. Huang, O. H. Maghsoudi,
Stokes, A. D. Mazzeo, X. Chen, M. Wang, and G. M. A. Spence, J. Zhao, H. Su, and J. Yin, “Leveraging
Whitesides, “Multigait soft robot,” Proceedings of the elastic instabilities for amplified performance: Spine-
National Academy of Sciences, vol. 108, pp. 20400– inspired high-speed and high-force soft robots,” Science
20403, nov 2011. Advances, vol. 6, p. eaaz6912, may 2020.
[8] Z. Wang, K. Or, and S. Hirai, “A dual-mode soft gripper [22] B. Gorissen, D. Melancon, N. Vasios, M. Torbati, and
for food packaging,” Robotics and Autonomous Systems, K. Bertoldi, “Inflatable soft jumper inspired by shell
vol. 125, p. 103427, 2020. snapping,” Science Robotics, vol. 5, p. eabb1967, may
[9] H.-C. Fu, J. D. Ho, K.-H. Lee, Y. C. Hu, S. K. Au, K.-J. 2020.
Cho, K. Y. Sze, and K.-W. Kwok, “Interfacing soft and [23] E. Milana, B. V. Raemdonck, K. Cornelis, E. Dehaerne,
hard: A spring reinforced actuator,” Soft Robotics, vol. 7, J. D. Clerck, Y. D. Groof, T. D. Vil, B. Gorissen,
no. 1, pp. 44–58, 2020. PMID: 31613702. and D. Reynaerts, “Eelworm: a bioinspired multimodal
[10] M. Asselmeier, R. L. Hatton, Y. Mengüç, and G. Olson, amphibious soft robot,” in 2020 3rd IEEE International
“Evaluation of a circumferential extending antagonist Conference on Soft Robotics (RoboSoft), pp. 766–771,
actuator in a soft arm,” in 2020 3rd IEEE International 2020.
Conference on Soft Robotics (RoboSoft), pp. 402–409, [24] M. Duduta, F. Berlinger, R. Nagpal, D. R. Clarke, R. J.
2020. Wood, and F. Z. Temel, “Tunable multi-modal locomo-
[11] A. Al-Ibadi, S. Nefti-Meziani, and S. Davis, “Design, tion in soft dielectric elastomer robots,” IEEE Robotics
kinematics and controlling a novel soft robot arm with and Automation Letters, vol. 5, pp. 3868–3875, jul 2020.
parallel motion,” Robotics, vol. 7, no. 2, 2018. [25] T. Park and Y. Cha, “Soft mobile robot inspired by
[12] Y. Hao, Z. Liu, J. Liu, X. Fang, B. Fang, S. Nie, Y. Guan, animal-like running motion,” Scientific Reports, vol. 9,
F. Sun, T. Wang, and L. Wen, “A soft gripper with oct 2019.
programmable effective length, tactile and curvature sen- [26] H. Peng, T. Mao, and X. Lu, “A small legged deformable
sory feedback,” Smart Materials and Structures, vol. 29, robot with multi-mode motion,” Journal of Intelligent
p. 035006, jan 2020. Material Systems and Structures, vol. 31, pp. 704–718,
[13] V. Subramaniam, S. Jain, J. Agarwal, and P. V. y Al- jan 2020.
varado, “Design and characterization of a hybrid soft [27] C. Tawk, M. in het Panhuis, G. M. Spinks, and G. Alici,
gripper with active palm pose control,” The International “Bioinspired 3d printable soft vacuum actuators for lo-
Journal of Robotics Research, vol. 39, no. 14, pp. 1668– comotion robots, grippers and artificial muscles,” Soft
1685, 2020. Robotics, vol. 5, pp. 685–694, dec 2018.
[14] J. F. Elfferich, D. Dodou, and C. D. Santina, “Soft robotic [28] T. Chen, O. R. Bilal, K. Shea, and C. Daraio, “Harnessing
grippers for crop handling or harvesting: A review,” IEEE bistability for directional propulsion of soft, untethered
Access, vol. 10, pp. 75428–75443, 2022. robots,” Proceedings of the National Academy of Sci-
[15] J. Guo, K. Elgeneidy, C. Xiang, N. Lohse, L. Justham, ences, vol. 115, pp. 5698–5702, may 2018.
and J. Rossiter, “Soft pneumatic grippers embedded [29] C. A. Aubin, S. Choudhury, R. Jerch, L. A. Archer, J. H.
with stretchable electroadhesion,” Smart Materials and Pikul, and R. F. Shepherd, “Electrolytic vascular systems
Structures, vol. 27, p. 055006, mar 2018. for energy-dense robots,” Nature, vol. 571, pp. 51–57, jun
[16] A. Fishman, M. S. Garrad, A. Hinitt, P. Zanini, T. Barker, 2019.
and J. Rossiter, “A compliant telescopic limb with [30] W.-S. Chu, K.-T. Lee, S.-H. Song, M.-W. Han, J.-Y.
anisotropic stiffness,” Frontiers in Robotics and AI, Lee, H.-S. Kim, M.-S. Kim, Y.-J. Park, K.-J. Cho, and
 S.-H. Ahn, “Review of biomimetic underwater robots Biomimetic supple and curvilinear spine driven by ten-
using smart actuators,” International Journal of Precision don,” in 2020 3rd IEEE International Conference on Soft
Engineering and Manufacturing, vol. 13, pp. 1281–1292, Robotics (RoboSoft), pp. 312–317, 2020.
jul 2012. [44] Z. Zhai, Y. Wang, K. Lin, L. Wu, and H. Jiang, “In
[31] A. Zolfagharian, L. Durran, S. Gharaie, B. Rolfe, A. Kay- situ stiffness manipulation using elegant curved origami,”
nak, and M. Bodaghi, “4d printing soft robots guided Science Advances, vol. 6, p. eabe2000, nov 2020.
by machine learning and finite element models,” Sensors [45] Y. Kim, Y. Lee, and Y. Cha, “Origami pump actuator
and Actuators A: Physical, vol. 328, p. 112774, sep 2021. based pneumatic quadruped robot (OPARO),” IEEE Ac-
[32] G. Singh and G. Krishnan, “Designing fiber-reinforced cess, vol. 9, pp. 41010–41018, 2021.
soft actuators for planar curvilinear shape matching,” Soft [46] Y. Xu, Q. Peyron, J. Kim, and J. Burgner-Kahrs, “Design
Robotics, vol. 7, pp. 109–121, feb 2020. of lightweight and extensible tendon-driven continuum
[33] X. Sui, H. Cai, D. Bie, Y. Zhang, J. Zhao, and robots using origami patterns,” in 2021 IEEE 4th Inter-
Y. Zhu, “Automatic generation of locomotion patterns for national Conference on Soft Robotics (RoboSoft), IEEE,
soft modular reconfigurable robots,” Applied Sciences, apr 2021.
vol. 10, p. 294, dec 2019. [47] X. Zou, T. Liang, M. Yang, C. LoPresti, S. Shukla,
[34] J. D. Hubbard, R. Acevedo, K. M. Edwards, A. T. M. Akin, B. T. Weil, S. Hoque, E. Gruber, and A. D.
Alsharhan, Z. Wen, J. Landry, K. Wang, S. Schaffer, Mazzeo, “Paper-based robotics with stackable pneumatic
and R. D. Sochol, “Fully 3d-printed soft robots with actuators,” Soft Robotics, aug 2021.
integrated fluidic circuitry,” Science Advances, vol. 7, [48] H. Lee, Y. Jang, J. K. Choe, S. Lee, H. Song, J. P.
p. eabe5257, jul 2021. Lee, N. Lone, and J. Kim, “3d-printed programmable
[35] J. Santoso, E. H. Skorina, M. Salerno, S. de Rivaz, tensegrity for soft robotics,” Science Robotics, vol. 5,
J. Paik, and C. D. Onal, “Single chamber multiple degree- no. 45, 2020.
of-freedom soft pneumatic actuator enabled by adjustable [49] J. Rieffel and J.-B. Mouret, “Adaptive and resilient soft
stiffness layers,” Smart Materials and Structures, vol. 28, tensegrity robots,” Soft Robotics, vol. 5, no. 3, pp. 318–
p. 035012, feb 2019. 329, 2018. PMID: 29664708.
[36] L. Meng, R. Kang, D. Gan, G. Chen, and J. S. Dai, [50] F. Corucci, N. Cheney, F. Giorgio-Serchi, J. Bongard,
“A shape memory alloy driven crawling robot utilizing and C. Laschi, “Evolving soft locomotion in aquatic and
a bistable mechanism,” in Volume 5B: 43rd Mechanisms terrestrial environments: Effects of material properties
and Robotics Conference, American Society of Mechan- and environmental transitions,” Soft Robotics, vol. 5,
ical Engineers, aug 2019. no. 4, pp. 475–495, 2018. PMID: 29985740.
[37] Y. Guo, L. Liu, Y. Liu, and J. Leng, “Review of dielectric [51] N. Cheney, R. MacCurdy, J. Clune, and H. Lipson,
elastomer actuators and their applications in soft robots,” “Unshackling evolution: Evolving soft robots with mul-
Advanced Intelligent Systems, p. 2000282, jul 2021. tiple materials and a powerful generative encoding,”
[38] B. Shin, J. Ha, M. Lee, K. Park, G. H. Park, T. H. Choi, SIGEVOlution, vol. 7, p. 11–23, aug 2014.
K.-J. Cho, and H.-Y. Kim, “Hygrobot: A self-locomotive [52] N. Cheney, , J. Clune, and H. Lipson, “Evolved elec-
ratcheted actuator powered by environmental humidity,” trophysiological soft robots,” in Artificial Life 14: Pro-
Science Robotics, vol. 3, p. eaar2629, jan 2018. ceedings of the Fourteenth International Conference on
[39] C. S. X. Ng, M. W. M. Tan, C. Xu, Z. Yang, P. S. Lee, the Synthesis and Simulation of Living Systems, The MIT
and G. Z. Lum, “Locomotion of miniature soft robots,” Press, jul 2014.
Advanced Materials, vol. 33, p. 2003558, dec 2020. [53] D. S. Shah, J. P. Powers, L. G. Tilton, S. Kriegman,
[40] M. W. M. Tan, G. Thangavel, and P. S. Lee, “Rugged J. Bongard, and R. Kramer-Bottiglio, “A soft robot that
soft robots using tough, stretchable, and self-healable adapts to environments through shape change,” Nature
adhesive elastomers,” Advanced Functional Materials, Machine Intelligence, vol. 3, no. 1, pp. 51–59, 2021.
p. 2103097, jun 2021. [54] J. Pinskier and D. Howard, “From bioinspiration to
[41] J. Tian, T. Wang, X. Fang, and Z. Shi, “Design, fabrica- computer generation: Developments in autonomous soft
tion and modeling analysis of a spiral support structure robot design,” Advanced Intelligent Systems, vol. 4, no. 1,
with superelastic ni-ti shape memory alloy for contin- p. 2100086, 2022.
uum robot,” Smart Materials and Structures, vol. 29, [55] D. E. Vogtmann, S. K. Gupta, and S. Bergbreiter, “Multi-
p. 045007, feb 2020. material compliant mechanisms for mobile millirobots,”
[42] F. Renda, C. Armanini, V. Lebastard, F. Candelier, and in 2011 IEEE International Conference on Robotics and
F. Boyer, “A geometric variable-strain approach for static Automation, IEEE, may 2011.
modeling of soft manipulators with tendon and fluidic [56] R. S. Pierre, N. Paul, and S. Bergbreiter, “3dflex: A
actuation,” IEEE Robotics and Automation Letters, vol. 5, rapid prototyping approach for multi-material compliant
no. 3, pp. 4006–4013, 2020. mechanisms in millirobots,” in 2017 IEEE International
[43] Y. Kakehashi, K. Okada, and M. Inaba, “Development Conference on Robotics and Automation (ICRA), IEEE,
of continuum spine mechanism for humanoid robot: may 2017.
[57] S. K. Patiballa and G. Krishnan, “On the design of three- tion, vol. 63, pp. 1351–1366, nov 2020.
dimensional mechanical metamaterials using load flow [71] P. Kumar, R. A. Sauer, and A. Saxena, “On topology
visualization,” Mechanics Based Design of Structures optimization of large deformation contact-aided shape
and Machines, pp. 1–26, feb 2020. morphing compliant mechanisms,” Mechanism and Ma-
[58] S. Patiballa, K. Uchikata, R. K. Ranganath, and G. Kr- chine Theory, vol. 156, p. 104135, feb 2021.
ishnan, “A conceptual design tool for synthesis of spatial [72] P. Kumar and M. Langelaar, “On topology optimization
compliant and shape morphing mechanisms,” in Volume of design-dependent pressure-loaded three-dimensional
5B: 42nd Mechanisms and Robotics Conference, Amer- structures and compliant mechanisms,” International
ican Society of Mechanical Engineers, aug 2018. Journal for Numerical Methods in Engineering, vol. 122,
[59] S. K. Patiballa, S. Satheeshbabu, and G. Krishnan, pp. 2205–2220, feb 2021.
“Load-flow based design of compliant mechanisms with [73] R. Ansola, E. Veguería, A. Maturana, and J. Canales,
embedded soft actuators,” in Volume 5A: 43rd Mech- “3d compliant mechanisms synthesis by a finite element
anisms and Robotics Conference, American Society of addition procedure,” Finite Elements in Analysis and
Mechanical Engineers, aug 2019. Design, vol. 46, pp. 760–769, sep 2010.
[60] G. Krishnan and S. K. Patiballa, “Conceptual insightful [74] X. Huang, Y. Li, S. Zhou, and Y. Xie, “Topology op-
synthesis of spatial compliant mechanisms using the timization of compliant mechanisms with desired struc-
load flow formulation,” Journal of Mechanical Design, tural stiffness,” Engineering Structures, vol. 79, pp. 13–
vol. 142, nov 2019. 21, nov 2014.
[61] M. I. Frecker, G. K. Ananthasuresh, S. Nishiwaki, [75] A. Milojević, M. Shin, and K. R. Oldham, “A novel
N. Kikuchi, and S. Kota, “Topological synthesis of design approach for micro-robotic appendages comprised
compliant mechanisms using multi-criteria optimization,” of active and passive elements with disparate properties,”
Journal of Mechanical Design, vol. 119, pp. 238–245, Journal of Intelligent Material Systems and Structures,
jun 1997. vol. 33, no. 1, pp. 136–159, 2022.
[62] S. Henning, S. Linß, P. Gräser, R. Theska, and L. Zent- [76] M. Grossard, C. Rotinat-Libersa, N. Chaillet, and
ner, “Non-linear analytical modeling of planar compliant M. Boukallel, “Mechanical and control-oriented design
mechanisms,” Mechanism and Machine Theory, vol. 155, of a monolithic piezoelectric microgripper using a new
p. 104067, jan 2021. topological optimization method,” IEEE/ASME Transac-
[63] N. T. Pavlović and N. D. Pavlović, “Compliant mecha- tions on Mechatronics, vol. 14, no. 1, pp. 32–45, 2009.
nism design for realizing of axial link translation,” Mech- [77] R. Parsons and S. Canfield, “Developing genetic pro-
anism and Machine Theory, vol. 44, no. 5, pp. 1082– gramming techniques for the design of compliant mech-
1091, 2009. anisms,” Structural and Multidisciplinary Optimization,
[64] M. P. Bendsoe and O. Sigmund, Topology Optimization: vol. 24, no. 1, pp. 78–86, 2002.
Theory, Methods and Applications. Springer, Feb. 2004. [78] D. E. Goldberg and M. P. Samtani, “Engineering opti-
[65] O. Sigmund, “On the design of compliant mechanisms mization via genetic algorithm,” in Electronic computa-
using topology optimization,” Mechanics of Structures tion, pp. 471–482, ASCE, 1986.
and Machines, vol. 25, no. 4, pp. 493–524, 1997. [79] S. Rajeev and C. Krishnamoorthy, “Discrete optimiza-
[66] B. Caasenbrood, A. Pogromsky, and H. Nijmeijer, “A tion of structures using genetic algorithms,” Journal of
computational design framework for pressure-driven soft structural engineering, vol. 118, no. 5, pp. 1233–1250,
robots through nonlinear topology optimization,” in 2020 1992.
3rd IEEE International Conference on Soft Robotics [80] S. Rajan, “Sizing, shape, and topology design opti-
(RoboSoft), IEEE, may 2020. mization of trusses using genetic algorithm,” Journal of
[67] E. M. de Souza and E. C. N. Silva, “Topology optimiza- structural engineering, vol. 121, no. 10, pp. 1480–1487,
tion applied to the design of actuators driven by pressure 1995.
loads,” Structural and Multidisciplinary Optimization, [81] K. Deb and S. Gulati, “Design of truss-structures for min-
vol. 61, pp. 1763–1786, mar 2020. imum weight using genetic algorithms,” Finite elements
[68] J. Hiller and H. Lipson, “Automatic design and manu- in analysis and design, vol. 37, no. 5, pp. 447–465, 2001.
facture of soft robots,” IEEE Transactions on Robotics, [82] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast
vol. 28, pp. 457–466, apr 2012. and elitist multi objective genetic algorithm: NSGA-II,”
[69] B. Zhu, X. Zhang, H. Zhang, J. Liang, H. Zang, H. Li, IEEE transactions on evolutionary computation, vol. 6,
and R. Wang, “Design of compliant mechanisms using no. 2, pp. 182–197, 2002.
continuum topology optimization: A review,” Mechanism [83] J. K. Pugh, L. B. Soros, and K. O. Stanley, “Quality
and Machine Theory, vol. 143, p. 103622, jan 2020. diversity: A new frontier for evolutionary computation,”
[70] P. Kumar, C. Schmidleithner, N. B. Larsen, and O. Sig- Frontiers in Robotics and AI, vol. 3, p. 40, 2016.
mund, “Topology optimization and 3d printing of large
deformation compliant mechanisms for straining biolog-
ical tissues,” Structural and Multidisciplinary Optimiza-
 A PPENDIX
A. Design parameter space
Design parameters for the synthesis of the appendages are
listed in Table A.1.
B. Material and 3D printing parameters
The materials used for experimentation and their corre-
sponding 3D printing parameters are listed in Table A.2.
C. Locomotion steps
Key frames from videos of the locomotion cycle for the
different robots are shown in Fig. A.1.
 TABLE A.1
D ESIGN PARAMETERS FOR SYNTHESIS OF SOFT COMPLIANT ROBOTIC APPENDAGES .

Design parameters Case 1 Case 2 Case 3

Domain size (length x width x height) 50 x 30 x 20 mm 50 x 30 x 20 mm 50 x 50 x 30 mm
node grid size nx × ny × nz = 3 × 3 × 2
degree of nodal connectivity 1
total number of elements 89
input node 13
end-effector node 9 9 3
support nodes 1, 4, 7, 10, 16
input displacement din = 5 mm
node wandering region size vx × vy × vz vx = 1 ÷ 2 mm, vy = 1 ÷ 2 mm, vz = 1 ÷ 2 mm
element dimensions (w x h x t) 1 x 1 x 1 mm
element material (Young modulus) E = 800 MPa
external loads Fx , Fy , Fz = 1 N
minimal desired end-effector displacement ddes
out = 1 mm
desired structure density Ldes
tot = 123 ÷ 230 mm

TABLE A.2
F ILAMENT MATERIALS AND 3D PRINTING PARAMETERS FOR ROBOT FABRICATION

Filament materials Ultimaker ABS Ultimaker PP Ultrafuse® PP Natural (BASF) Ultimaker PVA
Modulus (MPa) 2070 305 470 - 554 -
Nozzle 0.4 AA
Quality (mm) 0.06 0.08 0.08 0.08
Line width (mm) 0.35 0.38 0.38 0.35
Infill 20% - triangle 20% - octet 20% - octet 20% - triangle
Printing Temp 225 205 230 215
Plate temp 80 85 75 85
Printing speed (mm/s) 50 25 25 35

Fig. A.1. Locomotion principle: (a) Robot 1 design, (b) Robot 2 design. The columns containt key points in the locomotion cycle. Cases are shown for
manual control and when using a controller, and without and with pads to increase friction.