Robotic prefabrication of timber structures: towards automated large-scale spatial assembly

Abstract

Despite modern timber construction being on the forefront of digital technology in construction, subtractive CNC—fabrication technologies are still predominantly used in the industry. An important break in the digital chain occurs when prefabricated small building parts have to be assembled manually into functional modules. This can result in a loss of digital information in the process. Therefore, a robotic setup for timber construction was specifically developed by the authors enabling large-scale spatial fabrication possibilities using a combination of subtractive external tools for cutting and drilling and additive robotic operations. Through automatization techniques and innovative feedback processes, the system can minimize material waste by reacting to different material sizes even during the construction process. In a case study, which was undertaken in the course of the Master of Advanced Studies program in Digital Fabrication at ETH Zurich, a complete digital workflow using additive robotic fabrication processes in timber construction was realized. We demonstrate the conception of the worldwide first double-story robotically assembled timber structure, explain its fabrication processes including an integrated envelope, and conclude by analyzing the robotic fabrication technologies in terms of their efficiency and structural and functional capabilities and limits.

Full text

ORIGINAL PAPER
Robotic prefabrication of timber structures: towards automated
large-scale spatial assembly
Philipp Eversmann
1
•Fabio Gramazio
2
•Matthias Kohler
2
Received: 30 December 2016 / Accepted: 31 July 2017 / Published online: 14 August 2017
ÓSpringer International Publishing AG 2017
Abstract Despite modern timber construction being on the
forefront of digital technology in construction, subtractive
CNC—fabrication technologies are still predominantly
used in the industry. An important break in the digital chain
occurs when prefabricated small building parts have to be
assembled manually into functional modules. This can
result in a loss of digital information in the process.
Therefore, a robotic setup for timber construction was
specifically developed by the authors enabling large-scale
spatial fabrication possibilities using a combination of
subtractive external tools for cutting and drilling and
additive robotic operations. Through automatization tech-
niques and innovative feedback processes, the system can
minimize material waste by reacting to different material
sizes even during the construction process. In a case study,
which was undertaken in the course of the Master of
Advanced Studies program in Digital Fabrication at ETH
Zurich, a complete digital workflow using additive robotic
fabrication processes in timber construction was realized.
We demonstrate the conception of the worldwide first
double-story robotically assembled timber structure,
explain its fabrication processes including an integrated
envelope, and conclude by analyzing the robotic fabrica-
tion technologies in terms of their efficiency and structural
and functional capabilities and limits.
Keywords Robotic fabrication Digital design Timber
construction Prefabrication Additive fabrication 
Feedback process
1 Introduction
1.1 Prefabrication
Modern timber construction is already highly integrated
and digitally mastered (Internationale Konferenz 2006;
Jeska and Pascha 2015). An automated prefabrication of
singular timber elements has been demonstrated on various
research studies and projects in the last decades (Sass 2007;
Beyer 1991; Scheurer 2011). The degree of prefabrication
describes the size and complexity of prefabricated com-
ponents, which is directly related to the amount of on-site
construction labor, material use, construction quality, and,
therefore, sustainability performance (Boafo et al. 2016).
Even today, CNC technologies are used predominantly for
subtractive processes in the industry (Popovic et al. 2016),
which can also be noted in the available manufacturing
techniques of current timber production lines (Hans Hun-
degger 2016). These machines are used to precisely man-
ufacture small components like beams and plates, which
are later assembled manually into larger components. A
break of the digital chain occurs exactly before spatial and
functional building parts are assembled. This can result in a
loss of information and precision in the digital process, but
also in unexplored spatial, constructive, and fabrication
potentials. Through robotic prefabrication, an extremely
&Philipp Eversmann
studio@eversmann.fr
Fabio Gramazio
gramazio@arch.ethz.ch
Matthias Kohler
kohler@arch.ethz.ch
1
Eversmann Studio, Ohmstr.13, 80802 Munich, Germany
2
Gramazio Kohler Research, Professur fu
¨r Architektur und
Digitale Fabrikation, ETH Zu
¨rich HIB E 43, Stefano-
Franscini-Platz 1, 8093 Zurich, Switzerland
123
Constr Robot (2017) 1:49–60
https://doi.org/10.1007/s41693-017-0006-2

high global assembly precision
1
can be realized. Since
fabrication is directly connected to a precisely planned
virtual model, the danger of mistakes in construction is
very low and global precision extremely high, resulting in a
cost and construction efficient system.
1.2 Previous studies in digital timber construction
Subtractive digital fabrication and structural design for
timber construction have been extensively researched at
EPF Lausanne (Weinand 2009; Robeller et al. 2014;
Robeller and Weinand 2016), while the fabrication of
timber plate structures through robotic milling in combi-
nation with manual assembly has been studied on various
research projects at the ICD Stuttgart (Knippers and
Menges 2013; Menges 2012). Additive assembly processes
have been investigated notably by Gramazio Kohler
Research at the ETH Zurich, involving horizontal stacking
of simple linear wood slats to create complex geometry as
for the realisation of the roof of a new faculty building
(Apolinarska 2016a,b), and also the robotic assembly of
single-joint spatial structures in combination with gluing
technology (Zock et al. 2014; Helm 2016). A recent col-
laborative research project by Søndergaard and the authors
of this paper (Søndergaard 2016) investigated robotic
timber fabrication for topology optimized structures. Real-
time control and interaction during assembling wood sticks
of random geometry were explored by D}
orfler (2012)at
TU Vienna. Algorithmic techniques for using irregular
wood components to design non-standard structures were
explored by Monier et al. (2013), while scanning and
processing technologies for natural wood branches were
researched by Schindler et al. (2014) at CITA.
2
Techniques
for assembling wood slats autonomously using feedback
processes were recently experimented by Jeffers (2016)at
Carnegie Mellon University.
1.3 Large-scale robotic spatial assembly
The work presented in this paper builds on this develop-
ment, extending the scope to functional volumetric pre-
fabrication on an architectural scale. In Sect. 2, a new
robotic fabrication cell is explained, which was specifically
conceived for the involved fabrication technologies. We
demonstrate the spatial, structural, and functional possi-
bilities of highly diverse and customized timber frame
volumetric modules. For the implementation of these goals
in the 1:1 scale, a prototypical demonstrator was built
during a Masters program
3
in Digital Fabrication. In Sect.
3, we compare and analyze the robotic fabrication tech-
nologies in terms of their efficiency, as well as their
structural and functional capabilities and limits. Material
tolerances and variability are discussed with respect to
adaptive robotic processes. We conclude in Sect. 4by
showing potentials for future developments and comparing
spatial robotic prefabrication to on-site fabrication.
2 Methods
This section contains a detailed description of the robotic
technologies and their spatial arrangement necessary for
additive timber construction processes (Sect. 2.1). We
focus on four major subjects to demonstrate the construc-
tion capacities of the robotic cell with a large-scale case-
study fabrication project: the architectural prototype design
(Sect. 2.2), structure fabrication (Sect. 2.3), envelope
design and fabrication (Sect. 2.4), and on-site assembly
(Sect. 2.5). The goal of the case-study was to realize a
completely customized production in a continuous digital
workflow, with a robotic setup similar to ones used in
repetitive industrial applications.
2.1 Design of robotic setup
General setup The robotic cell was designed as a multi-
functional robotic tool for additive timber construction. It
was conceived to handle fairly unprocessed material of
various sizes as solid, slender timber beams of variable
cross section, and wood panels of variable dimensions.
Therefore, the installation of scanning devices and a real-
time connection of an external PC to the robot controller
for adaptive feedback control were necessary. The robots
were able to communicate with a range of external tooling
and also with each other for coordinated or synchronous
movements. The robotic setup was built on a movable
platform, so that the complete cell can be transported and
reconfigured in a matter of days.
Cell components Two industrial robots
4
were mounted
on a double carriage linear axis
5
of 5 m length. The cell
features a material feeding station configured for 5 m long
wood slats, a parallel pneumatic gripper
6
for centering and
fixing material during the saw cutting procedure, a CNC
saw equipped with servo motors for three controllable axes,
1
Global precision refers to an industrial robots capability and
precision of reaching coordinates and orientations in his workspace
(\1 mm).
2
Centre for Information Technology and Architecture, The Royal
Danish Academy of Fine Arts.
3
Master of Advanced Studies in Architecture and Digital Fabrication
of the NCCR Digital Fabrication at ETH Zurich.
4
ABB IRB 4600, reach: 2.55 m, payload: 40 kg.
5
ABB IRBT 2005.
6
Schunk PEH 30.
50 Constr Robot (2017) 1:49–60
123

and a custom designed worktable of 6 m 2.2 m with
integrated aluminum rails used for fixating wood structures
during build-up (see Fig. 1). A material feeding/scanning
table for wood panels was mounted on the left side of the
worktable. Robot 1’s (left) endeffector was equipped with a
measurement, scanning, vacuum gripping, and automatic
nailing tool. Robot 2 (right) was equipped with two parallel
electronic grippers controlled via a serial connection. They
could open and close on programmed distances, which was
especially important when performing movements while
releasing the work object in narrow parts of the structure.
All external tools were controlled directly through the
robot controller using a bus coupler
7
over a profinet con-
nection. Complex control routines were stored in system
modules (e.g., saw control/gripper control) and could be
accessed by simple functions in the programming interface
of Robotstudio.
8
Security concept In scientific studies, direct interaction
of the operators with the robots and the building process is
often needed to allow fast and continuous progress.
Therefore, a robust security protocol is necessary to limit
risks. Therefore, the main operating risks as the saw were
protected locally with polycarbonate shields, the maximum
speed was limited to manual mode (250 mm/s), and a
security distance to the robots during operation was defined
outside their maximum reach. Each building process was
precisely defined in a number of repetitive steps as
described later in detail in Sects. 2.3 and 2.4. These steps
were initiated and precisely controlled before full operation
mode by a professional operator. In addition, a precise
protocol was defined for operator interaction with the cell
especially for the saw operations and manual structural
fixation of the structure. In addition, each robot had its own
security workspace and some of the axes were limited in
range to limit collision potential between the robots and
tool heads.
2.2 Case study: architectural design
The aim of the prototype was to develop a design and
robotic fabrication system for a two-story structure with a
basic envelope. A complete digital workflow using additive
robotic fabrication processes was developed.
Programming workflow The general programming
workflow was defined in four steps. A set of input curves
controlled the overall geometry. The geometric detailing
occurred within a C# programming routine which was
directly connected with the structural calculation, enabling
optimization related to structural performance. All
fabrication data were generated in a Python routine. The
actual control, procedures, and operations of the robots
were written in rapid code within Robotstudio. In addition,
data acquired by the robots could be directly accessed
within the Python routines and could also be written on the
controller in real-time, enabling feedback loops during
construction. The set up as depicted in Fig. 2was used for
the envelope fabrication process explained in detail in Sect.
2.4.
Geometry An adjustable cuboid is the basis for the
geometry generation. When multiplied, the cuboids create
a spatially formed braced structure, which is capable of
translating seamlessly between different functions as wall,
slab, staircase, balustrade, etc. (Fig. 3).
The size and form of the modules were generated
regarding to functional, fabrication, and assembly con-
straints (Fig. 4).
2.3 Case study: structural design and fabrication
process
The geometry generation was directly linked to the struc-
tural design (FEM 2016) and optimisation of the structure.
Custom routines for multiple optimisations were integrated
to optimize for the defining load cases. We used the stan-
dard values from Swiss building code
9
for live loads, snow
and wind loads, and employed safety factors for the
calculation.
Optimization of bracings and cross sections We used
two strategies for general structural optimization: first by
adapting the overall bracing geometry and second by
allowing a range of different cross sections. As shown in
Fig. 3, each cuboids’ bracing orientations can be adapted
on its six faces and two diagonals. To use the bracings
mainly on compression, we developed an algorithm to
orient the bracings on the external shell-like faces follow-
ing the principal stress lines (Fig. 5right). We also per-
formed a custom-scripted cross section optimization,
resulting in four different cross sections (4, 6, 8, and 10
cm), reducing the amount of material by more than 30%
while maintaining visual integrity within the structure.
Structural connections Since each joint acts in a com-
bination between shear and axial forces, we used full-
threaded carbon steel screws. Considering the beam’s
thickness and the angle between the screw axis and the
directions of the beams’ fibres, we calculated the length
and orientation of the screw for each joint. We optimized
for four different sizes with similar diameter (Fig. 6). The
computational model also allowed us to deconstruct the
vectors of the forces that act on the screws in both the shear
7
Beckhoff EK9300 PROFINET-IO-Buskoppler for EtherCAT.
8
ABB controller specific programming environment in Rapid
language.
9
Self-weight, Safety Factor SF: 1.35; Live loads: 2 kN/m2(SF: 1.5);
Snow: 1 kN/m2(SF: 1.35) Wind: 1 kN/m2(SF: 1,5).
Constr Robot (2017) 1:49–60 51
123

Fig. 1 Design of a robotic setup featuring two robot arms and a 5 m
linear axis sitting on a mobile steel platform, a feeding station for 5 m
long wood slats equipped with a pneumatic gripper for centralizing
and holding slats during cutting procedure, a CNC saw of 600 mm
diameter with three controllable axes (360 horizontal orientation,
25–90 tilt, up/down 0–300 mm). Robot 1 (left) is provided with a
custom gripping and scanning tool and feeding station. Robot 2
(right) is equipped with two parallel electric grippers. A custom
working table was designed with rails for variable fixation of
elements of different sizes and scales
Geometry Detailing Fabrication Data Robot Control
Rhino GH C# - Karamba GH Python Robotstudio - Controller
Structure
Envelope
Axis Angles or Target Planes
Module Orientation
Saw Cut Angles
Drill Planes
Singularity Check
Collision Check
Shingle Gripping Plane
Shingle Placement Plane
Robot + Tool Data
Procedure Loop
External tool communication
Material Scanning
Optimization
Structural Detailing
(GH C#)
Structural Calculation
(Karamba)
Master Model
Control Curves
Feedback loop
Fig. 2 Programming Workflow: using provided design input curves,
all fabrication related geometry is generated in Rhino through a C#
script. Structural calculation and cross-section optimization are done
through the FEM software Karamba. A Python script organizes
fabrication data and calculates the assembly toolpath. In ABB
Robotstudio, the fabrication data are then used to drive all robotic
operations and control of external tools. Through direct computational
access to the controller, feedback is enabled during fabrication
52 Constr Robot (2017) 1:49–60
123

and axial components to check whether the load capacity of
the screws was not overreached. To avoid brittle fracture
behavior of the connections, all minimal distances between
the screw holes and the beams’ faces were verified.
Robotic Fabrication The large number of sub-proce-
dures integrating a range of external tooling of the struc-
tures fabrication processes makes it necessary to have an
extremely robust computational and robotic protocol
(Fig. 7). Information on each beam is stored in a general
array defining whether a fabrication procedure like cutting,
drilling, etc. will be performed (1) or not (0). Separate
arrays deliver the specific information on saw angles,
approach planes, and movement data. First, a wood slat is
placed on the roller rack, and centred by closing the saw
gripper to the reference position for robotic gripping. Next,
the saw gripper releases, allowing the robot to move to the
beam’s first programmed cutting position. During the saw
rotates axes 1 and 2 in position, the beam is held 5 mm
above the table, then the robot presses the slat on the table,
the saw gripper closes for additional stability, and the
cutting procedure is performed through moving axis 3 up
and down a programmed height depending on the saw’s
inclination and the beam size. For the second cut, the last
steps are repeated (Fig. 8). The robot can then lift the
workpiece to perform additional operations such as pre-
drilling or mount the beam directly on the structure. We
developed a path planning algorithm organizing the posi-
tions to grip, move, approach and place each element. The
algorithm sorted the beams depending on orientation, type,
and position of the beam in the structure. Since the struc-
ture is based on a simple geometric principle, as described
in Sect. 2.2, we were able to generate the approach
movements towards the final assembly positions in the
structure based on topological features (orientation, length,
position) rather than having to solve path planning for each
single element to avoid collisions with the already assem-
bled parts. In its final mounting position, the fixation
screws are attached manually (Fig. 8).
Fig. 3 Geometry generation: a cuboid serves as basic element
multiple geometrical connections and permutations are possible
within its faces
Fig. 4 Spatial module design: size, form, and sequential order of the
spatial modules produced by the robotic setup have to correspond to
functional, fabrication and assembly constraints. In our case study, the
complete structure was composed out of 46 separate modules of
maximum dimensions of 5 21:5m
Fig. 5 Structural design of double-story structure of dimensions of around 8:557 m (h). Left FE analysis showing resulting maximum
utilisation of members in relation with wall openings. Right Orientation of bracings following stress lines
Constr Robot (2017) 1:49–60 53
123

2.4 Case study: envelope design and fabrication
process
The design project was conceived as a minimal experimental
space, in which the envelope served primarily as weather
protection without consideration for thermal comfort (Fig. 9).
The idea was to use the maximum of material from the natural
form of a tree, resulting in variable widths of the timber panels
(shingles). Their exterior side got a rough and rich texture
through manual splitting, having the advantage of being much
more durable than machine sawn, since the material can dry
more easily(Niemiec and Brown 1993). We excluded any pre-
sorting into width categories, but instead used the intelligence
of the computational system to scan the shingle and calculate
and place each element according to its size. The final facade
geometry appeared only after a unique fabrication procedure.
For material and fabrication efficiency, we integrated the
envelope and primary structure in one system without the need
for substructure. This resulted in less material usage and less
fabrication time but also in a more distinct and visible logic
between envelope and structure: the primary structure
responds to the structural requirements of the envelope,
leading to a denser vertical span to the exterior side of the
trusses (max. 45 cm vertical span of bracings corresponding
to 60 cm shingle height). The facade directly reflects the
distribution of main stresses in the primary structure, resulting
in multi-layer cover of shingles in areas, where the bracings
are densely concentrated. The robotic setup consists of four
different components: (1) a sensor that allows scanning the
different widths of the shingles; (2) a scanning table; (3) a
vacuum gripper; (4) two nail guns that are activated by an air
compressed trigger (Fig. 10). The computati onal workflow for
10
6
4
46
Bracings
Verticals
8
VGZ 7x140
WT 6,5x65
VGZ 7x140
VGZ 7x140
VGZ 7x140
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
VGZ 7x100
WT 6,5x90
WT 6,5x90 WT 6,5x90
HBS 6X60WT 6,5x65
α = 30°α = 60°α = 90° α = 90° α = 60° α = 30°
in cm
VGZ 7x140
HBS 6X60
VGZ 7x100
WT 6,5x90
Fig. 6 Left Different types of full-threaded screws used for the joints. Right Calculation of screw type, length, and angle depending on
corresponding geometry and material thickness
Fabrication Data Programm Structure Specific Control
Rhino Robotstudio - ControllerRobotstudio - Controller
Gripping Planes
Saw Angles
Drill Target Planes
Approach Axis Angles
Placement Target Planes
Retract Target Planes
Initialise Cell
(Grippers, Saw, Variables)
loop:
Grip
Cut (Start - End)
Drill (Start - End)
Transfer to Place
Approach movements
Place / Release
Retract Movements
Call operations to perform:
GeneralBeamInfo{15}:= [1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1]
Example suboperations for gripping:
• move to beam specific cut length
• open Gripper
• close Saw clamp
• move to Gripping Position
• close Gripper
• open Saw clamp
Fig. 7 Data workflow for structure fabrication. Data for each procedure are read in data modules. The robotic procedures are standardized in
code and are repeated, even though movements and orientations are specific and customized for each element
54 Constr Robot (2017) 1:49–60
123

Fig. 8 Assembly sequence of wooden slat in truss element. a5 m long solid spruce slats are placed on the assembly table and the robot moves in
its home position. bThe robot grips the beam and moves it towards the saw, where it gets cut the programmed number of times
Fig. 9 Assembly sequence of wooden slat in truss element: the beam gets placed in the truss, where it is fixated manually
Constr Robot (2017) 1:49–60 55
123

the envelope consists of the following steps which will be
subsequently explained in more detail: acquiring the mea-
surement and reference data of the structure, getting the
geometry from shingle scanning, calculating custom gripping
and placing positions, and sending the data to thecontroller for
execution. Measurement data were necessary due to the high
precision requirements of the envelope process, and was
acquired through reference points of the robotically built
module of the primary structure. The scanningprocess using a
photoelectric proximity sensor works in the following way:
the robot moves linearly along a specific coordinate system
relative to the shingle-feeding station (Fig. 10b) until a signal
is triggered by sensor. The width of the shingle can be
deducted by the position coordinates. The width information
is sent to an external computer to calculate the new placement
and also thegripping position through a GH Python procedure.
Fig. 10 a Shingle tool is equipped with a light-reflex sensor, a
vacuum gripper fixated with additional springs and a soft gripping
form of the shingle, and two standard nailguns. bFeeding table used
for the robotic scanning process. During the scanning process, the
robot moves in relation with a specific coordinate system until
reaching the shingle edge
Scan
Grip Mount
Geometry Calculation Fabrication Procedure
Substructure Measurement
Feedback: Panel Size
Compressed air 1
Air suction
Compressed air 2
2
r
e
g
g
i
r
T
1
r
e
g
g
i
r
T
Sensor
Compressed air 1
Air suction
Compressed air 2
2
r
e
g
g
i
r
T
1
r
e
g
g
i
r
T
Sensor
Compressed air 1
Air suction
Compressed air 2
2
r
e
g
g
i
r
T
1
r
e
g
g
i
r
T
Sensor
(a) (b) (c)
(d)
Fig. 11 Robotic fabrication process of shingle facade of unknown
geometry: first, the shingle is scanned (a), the geometry is processed
and assembly information sent back to the controller, then the robot
grips the shingle at a specified position (b) and finally mounted and
fixated on the structure (c). dOverall procedure
56 Constr Robot (2017) 1:49–60
123

Once the data are sent back to the controller, the robot grips the
shingle on its newly defined position, orients it to the final
placement position, moves along the linear axis to the place-
ment position, presses the shingle on the structure, and actu-
ates the two nailguns for shooting the fixation staples (Fig.11).
2.5 Case study: on-site assembly
The modules arrived on site as described in the previous Sects.
2.3 and 2.4 completely prefabricated with structure and
envelope. Only at the joints, there was a space left uncovered
for easier access for on-site hoovering and joining. For the
foundations, we used removable earth screws of 1.6 m length
with footings for wood columns adjustable in three dimen-
sions. Steel bolts were used to clamp the modules structurally
together. The precise prefabrication of a large number of
highly unique modules allowed us to refrain from producing a
large number of paper plans normally being necessary for on-
site construction. Since the manual connections were extre-
mely simple and the specific geometry of each part could only
fit at one place in the building, assembly errors were practi-
cally impossible (Fig. 12). A simple three-dimensional view
of the numbering and form of parts and their connections
provided sufficient information on the construction site.
3 Results
The robotic system allows fabrication of customized three-
dimensional volumetric elements of complex geometry,
integrating functional and architectural requirements.
Adaptive robotic processes can be used in combination
with fairly unprocessed material of naturally variable sizes
to limit material waste. In our case-study project, over 4000
differently shaped elements could be assembled in only 5
weeks including testing and on-site assembly (Fig. 13).
3.1 Structure
Structural and fabrication constraints were negotiated in
optimizing the overall geometry, length, and orientation of
bracings. Even with 30 people on the top floor, the struc-
ture easily stayed within calculated maximal deflections.
With the current setup, the most efficient fabrication length
of timber slats was between 400 and 1500 mm. Shorter
elements were challenging to cut and assemble, while
longer elements led to tolerance issues. Since the robotic
mounting precision is consistently high (\1 mm), the
global assembly tolerance in the joints performs better
when using short slats and, therefore, small tolerance. The
performance of connections with full threaded screws is
directly related to the angle between the screw and the fibre
direction of the beams. Since, in the case-study project, all
angles are different, the effectiveness of each joint is
varying over the whole geometry. A spatial truss being
braced in all directions, therefore, assumes a certain
structural redundancy. Further research in connection
technology is needed to eliminate these redundancies and
safety factors for the structural detailing. Another impor-
tant topic was the definition of the robotic approach and
assembly movements in a densely populated structure.
There, the classification of beams by topological features
proved to be a simple way to avoid custom iterative path
planning for each element which is computationally
expensive. For a complete industrial process, the structure
would have to be either more regular and within the above-
mentioned beam size limits, or each assembly path has to
be computed separately, which is also subject to current
Fig. 12 a Pre-assembly of floor elements for tolerance control and beam connections for site assembly; bOn-site assembly of a large
prefabricated module using a small construction crane
Constr Robot (2017) 1:49–60 57
123

research in multi-robotic applications (Parascho and Gan-
dia 2017).
3.2 Envelope
The envelope fabrication of the case-study project
demonstrates the use of robotic feedback processes on
large-scale applications with thousands of different ele-
ments. The adaptability of these processes even allows the
use of highly imprecise material in a continuous fabrication
process. Accompanying computation, a physical robotic
‘‘softness’’ was also necessary. We used flexible elements
such as gripping foam and holding springs for this purpose.
Orientation tolerances can, however, still cause failures.
Fig. 14 Exterior image showing the shingle pattern and opening sequence of the structure
Fig. 13 Interior images of final structure on the upper floor, showing seamless integration of balustrade, floor elements, walls, roof structure, and
staircase (a). The size of the prefabricated chunks is still visible through the doubling of structural elements bimage courtesy of Kasia Jackowska
58 Constr Robot (2017) 1:49–60
123

Therefore, either a general physical flexibility of the end-
effector or computationally-induced soft robotic move-
ments can improve the general robustness of the robotic
process. Since we used a double overlay of the shingles in
the vertical and horizontal directions, we could also apply
fixation staples on the lower shingle edge. In traditional
shingle facades, a triple overlay in only the vertical
direction is usually applied (HOLZBAU 2017). For the
doubly curved base geometry, the additional bottom fixa-
tion greatly improved the stability of the shingles and also
their geometrical positioning, since the shingles are all
slightly cold-bent through the fixation procedure. Even
during heavy rain, the envelope remained waterproof
(Fig. 14).
4 Conclusion
Experimental results show that robotic fabrication of
unique, highly complex volumetric modules for on-site
assembly has significant potential. The modules can be
produced and integrated with all architectural, technical,
and functional parameters. Off-site prefabrication has the
advantage of a controlled and predictable fabrication
environment, resulting in high precision and high general
building quality. Even though transportation is less effi-
cient compared to on-site fabrication due to the abundance
of hollow forms, sustainability performance is still higher
compared to the conventional fabrication (Chao 2013).
Form and size of modules are also dependent on trans-
portation, which may affect structural and assembly
requirements. The current robotic setup is semi-mobile and
partly spatially configurable. Like in our case-study pro-
ject, the full spatial potential can be realized when using it
as an on-site prefabrication facility, liberated from trans-
portation constraints. This also provides the possibility of
integrating continuous adjustments and optimization even
during the building process. In terms of automation, it still
remains a large challenge to find efficient robotic processes
capable of integrating all functional requirements such as
thermal insulation, air-tightness, and technical systems in
continuous fabrication logic.
Acknowledgements The case study project was realized in the
framework of a Master of Advanced Studies class on digital fabri-
cation with the students Jay Chenault, Alessandro Dell’Endice,
Matthias Helmreich, Nicholas Hoban, Jesu
´s Medina, Pietro Odaglia,
Federico Salvalaio, and Stavroula Tsafou. This study was supported
by the NCCR Digital Fabrication, Funded by the Swiss National
Science Foundation SNSF. (Agreement # 51NF40-141853), and
builds directly on research findings and developments from the NRP-
66/ SNSF research project ‘‘Additive Robotic Fabrication of Complex
Timber Structures’’, established in collaboration between ETH Zur-
ich, Bern University of Applied Science and Nolax AG. We would
like to thank the companies Schilliger Holz AG, Rothoblaas, Krinner
Ag, ABB, and BAWO Befestigungstechnik AG for their generous
support. We would also like to thank P. Fleischmann and M.
Lyrenmann for their advice and continuous efforts for the robotic
setup and V. Helm and E. Schling for reviewing this article.
References
Apolinarska A et al (2016) The sequential roof. In: Menges A et al
(eds), Advancing wood architecture, Routledge, London
Apolinarska A et al (2016) Mastering the Sequential Roof. In:
Advances in Architectural Geometry 2016, vdf Hochschulverlag
AG an der ETH Zurich. doi:10.3218/3778-4_17
Beyer P-H (1991) Technologie von CNC-Holzbearbeitungsmaschi-
nen, 2nd edn. Cornelsen/Schwann-Girardet, Dusseldorf
Boafo F, Kim J-H, Kim J-T (2016) Performance of modular
prefabricated architecture: case study-based review and future
pathways. Sustainability 8:558. doi:10.3390/su8060558
Chao M et al (2013) Comparative study of greenhouse gas emissions
between off-site prefabrication and conventional construction
methods: two case studies of residential projects. Energy Build
66:165–176
D}
orfler K et al (2012) Interlacing, an experimental approach to integrating
digital and physical design methods. In: Robotic fabrication in
architecture, art and design, Springer, New York, pp 82–91
FEM - software (2016) Karamba. www.karamba3d.com. Accessed 25
Nov 2016
Hans Hundegger AG (2016). Products. www.hundegger.de. Accessed
25 Nov 2016
Helm V et al (2016) Additive robotic fabrication of complex timber
structures. In: Menges A et al (eds), Advancing wood architec-
ture, Routledge, London
Holzbau Entwicklungsgemeinschaft (2017) Regeln fu
¨r die Verwen-
dung von Holzschindeln fu
¨r Auenwandbekleidungen. Bauten mit
Holz, 6 Jg, S 86
Internationale Konferenz zur Automation in der Holzwirtschaft
(2006) 12. und 13. Oktober 2006, Biel, Schweiz, Biel : Berner
Fachhochschule - Architektur, Holz und Bau
Jeffers M (2013) Autonomous robotic assembly with variable
material properties, robotic fabrication in architecture, art and
design. Springer, New York, pp 48–61
Jeska S, Pascha KS (2015) Emergent timber technologies materials,
structures, engineering, projects. Birkhuser, Basel
Knippers J, Menges A (2013) ICD/ITKE research pavilion 2011. In:
Hu C (ed), Architectural material and Texture I, pp 266–273
(2013). ISBN 978-7-214-08683-9
Menges A (2012) Material computation higher integration in
morphogenetic design. Archit Des Vol 82, No 2, Wiley
Academy, London (2012). ISBN: 978 0470973301
Monier V, Bignon J-C, Duchanois G (2013) Use of irregular wood
components to design non-standard structures. Adv Mater Res
671–674:2337–2343. ISSN: 1662-8985
Niemiec SS, Brown TD (1993) Care and maintenance of wood
shingle and shake roofs. Oregon State University Extension
Service, EC 1271
Parascho S, Gandia A et al (2017) Cooperative fabrication of spatial
metal structures. In: Menges A, Sheil B, Glynn R, Skavara M
(eds), Fabricate 2017, UCL Press, London, pp 24–29
Popovic D, Fast-Berglund A, Winroth M (2016) Production of
customized and standardized single family timber houses A
comparative study on levels of automation. 7th Swedish
Production Symposium, vol 1
Robeller C, Hahn B, Mayencourt P, Weinand Y (2014) CNC-
fabricated dovetails for joints of prefabricated CLT components.
Bauingenieur 89:487–490
Constr Robot (2017) 1:49–60 59
123

Robeller C, Weinand Y (2016) Integrale Verbindungen fr Faltwerke
aus Holzwerkstoffplatten. Detail -Munchen 1:68–74
Sass L (2007) Synthesis of design production with integrated digital
fabrication. Automation in construction 16, Elsevier, Amster-
dam, pp 298–310
Scheurer F (2011) Digitaler Workflow im Freiform-Holzbau. Forum
Bois Construction Beaune, Biel, Forum Holzbau
Schindler C, Tamke M, Tabatabai A, Bereuter M, Yoshida H (2014)
Processing branches: reactivating the performativity of natural
wooden form with contemporary information technology. Int J
Archit Comput 12(2):101–115. doi:10.1260/1475-472X.12.2.101
Søndergaard A et al (2016) Topology optimization and robotic
fabrication of advanced timber space-frame structures. Robotic
fabrication in architecture, art and design, Springer, New York,
pp 190–203
Weinand Y (2009) Innovative timber constructions. J Int Assoc Shell
Spatial Struct 50(2):111–120
Zock P, Bachmann E, Gramazio F, Kohler M, Kohlhammer T,
Knauss M, Sigrist C, Sitzmann S (2014) Additive robot-
ergestu
¨tzte Herstellung komplexer Holzstrukturen, 46. Tagungs-
band Fortbildungskurs Holzverbindungen mit Klebstoffen fr die
Bauanwendung, pp 197–208
60 Constr Robot (2017) 1:49–60
123