Professional Documents
Culture Documents
Doctoral Thesis
Author(s):
Svetozarevic, Bratislav
Publication Date:
2018
Permanent Link:
https://doi.org/10.3929/ethz-b-000309667
Rights / License:
In Copyright - Non-Commercial Use Permitted
This page was generated automatically upon download from the ETH Zurich Research Collection. For more
information please consult the Terms of use.
ETH Library
bratislav svetozarevic
S O F T - R O B O T I C - D R I V E N A D A P T I V E P H O T O V O LTA I C
BUILDING ENVELOPES
diss. eth no. 25379
S O F T- R O B O T I C - D R I V E N A D A P T I V E
P H O T O V O LTA I C B U I L D I N G E N V E L O P E S
presented by
bratislav svetozarevic
2018
Bratislav Svetozarevic: Soft-Robotic-Driven Adaptive Photovoltaic Building
Envelopes , © 2018
doi:
To my beloved ones ...
P R E FA C E
Let me start off with some impressions about the environment in which
this work was done, the benefits and challenges it brought, and the aspects
in which I saw a beauty that made me quickly forget some hard moments
and kept me (and still keeps me) motivated to do research.
Doing research and engineering today is substantially different than how
it was just some 10 or 20 years ago, and already next year I would have
been able to write this preface differently. The world is changing at an
unprecedented pace driven by an immense amount of new discoveries
and knowledge created every day. This new information becomes available
around the globe just milliseconds after it is uploaded to the internet. At
the same time, novel tools, measurement devices, data banks, mathematical
models, algorithms, and software packages are being developed, deployed,
and adopted into practice within a fraction of a year. All of these allow
researchers to develop and test ideas faster, easier, and in greater detail
than ever before.
On the other hand, mastering any topic requires not only long hours of
studying, but more challengingly continued learning about new discoveries.
Furthermore, it is increasingly becoming the case that the most interesting
discoveries happen at the intersections of several disciplines rather than
within a single discipline. This requires researchers not only to be familiar
with one or more additional disciplines but also to start mastering them
in order to be able to more easily spot synergies. Consequently, a multidis-
ciplinary team is a fruitful environment that allows one to quickly learn
about other topics through direct interactions with researchers from other
disciplines, while having a chance to reflect and question learnings from
one’s own field.
Finally, mastering one or more disciplines and knowing their boundaries
is necessary but not sufficient for doing research. A researcher is supposed
to expand the boundaries of the known by envisioning a next feasible point
vii
beyond those boundaries. Such a point should be of benefit for everyone.
The excitement after finding such a point and seeing people reacting to it is
irreplaceable feeling. The soft-robotic-driven adaptive photovoltaic building
envelope was one such project.
viii
ABSTRACT
x
Z U S A M M E N FA S S U N G
xii
ACKNOWLEDGEMENTS
xiv
CONTENTS
1 introduction 1
1.1 Adaptive Building Envelopes 1
1.2 Soft-Material Actuators 2
1.3 Research Questions 3
1.4 Research Overview 4
1.4.1 Design of a single-body 2-DOF soft-material actuator
with corrugated walls 4
1.4.2 Fabrication of soft actuators 5
1.4.3 Hybrid soft-hard-material actuator with variable stiff-
ness 5
1.4.4 Control algorithms 6
1.4.5 Soft-robotic-driven adaptive photovoltaic envelopes 6
1.4.6 Modular Pneumatic Control System 7
1.4.7 Energy consumption 7
1.5 Organisation of the thesis 8
1.6 Contribution from other authors 9
2 soro-track: a two-axis soft robotic platform for so-
lar tracking and building-integrated photovoltaic
applications 11
2.1 Introduction 12
2.2 Design Specification and Fabrication 13
2.3 Modelling and characterisation 16
2.3.1 Kinematic Model 16
2.3.2 Dynamic Model 18
2.3.3 Robustness to Wind Loads 20
2.4 Solar Tracking 22
2.4.1 Orientation Control 22
2.4.2 Experiments 23
2.5 Conclusion 25
3 soft robotic driven photovoltaic building envelope
for adaptive energy and comfort management 27
3.1 Introduction 28
3.2 Working principle and use cases of soft robotic building
envelope 29
3.3 Two-axis hybrid soft-hard-material pneumatic actuator 30
xv
xvi contents
3.4Envelope control 32
3.5Envelope performance in real weather conditions 34
3.6Energy benefits of the envelope as part of the building service
system 37
3.7 Discussion 38
3.8 Methods 40
3.8.1 Fabrication of soft-material actuator 40
3.8.2 Finite-element analysis of soft actuator 41
3.8.3 Pneumatic control system 43
3.8.4 Pressure-deflection measurements 44
3.8.5 Repeatability of motion measurements 46
3.8.6 Variable stiffness experiment 47
3.8.7 Variable damping experiment 47
3.8.8 Cycles before break measurements 48
3.8.9 Orientation feedback control of a single facade mod-
ule 48
3.8.10 Solar tracking experiments 50
3.8.11 Electricity consumption measurements of the pneu-
matic control system 52
4 quick-cast: a method for fast and precise scalable
production of fluid-driven elastomeric soft actua-
tors 61
4.1 Introduction 62
4.2 Methods and materials 64
4.2.1 Actuator design 64
4.2.2 State-of-the-art manufacturing methods 65
4.2.3 Quick-cast manufacturing method in laboratory con-
ditions 66
4.2.4 Quick-cast manufacturing method in industry 69
4.3 Results and discussion 69
4.3.1 Pressure-deflection characterisation 70
4.3.2 Repeatability of motion 71
4.3.3 Comparison with state-of-the-art methods 71
4.4 Conclusions 73
4.5 Acknowledgements 75
4.6 Appendix A. Supplementary data 76
5 coupling of hard and soft robotics: hybrid metal-
elastomeric pneumatic 2-dof actuators with vari-
able stiffness 81
contents xvii
5.1 Introduction 82
5.2 Pressure-deflection analysis 86
5.2.1 Methodology 86
5.2.2 Results 87
5.3 Response to external distrubance 88
5.3.1 Methodology 88
5.3.2 Results 90
5.4 Stiffness analysis 92
5.4.1 Methodology 92
5.4.2 Results 95
5.5 Pressures-to-stiffness modeling 96
5.5.1 Methodology 96
5.5.2 Results 99
5.6 Pressures-to-angles modeling 102
5.6.1 Methodology 102
5.6.2 Results 103
5.7 Model-based feedforward control 106
5.7.1 Methodology 106
5.7.2 Results 107
5.8 Conclusion 108
5.9 Appendix A. System overview 112
5.10 Appendix B. Fabrication of 2-DOF joints 113
6 conclusion 115
6.1 Summary of results 115
6.2 Discussion 117
6.3 Outlook 119
bibliography 121
N O TAT I O N
symbol meaning
α, ϕ soft actautor roll angle
θ soft actautor pitch angle
pi pressures in chambers
k i , k ij , Stiffness coefficient
di , dij , δi , δi,j Damping coefficient
ωn Natural frequency
Φ0j moment of inertia of the SoRo-Track
Φmj moment of inertia of the mounted PV plate
Td Dominant time constant
s Complex frequency
G (s) Transfer function
xviii
notation xix
abbreviations
How people design and construct buildings has not changed in essence
ever since the construction of the first shelters. In its basic definition, a
building is considered as a barrier, an envelope, separating the inside from
the outside environment. Ideally, it protects the occupants from harsh
weather conditions, such as wind, rain, snow, hail, and cold, and also from
intense solar radiation, by keeping the internal space dry and at a tem-
perature comfortable for people.This is achieved by combining insulating
and permeable materials in a vast number of designs to utilise the outside
weather conditions in the best possible way. However, for a large part of
the year and for most places, it is also necessary to condition the internal
space, either by heating it up or cooling it down to reach the comfort range.
Over the years, insulation materials and building systems for heating and
cooling have improved. Today, about one-third of global primary energy
is used for conditioning indoor spaces, producing nearly a quarter of all
man-made greenhouse gas emissions [1]. According to reports from the In-
tergovernmental Panel on Climate Change [1] and the International Energy
Agency [2], the energy-saving potential has been estimated at 50-90%. This
could be achieved by improving buildings infrastructure, by using more
energy efficient technologies, and by more efficient operation of building
management systems, ideally without sacrificing occupants’ comfort. For
building envelopes in particular, efforts and investments need to scale up
dramatically to improve performance by 30% by 2025 [2].
Current building envelopes are predominantly static, without integrated
elements for harvesting solar energy, and manually operated by occupants.
This leaves a large potential for energy savings, on-site energy genera-
tion, and improvement of occupant comfort untapped. Realizing the full
potential of building envelopes requires a robust and cost-effective way
to actively change envelope behaviour at high spatio-temporal resolution.
However, even though various approaches to dynamic envelopes exist,
they are limited to theoretical studies, small-scale prototypes, and single,
exhibition-type prototypes [3, 4]. The existing kinetic envelopes are typically
driven by complex actuating mechanisms comprising of a large number of
1
2 introduction
Soft robotics is a rapidly growing field of robotics that makes use of elastic
and flexible materials and principles causing their deformation for the
construction of robotic components (e.g. actuators and end effectors), parts
of robot bodies (e.g. links and skin), and even entire robots [5, 6]. There is a
plethora of “soft” materials available from different material classes, such
as elastomers, which are intrinsically soft, and also metals and plastics that
may be ysed in specific designs to create extrinsically soft materials (e.g. thin
metal sheets, wires of shape-memory alloys, thermoplastics, and granules).
These materials can be designed to be deformed by a variety of actuation
principles, such as pressure, heat, humidity, magnetic field, and electric field.
This large design space has inspired researchers to develop soft machines
that, compared to their rigid-body counterparts, allow reduced control
complexity in complex tasks (e.g. grasping [7]), more natural, animal-like
motions [8], and new features (e.g. camouflage [9] and climbing walls [10]).
Among soft actuators, the most popular types are fluid-driven elas-
tomeric actuators (FEAs) [11]. FEAs contain voids in their bodies that are
used as channels for pressurised fluid (gas or liquid). When pressurised
fluid is applied, the elastomeric material surrounding the voids expands
according to Pascal’s law, causing a change in the shape of the soft actua-
tor. FEAs are popular because they exhibit a continuum of motion, large
strokes, low friction, low weight, high power-to-weight ratio, and intrinsic
compliance. Therefore, FEAs are very similar in functioning to biological
muscles, carrying great potential for applications on a scale directly useful
to humans, with examples successfully demonstrated in industrial automa-
tion [7], medicine [12] , and human-robot interaction [13]. Furthermore,
FEAs exhibit reduced control complexity due to their intrinsic compliance
1.3 research questions 3
• What are optimal 2-DOFs soft robotic actuators that can reliably orient
the single facade element in various weather conditions, particularly
in wind?
• How could one vary the stiffness of the soft actuator, without increas-
ing the complexity of the actuation system, only by just using the
available compressed-air?
• How could one mass produce these actuators at high quality and
repeatable performance?
• What are the optimal control algorithms of the soft actuator for au-
tonomous solar tracking?
Several state-of-the-art methods were tested with the aim of easily produc-
ing the corrugated single-body three-chambered actuator, called corrugated
SoRo-Track. However, all of them showed limitations, either due to diffi-
culties in achieving leak-free actuators, failing in achieving the required
geometrical complexity, limited support for different types of elastomers,
high process complexity, or long fabrication time. A novel method for fast
and precise scalable production of soft actuators with a moderate com-
plexity of fluidic pathways is proposed in this work. The method is called
Quick-cast. This method reduces the total fabrication time required for the
production of one actuator from 8-10 hours to less than one hour and it
supports a wide variety of elastomers.
Two variants of this method were developed for two settings. One method
was developed for fast prototyping in laboratory conditions with 3D printed
moulds using liquid multi-component silicone rubbers. The other method
was developed for an industrial setting and is based on a standard industrial
process, compression-moulding. It was tested with Neoprene rubber. Tens of
corrugated SoRo-Track actuators were produced using both processes and
then they were compared among each other within each batch to examine
the repeatability of quality and performance. The industrial process showed
advantages over the laboratory process over a large number of samples: the
pressure-deflection curves overlap very well for ten actuators, randomly
selected from a batch of more than 50.
Chapter 4 introduces a novel method for quick and precise mass produc-
tion of soft actuators, called Quick-cast.
Chapter 5 introduces novel hybrid soft-hard-material two degrees of
freedom actuator (DOFs) pneumatically-driven actuators with variable
stiffness, called Hybrid-pneuVSAs. This section also characterises Hybrid-
pneuVSAs in terms of pressure-deflection, stiffness, and damping. Finally,
a pressure-to-angle/stiffness model is developed and feedforward model-
based control is obtained.
Svetozarevic, B., Nagy, Z., Hofer, J., Jacob, D., Begle, M., Chatzi, E., & Schlueter, A. SoRo-Track:
A Two-Axis Soft Robotic Platform for Solar Tracking and Building-Integrated Photovoltaic
Applications. IEEE International Conference on Robotics and Automation (ICRA) (2016)
11
12 soro-track
2.1 introduction
Figure 2.1: Adaptive Solar Facade at the House Of Natural Resources, ETH Höng-
gerberg Campus in Zürich, Switzerland, actuated by 50 individually
addressable soft robotics solar trackers. (www.honr.ethz.ch)
Two-axis solar tracking requires orienting the PV panel along the azimuth
and altitude angles (see Section 2.3). For this purpose, three cylindrical
chambers are symmetrically distributed around the centre of the actuator
(Fig. 2.2a and 2.2b). The surrounding body that contains the chambers,
cylinder A, is expanded at two horizontal levels, forming two thick circular
disks, to allow mounting a PV module and mounting of the actuator onto
the cantilever (Fig. 2.2g).
When a single chamber is inflated, it expands the most where its walls
are the thinnest – which is to the outside, creating pressure on the top and
the bottom disks. This pressure causes the top platform to rotate around
14 soro-track
CIGS PV Panel
Backplate
Meshring
Junction-Box
Adapter
Soft - Actuator
Cantilever
Cable Net
PV Power Cable
the motion limiting part of the actuator, which is the remaining part of
the cylinder A that contains the other two chambers. Notice that the axis
of rotation is parallel to the line defined by the centres of the other two
chambers, and lies close to the edge of the cylinder A. However, it is not
fixed relative to the bottom disk, but changes with the pressure in the
chamber. Therefore, the rotation of the top platform is considered relative
to the platform orientation when all chambers are deflated and the two
disks are parallel.
The main performance requirements for solar tracking applications are
the range of motion of SoRo-Track in both angles and its ability to carry the
PV panel in outdoor weather conditions, which are expressed through the
stiffness and damping parameters of SoRo-Track. The design parameters
can be divided into geometrical and material parameters. The geometrical
parameters are the diameter dch and height hch of the cylindrical chambers,
the diameter dcyl and height hcyl of the cylinder A, and the diameter dcir of
the circle on which the centres of the chambers lie. The material parameters
are hardness Shore A index, tensile strength, and elongation at brake. The
thickness of the top and bottom disks should be chosen to prevent the
deformation of the chambers in the vertical direction. The diameters of the
top and bottom disks, as well as the diameter and height of the cylinder
B, where the inlet tubes are located, can be freely selected to achieve
mechanical stability of the fixations.
2.2 design specification and fabrication 15
In this section we discuss the kinematic and dynamic modelling of the actu-
ator. We use quasi-static pressure-deflection curves and shaker experiments
to identify the models. Then, the dynamic model is used to simulate the
behaviour of SoRo-Track in wind.
The two rotations of SoRo-Track, the azimuth (θ) – around the vertical
axis of the PV module and the altitude (ϕ) – around the horizontal axis
of the module, and their relation to the sun angle are shown in Fig. 2.4.
We develop the model phenomenologically and give the mathematical
description below.
When chamber 1 is being inflated, the top platform will start rotating
around the axis which is parallel to the line defined by the centres of
chambers 2 and 3, causing the altitude angle ϕ to increase. If the actuator is
produced perfectly radially symmetric, the azimuth angle θ should remain
zero. Therefore, by inflating chamber 1 solely, only positive altitude angles
can be achieved. This is represented on the azimuth-altitude graph as the
main direction of actuation of chamber 1 (Fig. 2.4b). Similar holds for
the other two chambers, by rotating the actuation direction by 120◦ . To
obtain a combined azimuth-altitude (θ, ϕ) orientation, a certain pressure
combination in two of the three chambers needs to be applied. Inflating the
third chamber while two other chambers are pressurised will counteract
the rotation already obtained by these two chambers. This implies that
the azimuth-altitude plane is divided in regions by the main directions of
actuation of individual chambers (Fig. 2.4b). In brief, the kinematic model
of the SoRo-Track is given by
" # " #
θ (p) θ(p)
= R·A· (2.1)
ϕ(p) ϕ(p)
φ
Equator
2.3 modelling and characterisation 17
S
θref θ
a)
N
Polar Axis Main direction φ (altitude)
of actuation
Sun of Ch 1
PV cell
" #
R1 R2 R3
R= (2.2)
R1 R2 R3
where
1
pi = 0 ∧ pmod(i,3)+1 ≥ 0
Ri = ∧ pmod(i+1,3)+1 > 0, i ∈ {1, 2, 3}, (2.3)
0 else.
where
0 m12 m13 0 n12 n13
A1 =
m21 0 ,A2 = n21
m23 0 ,
n23 (2.5)
m31 m32 0 n31 n32 0
18 soro-track
20 30 30 Single chamber
θ3(p3) Model
φ1(p1) Ch 1 Two chambers
20
20
10 Ch 1 & 2 Ch 1 & 3
θ (azimuth) [°]
φ (altitude) [°]
φ (altitude) [°]
10
10
θ1(p1)
0 0
0
−10
−10 φ3(p3) Ch 2 Ch 3
−10 −20
θ2(p2) φ2(p2)
Ch 2 & 3
−20 −20 −30
0 0.5 1 1.5 0 0.5 1 1.5 −30 −20 −10 0 10 20 30
a) Pressure [bar] b) Pressure [bar] c) θ (azimuth) [°]
k ij dij 1
κij = ; δij = ; λj = (2.9)
Φ0j + Φm
j Φ0j + Φm
j Φ0j + Φm
j
where {i, j} ∈ {θ, ϕ}, k ij are spring coefficients, dij are damping coeffi-
cients, Φ0j is moment of inertia of SoRo-Track, Φm j is moment of inertia
of the mounted PV plate, t is the total external momentum acting on the
actuator, and φ is the angle between x-axis and t. We assume that the
angular velocity of one rotation does not affect the other rotation, and set
δθ ϕ = δϕθ = 0. This leads to eight unknown parameters to identify.
To identify the parameters, two experiments were conducted using the
setup shown in Fig 2.6. First, the resonance frequency of SoRo-Track was
determined using a free oscillation experiment to be about 20 Hz. Then, a
forced oscillation experiment is performed using a shaker for parameter
identification over the frequency range [0, 4] kHz. The resonant frequency of
SoRo-Track including the PV module is approximately 3 Hz (Fig. 2.7a). The
system identification was performed in frequency domain using MATLAB
System Identification Toolbox.
Notice that the spring and damping coefficients are modelled as functions
of pressures in the chambers. The identification was carried out for different
pressure configurations: { p1 , p2 , p3 } ∈ {0.5, 0.8, 1.1} bar. The values of
spring and damping coefficients as functions of the pressures in chamber 1
and 2, for constant pressure in chamber 3, are shown in Fig. 2.7b. The plots
for a constant pressure in other chambers are similar. The plots show how
high damping and high stiffness are mutually exclusive.
20 soro-track
4 5 6 7
3
1 2
12 11 10
Pressure Ch 1
DO In
FPGA loop tank Out Valves Ch 2
module Ch 3
Figure 2.6: Experimental setup for identification of the dynamic model parame-
ters.
After the identification and validation of the model (by comparing the
frequency responses of both experimental methods), SoRo-Track behaviour
in wind can be simulated using appropriate wind data as input. Two
specific actuator configurations are selected for wind simulation: (i) high
stiffness - low damping, occurring at high pressures in the chambers, and
2.3 modelling and characterisation 21
Magnitude [dB]
0
High spring coeff.
−20 High damping coeff.
−40
−60
0
−45
Phase [°]
−90
−135
−180
a) 100 Frequency (Hz) 101
p2 [bar]
p2 [bar]
Average config. 2
25 1 High spring coeff.
20 High damping ceoff.
Angle [º]
15
10
0
10 11 12 13 14 15 16 17 18
c) Time [s]
(ii) low stiffness - high damping, occurring at low pressures in the chambers
(Fig. 2.7b). The transfer functions of these two actuator configurations show
that the most robust pressure configuration depends on the spectrum of the
wind that is exciting the actuator (Fig. 2.7a). For wind frequencies below
the resonance frequency of the actuator, high stiffness is favourable, i.e.,
the chambers should be pressurized. For frequencies above the resonant
frequency, the chambers should be deflated. This is also illustrated in
Fig. 2.7c, where in area 1 high damping is more beneficial than high
stiffness and in area 2 the opposite is the case.
22 soro-track
For given sun angles, solar tracking is achieved if the azimuth and altitude
angles of the SoRo-Track are aligned with the sun angles (Fig. 2.4). For this
purpose we developed an orientation control algorithm and tested it in
outdoor solar tracking experiments.
Figure 2.8: SoRo-Track – A two-axis SRA for solar tracking applications. Left:
Back view on the SoRo-Track. Top right: Front view on the SoRo-
Track. Bottom right: Sparkfun breadboard with IMU 9150 sensor from
InvenSense.
2.4 solar tracking 23
φ (altitude) [o]
(1) inflation Ch 1 5
Ch 2 R1
Ch 3 0
Start
θ [o]
−5
Ch 1 (azimuth)
210° 330°
−10 R2 R3
0 20
R4 R5
5 10
φ (altitude) [o]
R5 R4 R1
0 End 0
R2 R3
−10
Angle Reference
R1 −20
0 50 100 150 200 250 300
Start
5
10
θ (azimuth) [o]
R3 R4
0 R2 R3
0
5 R1 R2 R5
0 −10
20 −10 0 10 20 0 50 100 150 200 250 300
θ (azimuth) [o] c) Time [s]
chambers are constructed through A, called guiding lines (red dashed lines).
Then, one of the chambers is inflated until SoRo-Track reaches one of the
guiding lines (blue arrow 1). Then, the second chamber relevant for that
region is inflated, moving along the second guiding line (blue arrow 2),
until the reference point is reached. Analogous strategies are implemented
in the other two regions (points B and C). Typical experimental results of
the reference tracking are shown in Fig. 2.9b and 8c.
2.4.2 Experiments
The sun tracking experiment is set on the roof of the authors’ office building
in Zurich (47°240 N, E8°300 E) and was performed on a clear day in August.
24 soro-track
The sun position for this location and time is calculated with Rhinoceros
3D using the DIVA plugin [39]. During the experiment, the altitude and
azimuth angles of the module were adjusted every 5 minutes. For each time
step, the module DC power output at maximum power point (MPP) was
measured using the I-V-characteristic measurement device PVPM 2510C
from Photovoltaik Engineering.
In addition to solar tracking, the module power output at the reference
position (PV module tilted 12◦ w.r.t. the facade in the altitude angle, and
oriented to the South) was measured for each time step, which represents
the case of conventional building-integrated PVs (Case I, Fig. 2.10). For this
sunny day, the measured power gain from solar tracking was 36%. The
experimental results correspond very well to the model of the PV power
output. In the model, irradiance on the tilted module plane is calculated
considering the three components of beam, diffuse sky, and ground reflected
radiation [40]. The DC power output of the PV module is calculated by
taking into account the actuator range, the active PV area, and its electrical
conversion efficiency.
Changing the reference position to parallel to the facade of the building,
the model predicts a potential gain of 61% (Case II). And if the SoRo-Track
azimuth range could be expanded from ±20◦ to ±45◦ , the predicted power
gain is 73% (Case III). Note that the energy gain from solar tracking is
100
90
Normalised PV Power Output [%]
80
70
60
Case I (power gain 36 %)
50
Case II (power gain 61 %)
40 Case III (power gain 73 %)
30
measurements
20 reference orientation: 12º towards the facade
reference orientation: parallel to the facade
10 two-axis sun tracking (limited actuator range)
two-axis sun tracking (extended actuator range)
0
10 11 12 13 14 15 16 17 18 19
Hour of day [h]
specific for the meteorological condition at this location and day. (See video:
https://vimeo.com/152272960)
2.5 conclusion
Buildings account for the largest share of global primary energy consump-
tion and nearly one quarter of all greenhouse gas emissions. Thus, increas-
ing buildings’ energy efficiency and providing on-site renewable energy
generation is key to climate change mitigation and to creating sustainable
cities. Among building elements, the envelope has the most influence on
heating, cooling, and lighting demands, as well as on occupant comfort.
Current envelopes are predominantly static and are only marginally able
to adapt to changing conditions outside of the building and occupants’
needs inside of it. This leaves large potentials for energy savings, on-site
energy generation, and improvement of occupant comfort untapped. To
realize the full potential of building envelopes, a robust and cost-effective
way to actively change envelope behaviour in a high spatio-temporal res-
olution is needed. Here we report on a novel soft-robotic-driven adaptive
building envelope that can offset the entire energy demand of the space
behind it during day-lit hours, while using only 3% or less of the energy
generated for its operation over the year. Independently addressable soft
robotic actuators allow for an unprecedented fine-grained manipulation of
envelope elements for optimal local solar energy generation, passive heat-
ing, reduction of cooling demands, and daylight utilization. Expanding on
the concept, if equipped with reflecting elements, the adaptive envelope can
be also used to redistribute solar radiation among neighbouring buildings
for thermal energy generation or mitigation of urban heat-island effects.
Our compact and cost-effective soft actuators are applied in a distributed
way on a large scale, laying the foundation for robust, lightweight, and
modular adaptive envelopes. The possibility to manage energy and comfort
on building and urban scales is expected to significantly reduce the energy
consumption of the built environment, towards near zero-energy buildings,
and improve both indoor and outdoor comfort.
Svetozarevic, B., Nagy, Z., Begle, M., Jayathissa, P., Caranovic, S., Shepherd, R. F., Hischier,
I., Hofer, J., & Schlueter, A. Soft robotic driven photovoltaic building envelope for adaptive
energy and comfort management. Nature Energy. In Revision. (2018)
27
28 soft robotic driven adaptive building envelope
3.1 introduction
The building sector accounts for the largest share of global primary en-
ergy consumption (32% in 2010) and produces nearly one quarter of all
greenhouse gas emissions [1]. These emissions may double or triple by
mid-century due to further urbanisation [1, 2]. Likewise, the energy-saving
potential of buildings has been estimated at 50-90% [1]. Consequently, im-
proving building stock has been identified as key to meeting future energy
and climate targets [41–43]. The stringent energy and emission standards
recently adopted in many OECD countries require profound changes to
convert buildings from major energy consumers to energy-neutral or even
energy-positive entities [2, 44]. In particular for building envelopes, efforts
and investments need to scale up dramatically to improve performance by
30% by 2025, as recently stated by the International Energy Agency (IEA)2.
The envelope is one of the most relevant building elements for reduc-
ing operational energy consumption. It is responsible for the transfer of
energy in the form of daylight, heat, and air between the two stochastic
environments: external, weather-driven and internal, occupant-driven. By
improving response to the two dynamic environments, building energy
consumption can be significantly reduced while maintaining a comfort-
able indoor environment [3]. Of particular interest is the control of solar
radiation as it has a significant impact on thermal comfort and daylight
quality. To do so, dynamic envelopes have been proposed [3, 4]. Compared
to current envelopes, they are able to adjust their permeability and can
adapt to changing environmental conditions and user demands at high
spatio-temporal resolutions. Even though various approaches to dynamic
envelopes exist, they are limited to theoretical studies, small-scale proto-
types, or single exhibition-type prototypes [3, 4].
A major challenge in bringing dynamic envelopes into the mainstream is
the cost and reliability of the actuation system. The system must be compli-
ant with the external weather conditions, requiring minimal maintenance
over the lifetime of the building. Silicone-elastomer-based actuators, as com-
monly found in soft robotics [23, 45–47], have been identified as a simple,
low cost, and robust solution. The toughness, relative chemical inertness,
and low glass-transition temperature of silicone elastomers make them ideal
for tolerating external environmental conditions, such as rain, cold, heat,
and wind [48]. Pneumatics is commonly used to power these actuators,
due to the speed and specific power of these systems [23, 48, 49]. In the
past, soft-material actuators have mainly been used for robotic applications,
3.2 working principle and use cases of soft robotic building envelope 29
climate, the envelope was able to cover 62% of the buildings energy demand
on a sunny day in winter and 270% on a sunny day in summer [54].
Equipped with reflective elements (Fig. 3.1c), the envelope becomes a
unique tool to manage energy and comfort on an urban scale. Excess solar
irradiation is reflected off a building envelope towards a nearby building or
the sky. Thus, solar energy is optimally distributed in space to minimize
the overall energy need of the built environment. In building areas lacking
solar irradiation, reflected solar irradiation can be directed onto neighbour-
ing buildings for indirect daylighting and heating [55]. Alternatively, the
radiation can be redirected onto highly efficient building-integrated hybrid
photovoltaic-thermal (PVT) collectors for simultaneous production of heat
and electricity [56]. Sunlight can also be concentrated towards single or
multiple focal points and used for applications such as concentrating pho-
tovoltaics (CPV) [57] or optical fibre-based daylighting systems [58]. Lastly,
reflection towards the sky provides an efficient method to mitigate urban
heat island effects [59–61]. An initial prototype of the reflective dynamic
solar envelope on a building at ETH Zurich in Switzerland is shown in
Extended Data Fig. 3.16.
Figure 3.1: Working principle and use cases of the soft robotic building envelope.
a, b, A room with a dynamic solar envelope on summer (a) and
winter (b) days and qualitative representations of physical effects
that the envelope can control. c, A room with a reflective dynamic
envelope. This envelope allows redistribution of solar irradiation and
daylight within the built environment, by reflecting it to a neighbour-
ing building or to the sky. d, A two-axis soft-robotics-driven envelope
module mounted on a rod-net structure. (i) Thin-film CIGS PV panel
or mirror, (ii) orientation sensor (inertial measurement unit - IMU),
(iii) PV junction box with a bypass diode, (iv) soft robotic actuator, (v)
universal joint, (vi) cantilever, (vii) pneumatic-control module, and
(viii) tubes holding electric cables and pneumatic tubes.
the chambers to use to move the actuator from the current orientation to the
desired one (see Methods). The soft actuator was able to achieve reference
tracking with ±2.5◦ precision without substantial overshooting.
For high-level control, a simulation framework, considering building
space and building systems, is used to minimize the net-energy demand [64].
Solar irradiation and building system performance are coupled to find the
optimal module angles for given weather conditions.
The envelope also responds to occupants’ requirements for visual comfort,
such as opening the envelope for view, closing the envelope for privacy, or
closing some parts of the envelope e.g. to avoid glare. A simplified user
interface has been integrated which allows to run the envelope in open,
closed, or auto mode. The latter will optimize for energy performance.
Alternatively, a more advanced user interface is available, which allows to
address parts of the facade or individual modules. It is implemented as a
graphical application and runs on a smart hand-held device or on a touch-
panel mounted on a wall. Finally, if user inputs are monitored within the
facade control system, a learning-based algorithm could be implemented to
provide a personalized envelope response [65].
A full scale 3.9 x 3.2 m envelope prototype with 16 modules has been built
and installed on a building at ETH Zurich in Switzerland (Extended Data
Fig. 3.13) to test the control, dynamic actuation, PV performance, and wind
stability in realistic weather conditions. Each module is equipped with
thin-film CIGS PV cells. Performance in solar tracking was assessed by
measuring PV power output (at maximum power point), global horizontal
irradiation, and vertical and horizontal angles of the modules at regular
time intervals. For comparison, performance of the modules at a vertical tilt
(i.e. angle relative to the façade) of 10° and 45◦ was measured in between
each time interval. These angles are close to positions of a static BIPV
facade in closed mode and with yearly optimal tilt (55◦ for this location),
respectively, however restricted by the range of the actuator. Details of the
solar tracking algorithm are given in the Methods section.
Results for a clear summer day are shown in Fig. 3.4 for a single façade
element and the entire façade, respectively. The facade is oriented 23◦ from
the south towards the west. Therefore, the PV energy gains from solar
tracking were larger in the morning than in the afternoon. The facade
started receiving solar radiation only after 10 AM due to the shadow of a
3.5 envelope performance in real weather conditions 35
Figure 3.3: Orientation feedback control of the envelope. a, Diagram of the enve-
lope’s orientation feedback control system. b, Control performance of
one module during the figure-of-eight reference tracking. Zig-Zag ref-
erence tracking algorithm with ±2.5◦ precision is applied. c, Schematic
diagram of the pneumatic control system. d, Tilt and azimuth angles
over time during the figure-of-eight reference tracking.
Figure 3.4: Solar tracking experiments with a single module (a,b,c,d) and the
entire envelope (e,f,g,h). a,b,The results show comparison of the
dynamic solar envelope in three different modes: (i) two-axis solar
tracking, (ii) static parallel to the wall, i.e. tilted at about 10◦ from
the wall, and (iii) static tilted at about 45◦ , which is close to a yearly
optimal tilt. The envelope was cycled through these three states
throughout the day. b,f, Global horizontal irradiation on a horizontal
surface. c,d,g,h, Statistics of the last two minutes of roll and pitch
angles in each of the three states in each cycle. Each box shows the
median 50% of all samples closest to the median value (Interquartile
range), and the two lines outside the box show the lowest and highest
observations. For single module solar tracking, the angle limit was
±2.5◦ , and for the whole facade solar tracking ±3◦ .
ation, wind speed, temperature, and humidity were taken during several
days in spring. Compared to static BIPV facades, electrical energy gains
between 28% and 50%, due to solar tracking, were measured. The exper-
iments demonstrate successful and robust functioning of the facade in
typical weather conditions for Zurich, Switzerland at outside temperatures
between 3◦ C and 20◦ C and various wind speeds up to 30 km/h. Relative
humidity does not have a major influence on the functioning of the soft
robotic actuators, and the glass transition temperature of neoprene rubber
is -50◦ C, which is low enough for use in most climates.
100 gain = 28% gain = 42% gain = 36% gain = 33% gain = 50%
Normalised PV Power Output [%]
90
80
70
60
50
40
30
20
10
0
Wind [km/h] Sol. irr. [W/m 2 ]
500
40
20
30
Temp. [ºC]
100
20 Rel. Hum. [%] 80
10 60
0 40
10 12 14 16 18 10 12 14 16 18 10 15 20 10 15 20 10 15 20
Hour of day [h] Hour of day [h] Hour of day [h] Hour of day [h] Hour of day [h]
3.7 discussion
Figure 3.6: Envelope energy balance and integration into a nearly-zero energy
building, HiLo, at Empa, Switzerland. a, A building-scale prototype
(4.0 x 2.5 m) of the dynamic solar facade driven by hybrid soft-
hard-material pneumatic actuators with 30 individually addressable
modules. b, Integration of the dynamic envelope within a nearly-zero
energy building HiLo at NEST at Empa research center in Dübendorf,
Switzerland. c, Carpet plot detailing the net energy balance of the
room behind the solar façade. The x-axis represents the months of
the year and the y-axis represents the net energy demand over hours
of an average day of the month. Solid lines indicate the sunrise and
sunset times. d, Room net energy balance during day-lit hours. e,
Building energy savings with the adaptive solar envelope installed.
with mirrors, the dynamic envelope can redistribute solar irradiation within
the built environment for optimal energy harvesting among neighbouring
buildings or to the sky, thus mitigating heat island effects. The dynamic
solar envelope bring possibility to manage energy and comfort on building
and urban scales, which could significantly reduce the energy consumption
of the built environment, towards near-zero energy buildings, and improve
both indoor and outdoor comfort.
3.8 methods
chambers to release the rubber from the walls of the chamber-bullet. Finally,
chamber-bullets are pulled out easily and a unitary body soft actuator is
produced. The last step consisted of gluing the pneumatic fittings into the
three bottom inlets.
This fabrication process was further optimised for a large-scale produc-
tion in an industrial setting. The fabrication process remained similar in
its essence; However, several adaptations, in terms of mould shape, rubber
material, and mould material, were required. The actuator fabricated in the
industrial setting used Neoprene rubber (black colour) and it is shown in
Fig. 3.2a.
reached this angle at about 1.4 bar during experiments. This difference
comes from the fact that we use the softer rubber, M4601, for simulation.
Therefore, we used FEA for relative comparison of different designs and,
consequently, for reducing the maximum stress in the material.
state of the entire envelope every 15 minutes, cycling between closed state
and tracking the sun (see Methods section on Orientation feedback control).
Due to the minimal number of the pneumatic components, this pneumatic
control system has also minimal costs. Its drawback is the number and
length of pneumatic lines, as each chamber needs to connect to the common
pressure distribution chamber.
In contrast to centralised pneumatic control system, decentralised ones
allow reduction of pneumatic lines and enable modularity of the system
(Extended Data Figure 3.8). The valves used for pressure distribution to
the soft actuator are now part of the façade element, located close to the
soft actuator. Such a distributed system requires, though, the development
of the distributed control electronics to enable the communication with
the centralised PLC [73]. The decentralised pneumatic control system of
minimal complexity is shown in Extended Data Figure 3.8a. This system can
achieve the same motion complexity as the above centralised control system.
However, the pressure distribution manifold has a minimum complexity
in this case, containing only two 2/2-way valves. Therefore, it can be
easily replicated to obtain another pressure regulation component in the
system and, in that way, enable parallel control of two groups of soft-
actuators. For the HiLo envelope (Fig. 3.6a), each row is connected to one
minimal-complexity pressure distribution manifold. This control system
presents a balance between the number of pneumatic components used
and achievable complexity of the envelope motion; In this case, each row
of the envelope can be independently controlled. For the maximal motion
performance, where each actuator is independently addressed at the same
time, a pressure distribution chamber needs to be integrated close to each
soft actuator, which requires at least five valves and a check-in valve, to
reduce the cross-talking between different pressure distribution manifolds
(Extended Data Fig. 3.8b).
After obtaining the roll and pitch measurements, quasi-static points were
extracted. For this purpose, an algorithm was developed in Matlab R2106b,
taking the average angles of the last 2s in each 20s step (Extended Data Fig.
3.9). After the quasi-static points were obtained, they could be plotted as
pressure-deflection curve (Fig. 3.2e).
The goal of this experiment was to show how the stiffness of the hybrid
soft-hard-material actuator changes with changing the internal parameter
of the actuator, which is the pressure applied to all three chambers. The
experiment was performed in laboratory conditions and, therefore, isolated
from any unknown external disturbances, such as wind. The soft actuator
with the thin-film PV module was mounted on an elevated platform at 45◦
and different weights were hung on a string attached to the corner of the
panel. Several pressure points of interest were identified: 0, 0.8, 1.0, 1.2,
1.4, 1.6 bar. Pressures between 0 bar and 0.8 bar were not analysed as an
insignificant change of actuator stiffness occurred in this pressure range.
After applying a certain pressure level to all three chambers, the three valves
connecting the chambers to the common pressure distribution manifold
were closed to prevent the air escaping the chambers when external torque
was applied. Then, weights in increments of 86g were loaded onto the string
and the resulting angle deflections from the initial roll angle (45◦ ) were
recorded after reaching the steady-state, i.e. after the initial oscillations,
caused by loading the weight, had disappeared. At a certain weight, the
hybrid soft-hard actuator reached its maximum deflection limit, at about
37◦ . This limit is due to the mechanical design of the outer U-joint. The load
was then completely removed and the pressure in all three chambers was
increased to the next level, following the pressure sequence defined above.
The incremental loading of the weights was then repeated. The results are
shown in Fig. 3.2f.
The variable damping experiment was performed using the same experi-
mental setup as for the variable stiffness experiment, but the experiment
was performed differently. The goal of the experiment was to show how
the damping of the actuator changes with changing the internal parameter
of the actuator, which is the pressure in all three chambers. To test the
damping, a preselected weight was hung on the rope, passing through the
48 soft robotic driven adaptive building envelope
metal ring at the base of the elevated platform on which the soft actuator
with the thin-film PV panel was mounted. The weight was preloaded on
the rope and instantly released using a quick-release pin. In this way, a step
response is generated. After performing some initial tests, the weight of
400g was selected as the weight where the U-joint reaches its mechanical
limit, while no pressure is applied to the chambers. As the first step of
the experiment, the pressure in all chambers was set to 0 bar. The three
valves supplying the pressure to the chambers were turned off in order to
block the air escaping the chambers when external load was applied. Then,
the weight was released and the roll angle was recorded at 20 Hz. The
oscillations were stabilised (2% steady state was achieved) after about 10s
and this time period was used as a recording window for the subsequent
experiments. The pressure in all three chambers was then increased to the
next value, following this sequence 0, 0.8, 1.0, 1.2, 1.4, 1.6 bar. The results of
this experiment are shown in Fig. 3.2g.
The cycles before break measurements were performed with the soft ac-
tuator with attached 40 x 40 cm CiGS thin-film PV panel including the
PV bracket, without PV cables passing through the middle of the actuator.
The actuator was tilted vertically at 45◦ , as it is mounted on the façade.
The experiments were performed in the laboratory conditions, without any
external disturbances or torque applied. The actuator was mounted such
that one chamber was at the bottom and this chamber was used to lift
the PV panel up from the default orientation. The actuator reached 30,000
cycles without breaking at the pressure of 1.6 bar 3.10. Assuming 4 cycles a
day in average, this corresponds to a life time of the actuator of 20 years.
At this pressure, the actuator lifts the panel up for 25◦ .
Figure 3.10: Cycling tests of the corrugated soft actuator (Fig. 3.2a) made of
Neoprene rubber. Life time is calculated assuming 4 cycles/day.
datagram protocol (UDP) packets over the local area network (LAN) to NI
PLC. Orientation angles are used in the real-time control loop programmed
in LabVIEW Real-Time, running at 20 Hz.
The Zig-Zag control algorithm has been developed in order to map the
three available actuation directions (Fig. 3.2d), generated by the individual
inflation of the three chambers, to two angles, pitch and roll [75]. Starting
from the initial position in roll-pitch plane, the algorithm computes which
chambers and in which way they need to be inflated in order to reach the
reference orientation. If the current orientation lies on the main direction
of actuation of a certain chamber, only that chamber will be used to move
the panel. If the current orientation lies in between two main directions of
actuation, both that chambers will be used to rotate the module. Due to the
simplicity of the pneumatic control system, where only a single pressure
regulator exists for a single actuator, the two chambers are used sequentially,
one at a time. This generates zig-zag motion of the facade module, shown
50 soft robotic driven adaptive building envelope
in Extended Data Fig. 3.11a. If the soft actuator was supposed to follow the
reference trajectory consisting of straight lines between some points, for
example, as in the figure-of-eight (A-B-C-D-E-F-B) reference tracking, large
overshoots and drifting away from the straight trajectory occur. This over-
shoots could be reduced by introducing an upper limit on the maximum
difference between the pitch and roll angles and the straight line connecting
the initial and final orientation. When such a limit is introduced, the over-
shoot can be drastically reduced. Experimental results show that the control
algorithm was able to respect the upper bound of ±2.5◦ at almost all times
during the transition phase (Extended Data Fig. 3.11b). The small overshoots
of about 1◦ occur for a very short time due to the bang-bang control strategy
used. This could be avoided if more sophisticated control algorithms were
used, such as proportional integral derivative (PID) control or model predic-
tive control [76]. Due to the narrower angle band, the switching between two
chambers occurs more often, which extends the transition time from 160s to
250s for following the whole trajectory. As the solar tracking requires updat-
ing the orientation every 5-10 min for only few degrees, the time difference
introduced by setting the upper limit on the tracking precision is negligi-
ble. The algorithm is available in an open access GitHub repository [77]:
https://github.com/BratislavS/ZigZagControlWithUpperBound
Figure 3.11: Orientation feedback control of the soft actuator performing the
figure-of-eight reference tracking. a, Roll-pitch and time diagrams
of the zig-zag reference tracking without an upper bound on the
maximum distance from the reference orientation. b, Roll-pitch
and time diagrams of the zig-zag reference tracking with an upper
bound on the maximum distance from the reference orientation.
Introducing the upper bound on the maximum distance from the
reference orientation reduces the overshooting.
experiment were performed using the HPZ envelope: with half of the
envelope (Ext. Data Fig. 3.12a) and the entire envelope (Ext. Data Fig.
3.12b). In the first case, the zig-zag orientation control algorithm without an
upper bound on angle tracking difference was used, resulting in overshoots
during transitions. In the second case, the zig-zag orientation feedback
control with the upper bound on angle difference during tracking was used,
52 soft robotic driven adaptive building envelope
Figure 3.13: Solar tracking experiment with the dynamic envelope with 16 mod-
ules. The envelope is mounted on the roof of HPZ building at ETH
Zurich, Switzerland. a, Measurements of the global solar irradia-
tion on a horizontal surface. b, Comparison of PV power output of
the envelope in three states: two-axis solar tracking, static facade
parallel to the wall, and static facade tilted at 45◦ . The envelope
cycles through the three states every 30 min (10 min per state). c,
Characteristic moments of envelope orientation during the solar
tracking experiment.
coming from the control electronics and pneumatic valves, and short peaks
of high power consumption from the air-compressor. There is a constant
power consumption of the control electronics, NI PLC and Raspberry Pi,
of about 14.3 W. The variable low power consumption from the pneumatic
valves adds to the power consumption of the electronics as a variable part
with the maximum at 17.5 W. The air compressor generates peaks of about
750 W for about 45s, when it refills the tank from 6 to 8 bar. One solar
tracking transition step is shown in zoomed-in graphs (Extended Data Fig.
3.14e-h). The transition step lasted 65s and valves used 35 mWh. These
values are used to estimate the yearly energy consumption of the experi-
mental control setup (Extended Data Table 1). As indicated in the table, the
majority of the total power consumption (34.4W) of the pneumatic control
system is used by the control electronics (22.5W, or 65%). This is due to
the fact that the control electronics was not selected with intention of low
power consumption and final implementation of the control system, but
rather for quick development of the control system and control software
implementation.
Therefore, there is a huge potential to design much more efficient pneu-
matic control system with off-the-shelf components with intention for a
final product design. A control electronics board could be developed utilis-
ing ultra-low-power microcontroller, such as one from Texas Instruments
MSP430™ family. For example, MSP430FR59941 requires 0.4 mW in normal
regime, 0.1 mW in stand-by with very fast wakeup time of 7 µs, and it has
enough computational power and number of inputs and outputs to control
the envelope. Also, the envelope control algorithms are not computationally
intensive and run at 5 Hz. This means that the power consumption of the
control electronics could be reduced to around 1% of the current power con-
sumption; however, due to additional Ethernet communication electronics
for monitoring of the envelope, we assumed that the power consumption of
the control system is reduced to 10% of the values for experimental setup
at HPZ roof.
Further, it was found that only 60% of the power consumed by the
ABAC air-compressor was used in the pneumatic system; other 40% was
wasted due to the leakage in the compressor. This stand-by leakage of the
compressor could be avoided to a large part by using an air-compressor
with the sealing of a higher quality. Further, more energetically efficient
electro-pneumatic valves are available on the market from SMC Pneumatics
S070 series with only 0.1 W power consumption, instead of the current 1
W of Festo MHA1 series. In order to obtain the final energy balance of the
3.8 methods 55
envelope, we compare the electrical energy that the envelope can generate
over the year with the electrical energy it uses for its operation.
The generated electrical energy is calculated based on the solar irradi-
ation data for south oriented façade surface in Zurich [78]. Based on the
above measurements, the self-consumption of the experimental control
setup was estimated at 30%. If the control setup is designed for low-power
consumption using only off-the-shelf components as explained above, the
energy consumption of the pneumatic control system is estimated at 3%
or less (see Extended Data Table 1). This number further depends on the
climate region; in climates with more solar irradiation, the self-consumption
of the envelope will be lower than 3%, due to higher solar irradiation per
m2. Also, if thin-film PV cells with the efficiency higher than 10% are used,
this will further reduce the self-power consumption below 3%.
56 soft robotic driven adaptive building envelope
Figure 3.15: Soft robotic driven dynamic envelope at HiLo. a, Perspective view
on the dynamic solar envelope. The top three rows are closed, while
the bottom three rows are open. b, Front view on the dynamic solar
envelope with the top rows open.
58 soft robotic driven adaptive building envelope
Figure 3.16: Initial prototype of the Reflective Dynamic Solar Envelope. Bot-
tom three rows of the HPZ dynamic envelope were equipped with
mirrors, while the top row is equipped with the CIGS PV modules.
3.8 methods 59
§
Geographical Assessment of Solar Resource and Performance of Photovoltaic
Technology [78]
†
Experimental setup at HPZ roof. The consumption values are measured (see
Extended Data Figure 3.14) and then scaled up to represent values for a façade with
30 modules
‡
Calculations with respect to (A) in the Table
Performing minimisation of the building net energy consumption (see Section on
Energy benefits of the envelope as part of the building service system.)
¶
Page 9, Compressed Air Evaluation Protocol, National Renewable Energy
Laboratory [79].
Q U I C K - C A S T : A M E T H O D F O R FA S T A N D P R E C I S E
4
SCALABLE PRODUCTION OF FLUID-DRIVEN
E L A S T O M E R I C S O F T A C T U AT O R S
Svetozarevic, B., Begle, M., Caranovic, S., Nagy, Z., & Schlueter, Quick-cast: A method for fast
and precise scalable production of fluid-driven elastomeric soft actuators. Extreme Mechanics
Letters. In Review (2018)
61
62 quick-cast manufacturing method
4.1 introduction
Soft robotics is a rapidly growing field offering novel actuators [80], end ef-
fectors [24], skins [9], and even entire robots [6] made of compliant material.
They carry several important advantages compared to classical rigid-bodied
robots, such as intrinsic compliance, large power-to-weight ratio, low fric-
tion, simpler control, and cheaper fabrication, and have enabled solutions
to challenging problems in automation (e.g. universal grasping [7]), archi-
tecture (e.g. adaptive solar façade [81]), medicine (e.g. minimally invasive
surgery [52]), and human-robot interaction (e.g. safe continuum manip-
ulation [82]). As contrary to rigid-bodied robots, where components are
selected from the standard set of components (e.g. electromotor, gearbox,
spring) and designs follow rigid-body kinematics principles (rigid links
connected by discrete joints), soft robotics offers much more design free-
dom, with a plethora of soft materials available (e.g. intrinsically compliant
– elastomers, and extrinsically compliant, such as wires of shape mem-
ory alloy [83], thermoplastics [84], and jamming particles [7]), a range
of actuation principles (fluidic-, thermal-, humidity-, pH-, magnetic-, and
electric-driven), and freedom in defining components shape, structure, and
working principle. Consequently, the main challenge in working with soft
components is in their design and manufacturing. Compared to design,
manufacturing techniques are directly linked to the performance of the
actuators, which often brought major advances in the past [15, 23, 85, 86].
Among soft actuators, one of most popular types are fluid-driven elas-
tomeric actuators (FEAs). These actuators contain fluidic pathways within
their soft bodies. When they are filled up with pressurised fluid (gas or
liquid), the surrounding elastic material expands, leading to a change of the
outer shape of the actuator. Depending on the geometry of fluidic pathways
and shape of the actuator, as well as on the pressurisation method (inflation
or deflation), different output motion paths can be obtained, such as expan-
sion, contraction, bending, or twisting [18]. Besides motion diversity, FEAs
exhibit advanced actuation features, similar to those of biological muscles,
such as continuum of motion, high power-to-weight ratio, large strokes,
and little friction. Also, due to use of elastomers, FEAs are intrinsically
compliant and can passively mitigate external disturbances (acting as a
spring-damper system), allowing for reduction of control complexity and
safe interaction with the environment and humans. For existing reviews of
FEAs, please see [11, 85].
4.1 introduction 63
pouring, the mixture was degassed to remove the trapped air-bubbles (Step
2). Then, the curing of elastomer was done in an oven at 75◦ C for about
20 min (Step 3). This is much faster than curing at the room temperature,
68 quick-cast manufacturing method
which takes several hours. Finally, the mould is disassembled in two steps
(Step 4). First, the two parts forming the outer shape of the actuator are
taken apart. Second, the inner cores are pulled out of the soft actuator (Fig.
4.8 and Fig. 4.10). Pulling out of the hard cores is possible due to the large
elongation at break of elastomers (in this case it is around 500%; may be up
to 1000%, e.g. of Ecoflex™00-50 from Smooth-On, Inc.). To allow for easer
pulling out of the inner cores, we were slightly inflating the chambers. The
total fabrication time of a single actuator, including the mould assembling,
takes approximately one hour. To produce the next actuator, one needs to
repeat steps 1 to 4. Obtained actuators may be then tested for performance
(Step C) and the information fed back to the modelling software to inform
the change of the design. The images of this manufacturing process for
cylindrical SoRo-Track and using VARIO are given in Fig. 4.8. Also, we
show the manufacturing of the same actuator with ELASTOSIL©M4601
(Fig. 4.9).
4.3 results and discussion 69
1 Injection
CNC - milling of the moulding
metal mould for Quick-casting ~15 min
2
rubber Unmolding
(Neoprene) Elastic material allows
production time: ~1 hour to pull out the centercore
~10 min
two-axis
soft (FEA) actuator
3
Curing at an
elevated temperature
~20 min
15/40 mixture for each actuator before pouring. The second reason is in the
insufficient precision of the 3d printed nylon mould, hence small offsets in
the fixation of the inner cores during mould assembly, as well as a slight
bending of the metal rods on which these cores are mounted. These metal
rods are rather thin (3 mm stainless steel rod) and they bend during the
manual pulling out of the inner cores (Fig. 4.8).
The pressure-deflection curves for corrugated SoRo-Track actuators made
from Neoprene rubber and manufactured industrially are shown in Fig.
4.4e-g. We can see that the pressure-deflection curves overlap for all 10
of tested actuators is very good (1% in pressure change for 12%-16% in
achieved angles). As the mould is micromachined from stainless steel, there
are no geometrical differences between these actuators. The only source of
uncertainty is in the preparation of the rubber mix, which, due to the early
development phase of establishing of this process, is still done manually
for each actuator separately. However, the absolute angle error is about +/-
3◦ , which is acceptable for controlling the system in feed-forward mode,
i.e. only using pre-calculated control inputs, without the need for feedback
signals from sensors, thus simplifying the control and reducing the actuator
implementation costs.
One of the actuators from each batch is randomly chosen for testing the
repeatability of motion (Fig. 4.4d and Fig. 4.4h). Chamber 1 was inflated
and then deflated 50 times. Both actuators show very good repeatability
over number of cycles.
that all parts of the wax cores stayed intact during demoulding them and
afterwards while pouring of elastomers over them in the main fabrication
step. In terms of using the rotational casting method [70], we could not
tune it for ELASTOSIL®M4601 to obtain the precise geometry of the inner
cores. However, this could have been due to our lack of expertise in this
method. The total time for manufacturing of a single SoRo-Track actuator
4.4 conclusions 73
using the above methods was between 6h to 10h, after basic preparations
were done. We are confident in these numbers, as all of the above methods
require multiple steps (between 3 and 4).
Based on our review of the state-of-the-art methods provided in Section
4.2.2 and based on the results of our Quick-cast method, we provide the
summary of the comparison in Table 1. With the table fields shaded in
grey we emphasize which methods show the best performance for each
feature analysed. Overall, our method shows comparable or advanced
performance across all features besides one (geometrical complexity of
fluidic pathways), where it shows moderate performance. Our method
shows advance performance in terms of number of steps (reduction to a
single step) and total fabrication time (reduction from one working day to
one hour).
In terms of the geometrical complexity of fluidic pathways, our process
has certain limitations defined by the materials properties and certain
geometrical parameters of fluidic pathways. The difference between the
radius of the entrance of the fluidic pathways and the maximum radius of
the fluidic pathways at any point, should not be larger than the elongation at
break of a soft material. Given that the elongation at break of soft materials
is in the range of 100% to 1000%, this may not be a critical limitation. The
elongation at break is inversely correlated with the elastomer hardness. This
means that for stronger soft actuators (with higher elastomer hardness), the
elongation at break is getting smaller. A designer of a soft actuator would
need to check if the design satisfies this limitation before fabricating the
actuator.
4.4 conclusions
4.5 acknowledgements
Figure 4.7: Exploded view of the nylon mould for the two-axis three-chambered
soft actuators (SoRo-Track) with cylindrical walls. For corrugated
walls, the outer moulds are the same, but the “bullets” forming the
internal voids are different (see Fig. 4.10).
78 quick-cast manufacturing method
Svetozarevic, B., Stojanovic, N., Begle, M., Nagy, Z., & Schlueter, A. Coupling of hard and soft
robotics: Hybrid metal-elastomeric pneumatic 2-DOF actuators with variable stiffness. Soft
Robotics. In Review (2018)
81
82 hybrid metal-elastomeric pneumatic 2-dof actuators
5.1 introduction
Figure 5.1: Coupling of hard and soft robotics: Overview of features of actuators
built from hard, soft and hybrid soft-hard materials, such as Hybrid-
pneuVSA.
5.2.1 Methodology
The basic purely soft material actuator is compared with the three hybrid
designs (IFU, ICU, and OU-joints) in order to assess the advantages and
limitations of each design. The soft actuator receives pressures in the three
chambers as control inputs which lead to different pitch and roll orienta-
tions. As the soft actuator is radially symmetric, we just show results for the
roll angle, where the main axis of actuation of chamber 1 is aligned with the
pitch angle of the IMU sensor (see Fig. 5.6a). The goal of this analysis is to
determine the shapes of the pressure-deflection curves and the maximum
angle reach (Fig. 5.3a). We applied pressure in a “stair-steps” profile going
from 0 to 190 kPa in increments of 20 kPa. When the maximum pressure
is reached, the pressure is decreased following the same profile (Fig. 5.3b).
The duration of each stair step is defined such that the actuator reaches
the steady state after the transition, in this case 5 seconds. The pressure-
deflection points are extracted as the steady state values before the next
step occurs (Fig. 5.3c).
5.2 pressure-deflection analysis 87
5.2.2 Results
The results show that different hard joints introduce different bending
restrictions. At the maximum applied pressure of 190 kPa, the outer U-joint
introduces the highest reduction of 37.6% compared to the maximum angle
reach of the purely soft actuator, while the internal joints, the internal
flexible and chained internal, introduce only minor reductions of 3.5% and
12.3%, respectively. This is an expected result, given the joints kinematics.
The outer U-joint introduces the highest restriction to bending due to the
inconsistency of the two fixed rotational axes with the bending axis of the
soft actuator. The upper fixed rotation axis forces the top plate of the soft
actuator to bend around it, instead of allowing the soft actuator to expand
in the axial direction and bend around an axis which is outside of the
actuator’s body. Contrary to this, the chained internal U-joint consists of
four 1-DOF joints of which two by two have parallel axes, thus following
better the bending path of the soft actuator. Finally, the flexible internal joint
is a continuum joint, which also expands slightly when bent, restricting the
original bending by only 3.5%.
All three hybrid structures inherit a hysteresis effect from the basic
structure. However, under the same pressure, the amplitude of the hysteresis
88 hybrid metal-elastomeric pneumatic 2-dof actuators
does not change with the actuator structure and stays mostly around a
few degrees, up to maximal 6◦ , when the same pressure is considered (Fig.
5.3a).
5.3.1 Methodology
Due to very low effective inertia of FEAs, a step response to pressure (system
input) change is typically without large oscilations, as one can also see in
Fig. 5.3c for the pressure changes of 20 kPa. Hence, the input-to-output
(pressure-to-angle) model of the actuator can be described as a first-order
model with the system gain determined by the pressure-deflection curve
(Fig. 5.3a), as we shown in our previous work on cylindrical three-chamber
actuator [75]. However, due to the elastic properties of soft actuators, the
external distrubances which regularly occure in interaction with the real-
world environment, have significant influence on the actuator’s dynamics.
In our aplication case, the disturbance is wind, which can transfer a rather
large force to the soft actuator due to a large area (40 cm by 40 cm) of the
attached PV panel.
We analysed the step response of the Hybrid-pneuVSAs to an external
disturbance by attaching a weight of 500 g to one corner of the PV panel as
sketched in Fig. 5.4 (the image of the experimental setup used is shown in
Fig. 5.6a). The weight of 500 g was chosen to generate sufficient deflection
but still not to drive the actuator to its bending limits (45◦ ). Then, the string
which connected the PV panel and the weight was cut, thus simulating
the external step disturbance. This procedure triggers a PV panel rotation
in one (roll) direction which is recorded by the inertial measurement unit
(IMU) attached to the PV panel. We assumed that the response to external
disturbancess (also known as behaviour around equilibrium positions for
elastic systems) can be modeled as a second order linear time invariant
(LTI) system [124] with the following transfer function:
K
G (s) = (5.1)
s2 + 2ζωn s + ωn2
1 x (t)
δ= ln (5.3)
n x (t + nT )
where T is the oscillation period (Fig. 5.3b), x (t) is the amplitude at time t
and x (t + nT ) is the amplitude of the peak n periods away. The damping
ratio ζ is then found from the logarithmic decrement by:
1
ζ= q (5.4)
1 + ( 2π
δ )
2
2π
ωn = p (5.5)
T 1 − ζ2
The step response of each of the four actuators is recorded seven times.
Each time with different nominal pressures in all three chambers (henceforth
called common chamber pressure), resulting in 28 step experiments in total.
Note that after changing the common chamber pressure, the pneumatic
2/2-way valves that connect the chambers with a pressure regulator had
to be closed so that the air cannot escape one chamber and distrubute
into the other two when the disturbance occurs. The applied chamber
pressure values were 0.0, 0.5 , 0.8, 1.0, 1.2, 1.5 and 1.9 bar. As a result of
these 28 experiments, one can distinguish differences in damping ratio
and dominant time constant for different chamber pressures for all four
actuator types. We also analysed the maximumal deflection caused by the
external load at different pressures, in order to show the static behavior of
the actautors as well.
90 hybrid metal-elastomeric pneumatic 2-dof actuators
Figure 5.4: (a) Sketch of the external disturbance step experiment setup. (b)
Pointes of interenest for logarithmic decrement algorithm. (c) Val-
idation of second order system approximation. Measured system
response compared to simulated response.
5.3.2 Results
has the steepest growth and allows for the largest change in damping rate.
This suggests that the actuator with the chained internal joint allows for
a wide possibility of internal parameter changes through the setting of
corresponding pressures.
In the end, the dominant time constant Td is calculated from ζ and ωn
for all actuators and all chamber pressure values (Figure 5.5d). Clearly, all
actuators express shorter transients with higher pressures. The lowest Td
value has the outer sturcture, for all pressure values. However, the chained
internal structure decreseas Td much faster when pressure is increased,
and almost has the same value of Td for chamber pressures of 1.9 bar.
Therefore, even though initially the outer structure exhibits better transient
performance at lower pressures, when the pressure increses, the chained
internal structure can achieve a similar performance as the outer structure,
with the tendency to have even shorter transients at pressures higher than
1.9 bar.
Stiffness is the property of a body that describes its ability to resist structural
deformation when external load is applied. A common example is a long
stick of a certain material where an external force F is applied in the axial
direction and causes an extension ∆l in the same direction. For this example,
Hooke’s law defines the stiffness coefficient as k = F/∆l. By analogy, one
can also define a relation between the bending moment M and the angular
deformational response. In this way the bending stiffness is defined as
the ratio between applied moment and angle change as krot = M/α. Even
though stiffness represents a simple relation between applied load and
caused deformation, things can become complicated and ambiguous if the
loads and deformation directions are not defined in a strict way.
In the previous example, for most materials, the stick’s cross section will
contract while it expands axially under applied force. Therefore, one can
also search for a stiffness coefficient between the force applied in x-direction
and the structure deformation in any direction (e.g. y-direction, z-direction).
Furthermore, the origin and direction of a force vector F bring additional
degrees of freedom which creates numerous possible definitions for the
stiffness coefficient. Still, one can fully describe a local deformation by
using three direct strain components and three shear strain components
which yield a 6x6 material stiffness matrix [126]. When rotational types
of stiffness are also taken into account, things become even more intricate
and obtaining all stiffness coefficients for a full description of a system,
caan be extremely demanding and time consuming. Fortunately, in order to
draw some conclusions about a system only one specific or a few stiffness
coefficients are needed. To avoid confusion and ambiguity when the stiffness
coefficient values are analyzed, the force (moment) vector defined, as well
as the measured deformation, must be clearly defined.
5.4.1 Methodology
The experimenal setup for finding the stiffness coefficient of the actuator is
shown in Fig. 5.6a. It consists of a hybrid soft-hard actuator with a thin-film
PV panel, an IMU sensor MPU-9250 from InvenSenseTM for measuring
deflection angles, a stepper motor Nema23, a load cell AL6N-C3-10kg-3B6
from Variohm Euro Sensor, a metal string, and a metal ring for guiding the
metal string. During an experiment, the motor vertically translates the load
cell. As the load cell is connected to the corner of the PV panel via the metal
5.4 stiffness analysis 93
Figure 5.6: (a) Experimental setup for analysing the actuator’s stiffness. (b) Ge-
ometrical model of the setup during stiffness analysis. (c) Raw load
cell measurements during one stiffness analysis. (d) The stiffness
coefficient as the best linear least-squares fit of the torque/angle
samples.
string, it causes the actuator’s deflection, which is measured using the IMU
sensor. Also, the pulling force of the motor is measured by the load cell to
get the quantitative value of the applied force. A small ring is used to keep
94 hybrid metal-elastomeric pneumatic 2-dof actuators
the string perpendicular to the load cell (to match the load cell’s measuring
direction). Before conducting any experiments, the load cell was calibrated.
A coordinate frame and mathematical description of a system are in-
troduced in order to strictly define all vectors (Fig. 5.6b). Two points are
particularly important. The point P1 is the corner of the PV panel to which
the connecting string is attached and defines the displacement from the
rotation axis, while also serving as the force vector origin. The second
important point is P2 , which is the place of the guiding ring. It defines the
direciton of the force acting on P1 . After adopting the orientation of the roll
angle θ as shown in Fig. 5.6b, the coordinates of points P1 and P2 are:
~ = ~p × ~F
M (5.7)
~ ~ ~ ~
~ = P1 × | F |( P2 − P1 )
M (5.8)
~ − P1
| P2 ~ |
The moment M ~ has only a z-axis direction since ~p and ~F are both always
defined only in xy plane. Hence, the angle deflection around the z axis, θ is
directly measured by IMU as roll angle. Now, when all variables are defined,
the actuator’s stiffness coefficient in (in roll direction) can be defined as a
linear dependence between the angle θ and moment M:
M = k·θ (5.9)
5.4.2 Results
The stiffness coefficients for every actuator and every chamber pressure are
estimated from the measured torque-deflection curves. All torque-deflection
curves exhibited close to linear dependency. As one could anticipate from
the above pressure-deflection and transient analyses, the hybrid actuator
with the outer U-joint exhibits the highest stiffness, followed by ones with
the chained internal, the flexible internal, and the basic soft-material-only
actuator (Fig. 5.7). Therefore, all the hybrid actuators achieved higher
absolute stiffness than the purely soft actuator. However, analysing the
relative increase of the stiffness coefficient with respect to chamber pressures
yields another important conclusion (Fig. 5.8).
The stiffness coefficient of the outer structure increases notably less than
any other structure’s for the same change in input pressures. For the pres-
sure of 1.9 bar, the outer structure increases its stiffness by 57%, while the
chained internal structure increases its stiffness by 174%, compared to the
initial stiffness value at 0 bar. This shows a narrow range of possible stiff-
ness values achieved by the outer structure actuator. Therefore, the hybrid
actuator with the chained internal joint can achieve a three times wider
range of possible stiffness values than with the outer joint. Furthermore, the
hybrid actuator with the chained internal U-joint has a tendency to reach
the absolute stiffness value of the one with the outer joint for pressures
higher than 2 bar. Because of this, the hybrid actuator with the chained
internal joint was selected for pressure-to-angle/stiffness modeling and
feedforward control development.
96 hybrid metal-elastomeric pneumatic 2-dof actuators
5.5.1 Methodology
Figure 5.8: (a) Stifness increase and (b) relative stiffness increase with the change
in chamber pressures for soft-material-only and three hybrid soft-
hard-material actuators.
Figure 5.9: Sketch of the experimental setup for stiffness analysis in two axes.
98 hybrid metal-elastomeric pneumatic 2-dof actuators
The position of the corner of the PV panel, point P1, is calculated as:
while the ring position, point P2, is still constant in space. The angles α and
θ are measured directly by the IMU, h represents height of the actuator,
and d is the half-diagonal distance of the PV panel. The values of these
parameters are: h = 8 cm, d = 31.5 cm, x2 = 8.1 cm and y2 = −35.7 cm
z2 = 0 cm. Then, the corresponding bending moment is calculated as:
~ = ~p × ~F
M (5.11)
~ ~ ~ ~
~ = P1 × | F |( P2 − P1 )
M (5.12)
~ − P1
| P2 ~ |
Still, the stiffness coefficient of interest is the one relating the change in roll
angle to the corresponding moment around the z-axis. Therefore, when the
moment M is calculated it would have three components in general case,
and only the z-axis component is used for finding the stiffness coefficient:
Mz = k · θ (5.13)
A black box (data driven) approach is used for modeling the stiffness
coefficient with respect to the individual chamber pressures because of the
complex relation between the actuator shape, pressures, and output angle
values. Based on previous results, we made a general assumption that the
actuator’s stiffness will increase with higher pressures in the chambers.
Therefore, we examined several nonlinear multivariable growing functions
in order to find an adequate relation between the stiffness coefficient and
the chamber pressures. However, any data can be fitted arbitrarily well
with enough degrees of freedom in the fitting function. In order to avoid
overfitting the data, a cross-validation method [128] is used. The obtained
stiffness samples are randomly divided into two sets, a training set and a
testing set. The training set is used to find the function parameters, while
the test set is used for validation of the obtained function. Ten samples
(15.62%) of all 64 were used for testing and the root-mean-square-error
(RMSE) is used as a measure for validation. Then, the cross validation was
5.5 pressures-to-stiffness modeling 99
repeated ten times, and the mean of ten RMSE values was calculated. This
whole procedure was repeated for different multivariate exponential and
polynomial functions in order to find the fitting function F with the lowest
RMSE mean.
Once the function type was set, all data samples were used for the
function parameter estimation. We defined a mathematical optimization
problem in order to estimate the parameters [129]. Besides the convenience
of defining a custom cost function, the optimization approach enables a
simple way of defining constraints. The adopted cost function is defined as
a sum of absolute errors between measured samples and model predicted
values. Additional constraints requered for the first partial derivatives of F
to be positive. These constraints guarantee that a monotonically growing
function with respect to chamber pressures will be obtained. Mathematically
this optimization problem is expressed as:
64
∂F
J = arg max ∑ |YM − F ( p1 , p2 ,3 , C1:n )| such that > 0, ∀i=1:3
C1:n i =1
∂pi
(5.14)
where YM is measured value of the stiffness coefficient, C1:n are n unknown
function coefficients and pi are individual chamber pressures. Matlab’s
Optimization Toolbox is used for solving the optimisation problem.
5.5.2 Results
Figure 5.10: Stiffness coefficient w.r.t. Ch1 and Ch2 pressures when Ch3=0 bar
5.5 pressures-to-stiffness modeling 101
the experimental setup (Fig. 5.9), where chamber 1 receives larger stress than
chambers 2 and 3 under external distrubance. Therefore, the pressure inside
chamber 1 has the highest influence on the stiffness coefficient defined in
this way. If the actuator had been mounted rotated by 180◦ around y-axis,
then chambers 2 and 3 would have had the highest influence. This again
shows the importance of strict stiffness coefficient definition and possible
data misinterpretation if a system is not very well described .
The methodology for the function type selection resulted in a second
order polynomial function:
k = F ( p1 , p2 , p3 ) = C1 p21 + C2 p1 p2 + C3 p1 p3 + C4 p1 +
C5 p22 + C6 p2 p3 + C7 p2 + C8 p23 + (5.15)
C9 p3 + C10
and the optimization provided the following coefficient values: (C1 , ..., C10 ) =
(30.53, 0.02, 0.14, 0.09, 17.08, 5.92, 0.29, 15.84, 0.64, 100). In order to have
a better representation of the stiffness change, this function outputs relative
stiffness values in percentage, relative to the initial (the lowest) stiffness
coefficient at 0 bar, for all chambers. Absolute stiffness coefficient values
can be simply retrieved from relative stiffness when the minimal stiffness
coefficient of 1.38 Nm/rad is known.
Analyzing the individual coefficients, leads to the same conclusion as
one can visually notice. The pressure in Chamber 1 influences the stiffness
more than the pressures in chambers 2 and 3. One should also notice
the symmetry in the function’s coefficients related to the chambers 2 and
3. This symmetry is expected and validates the adopted model because
the positions of chambers 2 and 3 are symmetric to the external load.
Slight differences are probably caused by some minor differences in the
actuator’s body structure (e.g. different rubber thickness, chamber size,
nonsymmetrical chained rod binding, etc.).
The stiffness values predicted by the model (small balls) are shown
together with the measured values (big spheres) Fig. 5.10a). The histogram
of the relative model prediction error shows that more than 70% of data
samples have a relative error of less than 10% (Fig. 5.11).
102 hybrid metal-elastomeric pneumatic 2-dof actuators
5.6.1 Methodology
The starting point for angle deflection modeling is the hexagonal pressure-
deflection curve. This curve is obtained by setting different pressures in
individual chambers, as well as in combinations of two by two chambers.
The basic idea is to find the principal direction of actuation for each cham-
ber [75] and assume a superposition principle when multiple chambers are
actuated. The principal actuation directions ~di are streightforward to extract
from the hexagonal pressure-deflection curves as the direction of a linear
function defined by measurements (Fig. 5.12).
The other part is modeling the intensity of a vector. The intensity is
assumed to be the total deflection in both rotational directions, i.e., the
Euclidean distance from (0,0) to the measured data point. The goal is to
estimate the deflection in a chamber’s principal direction of actuation given
the pressure in that chamber. By doing so, three different functions (Ii ( pi ))
need to be found, one for each chamber. By combining both direction and
intensity, the model can be written ass:
3
[α, θ ]( p1 , p2 , p3 ) = ∑ ~di · Ii ( pi ), where |~di | = 1 ∀i=1:3 (5.16)
i =1
5.6 pressures-to-angles modeling 103
30 0.5 bar
0.8 bar
1.0 bar
20 d1 1.2 bar
1.5 bar
1.9 bar
0
d2 d3
-10
-20
-30 -20 -10 0 10 20 30
Pitch angle [°]
Figure 5.12: Deflection angles with the principal directions of actuation of each
chamber
5.6.2 Results
The obtained principal actuation directions, ~di , and their defining values
are shown in Fig. 5.13a. When the intensity Ii functions are considered,
a quadratic function is adopted to model the deflection intensity of each
chamber for a given pressure (Fig. 5.13b). For the measured data samples,
the quadratic function turned out to have the least RMSE compared to
exponential functions, while higher order polynomial functions tend to
overfit the data. The analytical forms of obtained quadratic function are:
a 30
Measurement
Model output
20
10
Roll angle [°]
10
-20
-30
-30 -20 -10 0 10 20 30
Pitch angle [°]
b 60
55
50
45
Sample ratio [%]
40
35
30
25
20
15
10
5
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Apsolute error [°]
5.7.1 Methodology
A model of a system enables predicting its output signals for a set of given
input signals. Often, it is of interest to know which input signals are needed
to obtain certain desired output values. This problem is known as a feedfor-
ward control problem and it requires solving an inverse problem, i.e. given
the system’s input-to-ouput model, find the output-to-input relation [130].
In the case of the hybrid soft-hard actuator, one should find the three cham-
ber pressures values ( p1 , p2 , p3 ) for a set of given deflection angles (α, θ )
and actuator’s stiffness k.
Since the pressure-to-deflection (5.17) and pressure-to-stiffness (5.15)
models are nonlinear (with the hard limits on the maximum pressure of 2.0
bar), there may or may not be a feasible solution to the inverse modeling
problem. In order to solve this problem, we applied again mathematical
optimization. By doing so, one will always receive some pressure values
which satisfy the requirements in some way. Maybe the most significant
benefit of using an optimisation approach is being independent of adopted
models. Therefore, one can change or adopt system models easily, without
any change in the solving algorithm. Another benefit of optimisation is the
freedom of assigning higher importance to some variable(s) of interest if all
requirements cannot be met.
The adopted cost function has the following form:
where ∆α, ∆θ, and ∆k are differences between the referece values and
predicted values of pitch angle, roll angle and relative stiffness, respectively.
The positive constants C¯1 , C¯2 , C¯3 are weighting factors. For examples, if
it is more important to precisely control the deflection angles, while the
stiffness value should be as close as possible to the reference value, then
the weighting factors could be selected as C¯1 = 100, C¯2 = 100, C¯3 = 1. The
optimisation problem can be formulated as:
[ p1 , p2 , p3 ] =arg min J
p1 , p2 , p3 (5.19)
subject to 0 ≥ Pi ≥ Pmax , Pmax = 2.0 bar, ∀i=1:3
Figure 5.15: Inverse model solution of the hybrid actuator with the chained
internal universal joint. For given reference angles ((a)) and stiffness
values (b), pressure values (c) are obtained.
5.7.2 Results
In the first example (Figure 5.15), a stiffness of about 130% to 240% can be
achieved without disturbing the angle values. Outside of this stiffness range,
one or more chamber pressures reach saturation. It can be noticed that both
angles are tightly kept at the reference values when the chamber pressures
are below the saturation limit. However, the angles slightly vary when the
reference stiffness cannot be achieved. How much they deviate from the
angle setpoints is determined by the coefficients C¯1 , C¯2 , C¯3 . Choosing lower
values for C¯1 and C¯2 would allow a wider angle band, whereas higher
values of these constants guarantee precise angle tracking.
In the second example (Figure 5.16) we show a challenging combination
of angles-stiffness setpoints from the actuator’s point of view. In order to
cause a roll angle of 20◦ , p1 must be very close to the maximal pressure.
This allows only a narrow stiffness range around 250%.
To illustrate these saturation effects caused by the maximum pressure in
the system, we calculated the maximum attainable stiffness for the whole
angle range, based on the above inverse model. The attainable stiffness
range depends on the setpoints of the pitch and roll angles. As the above
examples showed, the further an angle setpoint is from the initial position
(0◦ , 0◦ ), the higher the pressure required to reach that position is, hence
it reduces the availble pressure range up to the maximum pressure for
varying the stiffness. All possibly attainable stiffness values are shown
by the polygonal envelope surface (Fig. 5.17a). Its volume contains all
attainable stiffness values, while the polygonal edges determine the upper
and lower extreme values (Fig. 5.17b,c). The difference between the maximal
and minimal stiffness values creates the cone shape figure of the attainable
stiffness change (Fig. 5.17d). Clearly, its top is around (0◦ , 0◦ ) for both
angles (Fig. 5.17d).
5.8 conclusion
In this article we have discussed the benefits of coupling soft- and hard-
material robotic components and we introduced a novel hybrid soft-hard-
material 2-DOF pneumatically-driven actuator with variable stiffness. It
consist of a single-body three-chambered elastomeric actuator and a metal
U-joint from classical mechanics. The U-joint provides structural stability to
the soft actuator, thus enabling more precise and controlled movements. On
the other hand, all advantages of fluid-driven elastomeric actuators, such as
inherent compliance, high power-to-weight ratio, large stroke, low inertia,
and simple manufacturing are retained. In addition, due to antagonistic
5.8 conclusion 109
Figure 5.16: Inverse model solution of the hybrid actuator with the chained
internal universal joint when one of the chambers has almost reached
the maximum allowed pressure in the system. For given reference
angles ((a)) and stiffness values (b), pressure values (c) are obtained.
Figure 5.17: (a) Attainable stiffness range for different deflection angles. (b) Max-
imal stiffness values for different deflection angles. (c) Minimal
stiffness values for different deflection angles. Region of possible
change in stiffness for different deflection angles in 3D (d) and 2D
(e).
5.8 conclusion 111
scalled up, they could be used as cost-efficient solar trackers for commercial
PV fields. Furthermore, the safety, lightweightness, simple control, and
low cost of these actuators makes them in particularly interesting for con-
structing soft arms and soft legged robots for applications in real-world
environments, such as for human-safe robot interaction or locomotion over
rough terrains. Due to the inherent compliance, the control complexity
of locomotion over rough terrains could be reduced, allowing for more
feedforward-like control. Finally, beside the low mechanical impendance
and inherent compliance, the pressure-based feedback control allows for
detection of external impact and depressurising the actuator immediately,
thus further reducing the risk of human injury.
The system and experiment setup consist of numerous elements (Fig. X). We
present the details of it in Fig. NI CompactRIO real-time controller(1) is the
main controller of the system. It runs the controlling algorithm and governs
the system states. NI CompactRIO communicates with other controlling
devices through LAN infrastructure using Ethernet router(3). First, it ex-
changes information with PC(2), which provides user interface for system
monitoring and control. Next, NI CompactRIO communicates with two
Raspberry PI devices(4)(15). One Raspberry PI receives IMU(6) measure-
ments, process raw data and returns IMU angles to the NI CompactRIO.
The other Raspberry PI, receives measurements from load cell(14), process
them and returns force values to main controller. The second Raspberry
PI(15) sends motor movement references to Arduino Uno(18), which runs
step motor(17) control algorithm. Step motor shifts load cell vertically using
mechanical belt(13) and causes different load values. Metal string(8) con-
nects PV panel(5) to the load cell. The metal string goes through guiding
ring(9) in order to have perpendicular forces to the load cell. NI Com-
pactRIO directly controls individual chamber valves(10) as well as pressure
IN/OUT valves (12). All valves have purpose to open/close pneumatic
lines between hybrid soft-hard actuator(7) and air compressor(16) which
provides necessary high pressure in the system. In order to measure pres-
sure in chambers, a pressure sensor is used and connected to the main NI
CompactRIO controller.
5.10 appendix b. fabrication of 2-dof joints 113
Figure 5.18: (1) NI CompactRIO real-time controller, (2) Personal computer, (3)
Ethernet router, (4) Raspberry PI for obtaining angle measuremetns
from the IMU sensor and forwarding it to the NI CompactRIO con-
troller, (5) Photovoltaic panel, (6) Inertial measurement unit (IMU),
(7) Hybrid soft-hard actuator, (8) Metal connecting string, (9) Guid-
ing ring, (10) Valves for individual chambers, (11) Pressure sensor,
(12) Pressure IN/OUT valve, (13) Mechanical belt, (14) Load cell,
(15) Raspberry PI for controlling the step motor, (16) Air compressor,
(17) Step motor, and (18) Arduino Uno
From the three types of mechanical joints that give the hybrid actuator
its rigidity, only one can be found as an off-the-shelf component, namely
the flexible shaft. It is composed of multiple layers of spiralled spring
wire, coiled up in opposite directions on adjacent layers. This construction
ensures that the shaft does not expand or collapse in itself radially, thus
allowing the transmission of torsion moments in both directions. Axially,
114 hybrid metal-elastomeric pneumatic 2-dof actuators
the shaft can present a slight expansion under tension forces, which fits
better to the expansion behaviour of the basic soft-material-only actuator.
The ICU joint is a modified version of a universal joint, where the bearing
cross is reduced to only one axis of rotation, allowing the design of slender
shafts with high torque transmission capabilities. To achieve a similar
behaviour of a conventional universal joint, our design uses three links and
two rotation axes oriented perpendicular. In order to achieve a mechanical
motion that matches closer the continuous bending of the soft actuator, we
added two more links, resulting in total of four rotational axes oriented
alternatively at 0◦ and 90◦ . The forces are not transmitted only through
the rotation pin, but also through the direct interlocking of the chain links,
giving the very thin ICU joint a superior torque transmission compared to
standard universal joint with a bearing cross of that size. The ICU was only
constructed as a prototype, conventionally milled from a hexagonal SW8
profile and assembled with steel rivets.
The OU joint is conceptually very similar to the ICU, but the structural
material is placed on the outside, around an empty core - where we place
the actuator. The two rotational axes, oriented at 0◦ and 90◦ , are placed
at the extremes of the joint in order to avoid collision with the inflating
(and bending) actuator. For this, the link between those axes (what would
be the bearing cross of a universal joint) has a very specific wave shape
that maximizes the range of motion. Although this construction seems
rather complex, the availability of laser sheet metal cutting and 3D CNC
laser cutting enabled an easy, fast and cost-efficient production of these
OU joints. The upper and lower components were laser cut from sheets
1.5mm stainless steel and the two wings for the rotation pin were bent 90◦ .
The middle link was designed to be cut on a 3D CNC laser directly from
a piece of stainless steel pipe. This method was preferred over the regular
2D laser cutting and bending, as it is two to three times more cost efficient
due to the one-step production and the possibility to stack multiple links
on a single pipe, thus minimizing material waste. Finally, the pieces were
assembled manually with steel rivets.
CONCLUSION
6
6.1 summary of results
stability to the soft actuator, thus enabling its more precise and controlled
movements, higher forces, and variable stiffness, while all the advantages
of pneumatically-driven elastomeric actuators are retained. The chained
internal structure leads to the most versatile actuator that can vary stiffness
the most. Its maximal deflection angle was only 3◦ less than the initial
soft actuator that had no hard-material components. Moreover, the chained
internal structure achieves the widest range of values for the damping
coefficient, expresses the highest increase in stiffness values of around three
times (180% increase) at 1.9 bar. On the other hand, the outer structure
has the highest absolute stiffness over the whole chamber pressure range,
meaning the highest structural stability for the whole range of input signals.
For wind stabilisation purposes in real weather, the outer structure should
provide the best results.
These lightweight, compliant actuators enabled reduction of the actuator
size and construction of lightweight adaptive building envelopes with inte-
grated thin-film photovoltaic panels. In total three building-scale prototypes
were constructed. The first envelope prototype, with 50 panels actuated by
cylindrical actuator fabricated from Elastosil™Vario, was constructed in
2015 at the House of Natural Resources at the ETH Zurich Hoenggerberg
Campus. The second prototype, with 16 panels actuated by the corrugated
soft actuator industrially fabricated from Neoprene rubber, was constructed
in 2016 at the roof of HPZ building at the same campus. The final proto-
type, with 30 panels actuated by hybrid actuators with outer joints, was
constructed in 2017 at the NEST building at the Swiss Federal Laboratories
for Materials Science and Technology (Empa).
I then presented a novel distributed modular pneumatic control system.
It allows the pressure at each envelope module to be controlled locally
by receiving the control signals from the central controlling unit, while
the measurements from the orientation sensor (inertial measurement unit)
are sent to the central unit. Such a pneumatic-orientation control system
enables expansion of the facade to a wide range of shapes and sizes, with
only a few steps required to connect a new envelope module with the soft
actuator.
Finally, the performance of HPZ envelope was measured in real-weather
conditions in solar tracking experiments. Electricity gains of 30-50% have
been reported. The gain in solar tracking depends largely on envelope
orientation, climatic region, and the distribution of the elements within the
envelope. However, the total building energy savings also include the effects
of active shading. Estimated energy savings for an office in Zurich with
6.2 discussion 117
6.2 discussion
lack of transparency. This has been improved in the design for the NEST
building by having a large spread between the panels.
Finally, a decisive criterion for a potential buyer is likely to be the overall
benefit it creates for the user. The system can provide a unique combination
of energy saving, autonomous operation for maximizing solar energy har-
vesting during the day, storage of excess energy in a battery, and provision
of adaptive personalised comfort.
The promising next steps are testing the latest envelope prototype within
a real-building setting, in front of an office window. This will enable testing
of control strategies for maximising the energy gains and testing of vari-
ous occupant-envelope interaction means and data-driven learning-based
control strategies. Further, architects can also program open-loop demo
modes of the envelope, to create particular building expressions, not only
by quickly transitioning between static states, but also creating dynamic
expression.
6.3 outlook
1. Lucon, O., Ürge-Vorsatz, D., Ahmed, A., Akbari, H., Bertoldi, P.,
Cabeza, L., et al. Buildings. Working Group III contribution to the IPCC
5th assessment report, climate change 2014: mitigation of climate change
00002 (Cambridge, UK/New York, NY, USA: Cambridge University
Press, 2014).
2. International energy agency (iea), energy technology perspectives, tracking
clean energy progress 2017, building envelopes 00000.
3. Loonen, R., Trčka, M., Cóstola, D. & Hensen, J. Climate adaptive
building shells: State-of-the-art and future challenges. Renewable and
Sustainable Energy Reviews 25. 00135, 483 (2013).
4. Linn, C. Kinetic architecture: design for active envelopes 00026 (Images
Publishing, 2014).
5. Laschi, C., Mazzolai, B. & Cianchetti, M. Soft robotics: Technologies
and systems pushing the boundaries of robot abilities. Sci. Robot. 1,
eaah3690 (2016).
6. Wehner, M., Truby, R. L., Fitzgerald, D. J., Mosadegh, B., Whitesides,
G. M., Lewis, J. A. & Wood, R. J. An integrated design and fabrication
strategy for entirely soft, autonomous robots. Nature 536, 451 (2016).
7. Brown, E., Rodenberg, N., Amend, J., Mozeika, A., Steltz, E., Zakin,
M. R., Lipson, H. & Jaeger, H. M. Universal robotic gripper based on
the jamming of granular material. Proceedings of the National Academy
of Sciences 107, 18809 (2010).
8. Laschi, C., Cianchetti, M., Mazzolai, B., Margheri, L., Follador, M. &
Dario, P. Soft robot arm inspired by the octopus. Adv. Robotics 26, 709
(2012).
9. Morin, S. A., Shepherd, R. F., Kwok, S. W., Stokes, A. A., Nemiroski,
A. & Whitesides, G. M. Camouflage and display for soft machines.
Science 337, 828 (2012).
10. Menon, C. & Sitti, M. A biomimetic climbing robot based on the
gecko. Journal of Bionic Engineering 3, 115 (2006).
121
122 bibliography
11. Gorissen, B., Reynaerts, D., Konishi, S., Yoshida, K., Kim, J.-W. & De
Volder, M. Elastic Inflatable Actuators for Soft Robotic Applications.
Advanced Materials 29. 00000, n/a (2017).
12. Cianchetti, M., Ranzani, T., Gerboni, G., Falco, I. D., Laschi, C. &
Menciassi, A. STIFF-FLOP surgical manipulator: Mechanical design and
experimental characterization of the single module in 2013 IEEE/RSJ In-
ternational Conference on Intelligent Robots and Systems 00073 (2013),
3576.
13. Manti, M., Pratesi, A., Falotico, E., Cianchetti, M. & Laschi, C. Soft
assistive robot for personal care of elderly people in Biomedical Robotics
and Biomechatronics (BioRob), 2016 6th IEEE International Conference on
(2016), 833.
14. Pfeifer, R., Lungarella, M. & Iida, F. Self-organization, embodiment,
and biologically inspired robotics. science 318. 00647, 1088 (2007).
15. Rus, D. & Tolley, M. T. Design, fabrication and control of soft robots.
Nature 521. 00362, 467 (2015).
16. Manti, M., Cacucciolo, V. & Cianchetti, M. Stiffening in Soft Robotics:
A Review of the State of the Art. IEEE Robotics Automation Magazine
23. 00001, 93 (2016).
17. Mosadegh, B., Polygerinos, P., Keplinger, C., Wennstedt, S., Shepherd,
R. F., Gupta, U., Shim, J., Bertoldi, K., Walsh, C. J. & Whitesides, G. M.
Pneumatic Networks for Soft Robotics that Actuate Rapidly. Advanced
Functional Materials 24. 00191, 2163 (2014).
18. Martinez, R. V., Fish, C. R., Chen, X. & Whitesides, G. M. Elastomeric
origami: programmable paper-elastomer composites as pneumatic
actuators. Advanced functional materials 22, 1376 (2012).
19. Mousazadeh, H., Keyhani, A., Javadi, A., Mobli, H., Abrinia, K. &
Sharifi, A. A review of principle and sun-tracking methods for maxi-
mizing solar systems output. Renewable and sustainable energy reviews
13, 1800 (2009).
20. Block, P., Schlueter, A., Veenendaal, D., Bakker, J., Begle, M., Hofer,
J., Jayathissa, P., Lydon, G., Maxwell, I., Mendez Echenagucia, T.,
Nagy, Z., Pigram, D., Svetozarevic, B., Torsing, R., Verbeek, J. &
Willmann, A. NEST HiLo: Research & innovation unit for lightweight
construction and building systems integration. Journal of Building
Engineering. under review (2017).
bibliography 123
21. Lamoureux, A., Lee, K., Shlian, M., Forrest, S. R. & Shtein, M. Dy-
namic kirigami structures for integrated solar tracking. Nature com-
munications 6 (2015).
22. Suzumori, K., Iikura, S. & Tanaka, H. Applying a flexible microactua-
tor to robotic mechanisms. Control Systems, IEEE 12, 21 (1992).
23. Ilievski, F., Mazzeo, A. D., Shepherd, R. F., Chen, X. & Whitesides,
G. M. Soft robotics for chemists. Angewandte Chemie 123, 1930 (2011).
24. Deimel, R. & Brock, O. A novel type of compliant, underactuated
robotic hand for dexterous grasping. Robotics: Science and Systems,
Berkeley, CA, 1687 (2014).
25. McMahan, W., Chitrakaran, V., Csencsits, M., Dawson, D., Walker,
I. D., Jones, B. A., Pritts, M., Dienno, D., Grissom, M. & Rahn, C. D.
Field trials and testing of the OctArm continuum manipulator in Proc.
IEEE Int’l Conf. on Robotics and Automation (ICRA) (2006).
26. Sanan, S., Lynn, P. S. & Griffith, S. T. Pneumatic torsional actuators
for inflatable robots. J. of Mechanisms and Robotics 6, 031003 (2014).
27. Marchese, A. D., Komorowski, K., Onal, C. D. & Rus, D. Design and
control of a soft and continuously deformable 2d robotic manipulation system
in Proc. IEEE Int’l Conf. on Robotics and Automation (ICRA) (2014).
28. Maghooa, F., Stilli, A., Noh, Y., Althoefer, K. & Wurdemann, H. A.
Tendon and pressure actuation for a bio-inspired manipulator based on an
antagonistic principle in Proc. IEEE Int’l Conf. on Robotics and Automation
(ICRA) (2015).
29. Wang, L. & Iida, F. Deformation in Soft-Matter Robotics: A Cate-
gorization and Quantitative Characterization. Robotics Automation
Magazine, IEEE 22, 125 (2015).
30. Asbeck, A. T., Dyer, R. J., Larusson, A. F. & Walsh, C. J. Biologically-
inspired soft exosuit in Int’l Conf. on Rehabilitation robotics (ICORR)
(2013).
31. Polygerinos, P., Galloway, K. C., Savage, E., Herman, M., Donnell,
K. O. & Walsh, C. J. Soft robotic glove for hand rehabilitation and task
specific training in Proc. IEEE Int’l Conf. on Robotics and Automation
(ICRA) (2015).
32. Rus, D. & Tolley, M. T. Design, fabrication and control of soft robots.
Nature 521, 467 (2015).
124 bibliography
33. Tolley, M. T., Shepherd, R. F., Mosadegh, B., Galloway, K. C., Wehner,
M., Karpelson, M., Wood, R. J. & Whitesides, G. M. A resilient,
untethered soft robot. Soft Robotics 1, 213 (2014).
34. Shepherd, R. F., Stokes, A. A., Nunes, R. & Whitesides, G. M. Soft
machines that are resistant to puncture and that self seal. Advanced
Materials 25, 6709 (2013).
35. Terryn, S., Mathijssen, G., Brancart, J., Van Assche, G., Vanderborght,
B. & Lefeber, D. Investigation of self-healing compliant actuators for
robotics in Proc. IEEE Int’l Conf. on Robotics and Automation (ICRA)
(2015).
36. Pfeifer, R., Iida, F. & Lungarella, M. Cognition from the bottom up: on
biological inspiration, body morphology, and soft materials. Trends in
cognitive sciences 18, 404 (2014).
37. Rossi, D., Nagy, Z. & Schlueter, A. Adaptive distributed robotics
for environmental performance, occupant comfort and architectural
expression. Int’l J Arch Comp 10, 341 (2012).
38. Jayathissa, P., Nagy, Z., Offedu, N. & Schlueter, A. Numerical Simula-
tion of Energy Performance, and Construction of the Adaptive Solar Façade
in Advanced Building Skins (2015).
39. Jakubiec, J. A. & Reinhart, C. F. DIVA 2.0: Integrating daylight and
thermal simulations using Rhinoceros 3D, Daysim and EnergyPlus in
Building Simulation-12th Conf. of Int’l Building Performance Simulation
Association (2011).
40. Häberlin, H. Photovoltaics system design and practice (John Wiley &
Sons, 2012).
41. Kammen, D. M. & Sunter, D. A. City-integrated renewable energy for
urban sustainability. Science 352. 00017, 922 (2016).
42. Mathiesen, B. V., Lund, H. & Karlsson, K. 100% Renewable energy
systems, climate mitigation and economic growth. Applied Energy 88.
00363, 488 (2011).
43. Rogelj, J., Luderer, G., Pietzcker, R. C., Kriegler, E., Schaeffer, M.,
Krey, V. & Riahi, K. Energy system transformations for limiting end-
of-century warming to below 1.5 °C. Nature Climate Change 5. 00115,
519 (2015).
44. Recast, E. Directive 2010/31/EU of the European Parliament and of
the Council of 19 May 2010 on the energy performance of buildings
(recast). Official Journal of the European Union 18. 00122, 2010 (2010).
bibliography 125
45. Trivedi, D., Rahn, C. D., Kier, W. M. & Walker, I. D. Soft robotics:
Biological inspiration, state of the art, and future research. Applied
Bionics and Biomechanics 5. 00643, 99 (2008).
46. Majidi, C. Soft Robotics: A Perspective—Current Trends and Prospects
for the Future. Soft Robotics 1. 00207, 5 (2014).
47. Wang, L., Nurzaman, S. G., Iida, F., et al. Soft-Material Robotics.
Foundations and Trends® in Robotics 5. 00001, 191 (2017).
48. Tolley, M. T., Shepherd, R. F., Mosadegh, B., Galloway, K. C., Wehner,
M., Karpelson, M., Wood, R. J. & Whitesides, G. M. A Resilient,
Untethered Soft Robot. Soft Robotics 1. 00153, 213 (2014).
49. Trimmer, B. A Practical Approach to Soft Actuation. Soft Robotics 4.
00000, 1 (2017).
50. Deimel, R. & Brock, O. A novel type of compliant and underactu-
ated robotic hand for dexterous grasping. The International Journal of
Robotics Research 35. 00163, 161 (2016).
51. Asbeck, A. T., Dyer, R. J., Larusson, A. F. & Walsh, C. J. Biologically-
inspired soft exosuit in 2013 IEEE 13th International Conference on Reha-
bilitation Robotics (ICORR) 00068 (2013), 1.
52. Cianchetti, M., Ranzani, T., Gerboni, G., Nanayakkara, T., Althoe-
fer, K., Dasgupta, P. & Menciassi, A. Soft Robotics Technologies to
Address Shortcomings in Today’s Minimally Invasive Surgery: The
STIFF-FLOP Approach. Soft Robotics 1. 00056, 122 (2014).
53. Nagy, Z., Svetozarevic, B., Jayathissa, P., Begle, M., Hofer, J., Lydon,
G., Willmann, A. & Schlueter, A. The Adaptive Solar Facade: From
concept to prototypes. Frontiers of Architectural Research 5. 00011, 143
(2016).
54. Jayathissa, P., Luzzatto, M., Schmidli, J., Hofer, J., Nagy, Z. & Schlueter,
A. Optimising building net energy demand with dynamic BIPV
shading. Applied Energy 202. 00002, 726 (2017).
55. Kim, J. T. & Kim, G. Overview and new developments in optical day-
lighting systems for building a healthy indoor environment. Building
and Environment 45. 00101, 256 (2010).
56. Chow, T. A review on photovoltaic/thermal hybrid solar technology.
Applied Energy 87. 00751, 365 (2010).
57. Chemisana, D. Building Integrated Concentrating Photovoltaics: A
review. Renewable and Sustainable Energy Reviews 15. 00164, 603 (2011).
126 bibliography
68. Lydon, G., Hofer, J., Svetozarevic, B., Nagy, Z. & Schlueter, A. Cou-
pling energy systems with lightweight structures for a net plus energy
building. Applied Energy 189, 310 (2017).
69. Nest – Exploring the Future of Buildings, Swiss Federal Laboratories for
Materials Science and Technology - EMPA, Duebendorf, Switzerland 00000.
70. Zhao, H., Li, Y., Elsamadisi, A. & Shepherd, R. Scalable manufacturing
of high force wearable soft actuators. Extreme Mechanics Letters 3.
00021, 89 (2015).
71. Galloway, K. C., Becker, K. P., Phillips, B., Kirby, J., Licht, S., Tchernov,
D., Wood, R. J. & Gruber, D. F. Soft Robotic Grippers for Biological
Sampling on Deep Reefs. Soft Robotics 3. 00048, 23 (2016).
72. Moseley, P., Florez, J. M., Sonar, H. A., Agarwal, G., Curtin, W. & Paik,
J. Modeling, Design, and Development of Soft Pneumatic Actuators
with Finite Element Method: Modeling, Design, and Development
of SPAs with FEM . . . Advanced Engineering Materials 18. 00000, 978
(2016).
73. Jayathissa, P., Caranovic, S., Begle, M., Svetozarevic, B., Hofer, J., Nagy,
Z. & Arno, S. Structural and Architectural Integration of Adaptive
Photovoltaic Modules. Proceedings of the Advanced Building Skins. 00000,
C6 (2016).
74. Madgwick, S. O. H., Harrison, A. J. L. & Vaidyanathan, R. Estimation
of IMU and MARG orientation using a gradient descent algorithm in 2011
IEEE International Conference on Rehabilitation Robotics 00654 (2011), 1.
75. Svetozarevic, B., Nagy, Z., Hofer, J., Jacob, D., Begle, M., Chatzi, E. &
Schlueter, A. SoRo-Track: a two-axis soft robotic platform for solar tracking
and building-integrated photovoltaic applications in IEEE International
Conference on Robotics and Automation (ICRA) (2016).
76. Mayne, D. Q., Rawlings, J. B., Rao, C. V. & Scokaert, P. O. M. Con-
strained model predictive control: Stability and optimality. Automatica
36. 06070, 789 (2000).
77. Svetozarevic, B. Zig-Zag orientation feedback control with the upper angle
bound of 2-DOFs soft pneumatic actuator 00000. 2017.
78. Joint Research Centre, Institute for Energy and Transport (IET), Photovoltaic
Geographical Information System (PVGIS) 00000.
79. Compressed Air Evaluation Protocol, National Renewable Energy Labora-
tory 00000.
128 bibliography
91. Rodrigue, H., Bhandari, B., Wang, W. & Ahn, S.-H. 3D soft lithog-
raphy: A fabrication process for thermocurable polymers. Journal of
Materials Processing Technology 217, 302 (2015).
92. Zolfagharian, A., Kouzani, A. Z., Khoo, S. Y., Moghadam, A. A. A.,
Gibson, I. & Kaynak, A. Evolution of 3D printed soft actuators. Sensors
and Actuators A: Physical 250, 258 (2016).
93. Siciliano, B. & Khatib, O. Springer Handbook of Robotics (Springer,
2016).
94. Kim, S., Laschi, C. & Trimmer, B. Soft robotics: a bioinspired evolution
in robotics. Trends in Biotechnology 31, 287 (2013).
95. Kardong, K. V. Vertebrates: comparative anatomy, function, evolution
QL805 K35 2006 (McGraw-Hill New York, 2002).
96. Dickinson, M. H., Farley, C. T., Full, R. J., Koehl, M. A. R., Kram, R.
& Lehman, S. How animals move: an integrative view. Science 288.
01079, 100 (2000).
97. Mengüç, Y., Correll, N., Kramer, R. & Paik, J. Will robots be bodies
with brains or brains with bodies? Science Robotics 2. 00000, eaar4527
(2017).
98. Zinn, M., Khatib, O., Roth, B. & Salisbury, J. K. Playing it safe [human-
friendly robots]. IEEE Robotics & Automation Magazine 11. 00325, 12
(2004).
99. Wolf, S., Grioli, G., Eiberger, O., Friedl, W., Grebenstein, M., Hoppner,
H., Burdet, E., Caldwell, D. G., Carloni, R., Catalano, M. G., Lefeber,
D., Stramigioli, S., Tsagarakis, N., Van Damme, M., Van Ham, R.,
Vanderborght, B., Visser, L. C., Bicchi, A. & Albu-Schaffer, A. Variable
Stiffness Actuators: Review on Design and Components. IEEE/ASME
Transactions on Mechatronics 21. 00030, 2418 (2016).
100. Boston Dynamics is changing your idea of what robots can do. | Boston
Dynamics 00000.
101. Seok, S., Wang, A., Michael Chuah, M. Y., Hyun, D. J., Lee, J., Otten,
D. M., Lang, J. H. & Kim, S. Design Principles for Energy-Efficient
Legged Locomotion and Implementation on the MIT Cheetah Robot.
IEEE/ASME Transactions on Mechatronics 20, 1117 (2015).
130 bibliography
102. Hutter, M., Gehring, C., Jud, D., Lauber, A., Bellicoso, C. D., Tsounis,
V., Hwangbo, J., Bodie, K., Fankhauser, P., Bloesch, M., Diethelm, R.,
Bachmann, S., Melzer, A. & Hoepflinger, M. ANYmal - a highly mobile
and dynamic quadrupedal robot in 2016 IEEE/RSJ International Conference
on Intelligent Robots and Systems (IROS) (2016), 38.
103. Grebenstein, M., Albu-Schäffer, A., Bahls, T., Chalon, M., Eiberger, O.,
Friedl, W., Gruber, R., Haddadin, S., Hagn, U., Haslinger, R., et al. The
DLR hand arm system in Robotics and Automation (ICRA), 2011 IEEE
International Conference on (2011), 3175.
104. Guizzo, E. & Ackerman, E. The hard lessons of DARPA’s robotics
challenge [News]. IEEE Spectrum 52. 00019, 11 (2015).
105. Wall, V., Deimel, R. & Brock, O. Selective stiffening of soft actuators
based on jamming in 2015 IEEE International Conference on Robotics and
Automation (ICRA) (2015), 252.
106. Cianchetti, M., Licofonte, A., Follador, M., Rogai, F. & Laschi, C.
Bioinspired Soft Actuation System Using Shape Memory Alloys.
Actuators 3, 226 (2014).
107. Yang, Y., Chen, Y., Li, Y., Chen, M. Z. & Wei, Y. Bioinspired Robotic
Fingers Based on Pneumatic Actuator and 3D Printing of Smart
Material. Soft Robotics 4. 00003, 147 (2017).
108. Carpi, F., Frediani, G., Gerboni, C., Gemignani, J. & De Rossi, D.
Enabling variable-stiffness hand rehabilitation orthoses with dielectric
elastomer transducers. eng. Medical Engineering & Physics 36, 205
(2014).
109. Maghooa, F., Stilli, A., Noh, Y., Althoefer, K. & Wurdemann, H. A.
Tendon and pressure actuation for a bio-inspired manipulator based on an
antagonistic principle in 2015 IEEE International Conference on Robotics
and Automation (ICRA) (2015), 2556.
110. Immega, G. & Antonelli, K. The KSI tentacle manipulator in Proceedings
of 1995 IEEE International Conference on Robotics and Automation 3
(1995), 3149.
111. Correll, N., Önal, Ç. D., Liang, H., Schoenfeld, E. & Rus, D. Soft
autonomous materials—using active elasticity and embedded distributed
computation in Experimental Robotics (2014), 227.
bibliography 131
112. Suzumori, K., Wakimoto, S., Miyoshi, K. & Iwata, K. Long bending
rubber mechanism combined contracting and extending tluidic actuators in
2013 IEEE/RSJ International Conference on Intelligent Robots and Systems
(2013), 4454.
113. Daerden, F. & Lefeber, D. Pneumatic Artificial Muscles: actuators for
robotics and automation. European journal of mechanical and environ-
mental engineering 47, 11 (2002).
114. Tondu, B., Ippolito, S., Guiochet, J. & Daidie, A. A Seven-degrees-
of-freedom Robot-arm Driven by Pneumatic Artificial Muscles for
Humanoid Robots. The International Journal of Robotics Research 24.
00206, 257 (2005).
115. Vanderborght, B., Van Ham, R., Verrelst, B., Van Damme, M. &
Lefeber, D. Overview of the Lucy Project: Dynamic Stabilization of a
Biped Powered by Pneumatic Artificial Muscles. Advanced Robotics 22.
00064, 1027 (2008).
116. Felt, W., Chin, K. Y. & Remy, C. D. Smart Braid Feedback for the
Closed-Loop Control of Soft Robotic Systems. Soft Robotics 4, 261
(2017).
117. Yi, J., Chen, X., Song, C. & Wang, Z. Fiber-Reinforced Origamic
Robotic Actuator. Soft Robotics. 00000 (2017).
118. Ohta, P., Valle, L., King, J., Low, K., Yi, J., Atkeson, C. G. & Park, Y.-L.
Design of a Lightweight Soft Robotic Arm Using Pneumatic Artificial
Muscles and Inflatable Sleeves. Soft Robotics. 00001 (2017).
119. Hawkes, E. W., Christensen, D. L. & Okamura, A. M. Design and imple-
mentation of a 300% strain soft artificial muscle in 2016 IEEE International
Conference on Robotics and Automation (ICRA) (2016), 4022.
120. Han, K., Kim, N.-H. & Shin, D. A Novel Soft Pneumatic Artificial
Muscle with High-Contraction Ratio. Soft Robotics (2018).
121. Skorina, E. H., Luo, M., Ozel, S., Chen, F., Tao, W. & Onal, C. D.
Feedforward augmented sliding mode motion control of antagonistic soft
pneumatic actuators in 2015 IEEE International Conference on Robotics
and Automation (ICRA) (2015), 2544.
122. Skorina, E. H., Tao, W., Chen, F., Luo, M. & Onal, C. D. Motion
control of a soft-actuated modular manipulator in 2016 IEEE International
Conference on Robotics and Automation (ICRA) (2016), 4997.
132 bibliography
123. Svetozarevic, B., Begle, M., Caranovic, S., Nagy, Z. & Schlueter, A.
Quick-cast: A method for fast and precise mass production of fluid-
driven elastomeric soft actuators. Extreme Mechanics Letters (under
review) (2018).
124. Franklin, G. F., Powell, J. D., Emami-Naeini, A. & Powell, J. D. Feedback
control of dynamic systems (Addison-Wesley Reading, MA, 1994).
125. Inman, D. J. & Singh, R. C. Engineering vibration (Prentice Hall Engle-
wood Cliffs, NJ, 1994).
126. Baumgart, F. Stiffness-an unknown world of mechanical science?
Injury-International Journal for the Care of the Injured 31, 14 (2000).
127. Chawla, K. K. & Meyers, M. Mechanical behavior of materials (Prentice
Hall, 1999).
128. Bishop, C. M. Pattern recognition and machine learning (springer, 2006).
129. Lange, K. Optimization (Springer-Verlag New York, 2013).
130. Haugen, F. Basic Dynamics and Control (2009).
C U R R I C U L U M V I TA E
personal data
education
experience
133
P U B L I C AT I O N S
Conference contributions:
4. Svetozarevic, B., Hofer, J., Hischier, I. & Schlüter, A. Flexible Pneumatic Actuator for PV
Solar Tracking Applications in Proceedings of the EU PVSEC (2017).
5. Svetozarevic, B., Nagy, Z., Hofer, J., Jacob, D., Begle, M., Chatzi, E. & Schlueter, A. SoRo-
Track: a two-axis soft robotic platform for solar tracking and building-integrated photovoltaic
applications in IEEE International Conference on Robotics and Automation (ICRA) (2016).
6. Svetozarevic, B., Nagy, Z., Rossi, D. & Schlueter, A. Experimental Characterization of a
2-DOF Soft Robotic Platform for Architectural Applications in Robotics: Science and Systems,
Workshop on Advances on Soft Robotics (RSS), Berkeley, CA, USA (2014).
Patents:
7. Svetozarevic, B., Schlueter, A., Begle, M., Jayathissa, P. & Caranovic, S. Hybrid hard-
soft-material pneumatic actuator with adjustable mechanical impedance European patent,
EP17201677.6. (2017).
? Awarded: Spark Award 2018 for Top 20 patents at ETH Zurich (out of 86 filed in 2017).
Technical reports:
8. Svetozarevic, B., Lydon, G., Nagy, Z. & Schlueter, A. HiLo Home Automation Research
Platform Architecture and Building Systems, ETH Zurich. (2016).
135
C O N T R I B U T E D T O P U B L I C AT I O N S
Conference contributions:
5. Hofer, J., Svetozarevic, B., Nagy, Z. & Schlüter, A. DC building networks and local
storage for BIPV integration in Proceedings of International Conference CISBAT 2015 Future
Buildings and Districts Sustainability from Nano to Urban Scale (2015), 681.
6. Jayathissa, P., Caranovic, S., Begle, M., Svetozarevic, B., Hofer, J., Nagy, Z. & Arno, S.
Structural and Architectural Integration of Adaptive Photovoltaic Modules in Proceedings of
the Advanced Building Skins (Advanced Building Skins GmbH, 2016), C6.
7. Hofer, J., Svetozarevic, B. & Schlueter, A. Hybrid AC/DC building microgrid for solar PV
and battery storage integration in DC Microgrids (ICDCM), 2017 IEEE Second International
Conference on (2017), 188.
137