Use of a Low-Cost Humanoid for Tiling as a Study in On-Site Fabrication Techniques and Methods

Abstract

Since the time architecture and construction began embracing robotics, the prefab movement has grown rapidly. As the possibilities for new design and fabrication emerge from creativity and need, the application and use of new robotic technologies becomes vital. This movement has been largely focused on the deployment of industrial-type robots used in the (automobile) manufacturing industry for decades, as well as trying to apply these technologies into off-site building construction. Beyond the prefab (off-site) conditions, on-site fabrication offers a valuable next step to implement new construction methods and reduce human work-related injuries. The main challenge in introducing on-site robotic fabrication/construction is the difficulty in calibrating robot navigation (localization) in an unstructured and constantly changing environment. Additionally, advances in robotic technology, similar to the revolution of at-home 3D printing, shift the ownership of modes of production from large industrial entities to individuals, allowing for greater levels of design and construction customization. This paper demonstrates a low-cost humanoid robot as highly custom-izable technology for floor tiling. A novel end-effector design to pick up tiles was developed, along with a localization system that can be applied to a wide variety of robots.

Full text

214
Use of a Low-Cost Humanoid
for Tiling as a Study in On-Site
Fabricaon
Mathew Schwartz
Advanced Instutes of
Convergence Technology,
Seoul Naonal University
Techniques and Methods
ABSTRACT
Since the me architecture and construcon began embracing robocs, the pre-fab movement
has grown rapidly. As the possibilies for new design and fabricaon emerge from creavity and
need, the applicaon and use of new roboc technologies becomes vital. This movement has been
largely focused on the deployment of industrial-type robots used in the (automobile) manufacturing
industry for decades, as well as trying to apply these technologies into o-site building construc-
on. Beyond the pre-fab (o-site) condions, on-site fabricaon oers a valuable next step to
implement new construcon methods and reduce human work-related injuries. The main challenge
in introducing on-site roboc fabricaon/construcon is the diculty in calibrang robot navigaon
(localizaon) in an unstructured and constantly changing environment. Addionally, advances in
roboc technology, similar to the revoluon of at-home 3D prinng, shi the ownership of modes
of producon from large industrial enes to individuals, allowing for greater levels of design and
construcon customizaon. This paper demonstrates a low-cost humanoid robot as highly custom-
izable technology for oor ling. A novel end-eector design to pick up les was developed, along
with a localizaon system that can be applied to a wide variety of robots.
1 The NAO robot with moon
capture markers aached while
inside the capture volume of the
moon capture studio.
1

215
GENERATIVE ROBOTICS
INTRODUCTION
Construction Injury
While robocs plays an important role in increasing eciency
and quality in manufacturing, it is also vital in reducing workers’
exposure to dangerous and hazardous manufacturing and
construcon processes. Although the displacement of jobs
connues to be a highly controversial topic, the reducon of
construcon injuries is of great benet, as manufacturing and
construcon industries represent a large share of both fatal and
non-fatal work accidents (OECD 2007). While many factors,
such as regulaons, have reduced workplace deaths, the early
1990s had seen a risk of death in construcon three mes higher
than all other workers combined (Jeong 1998). Based on current
conversaons within manufacturing about compliant robocs
being necessary for human-robot coexistence, workplace inju-
ries in construcon have been in large part accidents caused
by slipping or falling objects, rather than direct contact with
machinery (Jeong 1998). In 2002, falling totaled 56% of injuries,
while being hit by falling objects was 17%, suggesng that such
injuries connue to be a common on-site issue (Haslam et al.
2005). Furthermore, injuries are not limited to an immediate loss
of life or health, but include long-term eects from sicknesses
and diseases from exposure to lead, asbestos, or even radiaon.
As seen from the March 11, 2011 earthquake disaster in Japan,
human exposure during reconstrucon and debris removal is
extremely dangerous and results in a wide range of health condi-
ons. For these and other reasons, it would be highly benecial
for society to develop strategies for on-site roboc construcon,
even without considering economic and design benets.
Teleoperation
In this regard, the robot can be controlled with either
one-to-one mapping or from a remote site with some degree
of autonomy. A popular applicaon of teleoperaon is in the
medical eld of surgery. The surgeon is able to control the
robot through an interface, such as a hapc feedback pen,
allowing for more accurate and ne-tuned control during the
procedure (Marohn and Hanly 2004). While this is close to a
one-to-one mapping, other aspects of medical robocs, such
as wrist injuries during ultrasounds, may benet from a more
intelligent teleoperated roboc assistant (Con, Park, and
Khab 2014). In the case of surgery and ultrasound imaging,
the roboc system is in a xed locaon. In unstructured envi-
ronments, such as buildings, a dierent type of robot would
be required.
Immediately aer the Japan disaster, teleoperated systems were
originally deployed using unmanned construcon technologies
(OSUMI 2014). These remote-control technologies may include
greater autonomy, to the degree that a single human operator
could concurrently control mulple machines (Bock, Linner, and
Ikeda 2012).
During 2014 and 2015, researchers had prepared for the DARPA
Robocs Challenge (DRC) (Johnson et al. 2015), which aimed to
advance the state-of-the-art for robocs research. Based on the
disaster in Fukushima, the need for robots to navigate unstruc-
tured environments that were designed for people became
apparent. Oen, the general public views humanoid robots as
futurisc servants dealing with house chores and taking coats
from guests. Even so, current household robots for chores such as
vacuum cleaning have been reduced to simple cylinders on wheels.
However, as the built environment is designed for people, the
simplest way for robots to ulize the same space is to share
similar features and general physical qualies of humans. This
was seen as a crical part of the DRC, since robots were required
to climb stairs or even a ladder, and be able to operate (shut o)
a valve. A secondary aspect to the DRC was the intermient
communicaon between the operator and robot. The operator
was able to receive a limited amount of data from the robot,
requiring the implementaon of three general approaches: long
periods of waing (in which a me rule was implemented), a fully
autonomous robot able to complete the tasks, or a combinaon
of both. In all cases, the human was removed from danger, and
in each case, various levels of research were needed. While this
research is sll ongoing, the goal is to develop diversied robots
to navigate an unspecied space with various obstacles, rather
than develop special machinery for each task in which it may
rarely need to be teleoperated (Hasunuma et al. 2002).
On Site
One reason pre-fab has more successfully implemented robocs is
the presence of structured environments. When robots move on
site to an unstructured environment, similar tasks become more
dicult (Feng et al. 2015). While the use of a robot on site was
acknowledged in the research of Ariza and Gazit (2015), the robot
relied on a structured environment through predened rotaons.
The discussion on using robots for on-site construcon is not
new. In 1985, while recognizing that commercial robots for
construcon in any sense were non-existent, the feasibility of
an on-site robot was doubted due to the need for “program-
ming and installaon for each parcular case” (Warszawski and
Sangrey 1985). Understandably, the idea of automaon in terms
of roboc autonomy was not yet suciently developed to the
point that it would seem valuable.
Addionally, manufacturing of large complex systems can be
viewed in a similar way as on-site construcon, as in the case of

216
Tiling Robots
There has been a sporadic history in the use of robots for ling
or mosaic operaons. When the panel is already available, albeit
a single ceramic le or a composed mosaic, researchers have
explored the process of laying the le on site. In the work of
Ahamed Khan et al. (2011), the oor ling system is semi-auton-
omous. In this case, the roboc system is a method of reducing
strain on the worker as well as speeding up the process. In
contrast, other researchers theorize an autonomous machine
to lay les by comparing specic aspects of a robot’s capability
with small-scale experiments with studies of human labor cost
and fague (Navon 2000; Apostolopoulos, Schempf, and West
1996). These robots are large machines intended for large-scale
installaons and have not demonstrated the capability for actual
development and implementaon.
Current State
This research demonstrates the current ecacy and capacity of
a low-cost humanoid meant for consumer experimentaon. The
basis for replacing humans on the construcon site is due to the
dangers of the construcon industry, and the benets of robots
being able to perform construcon tasks directly on site aords
value beyond safety. In keeping with the target point of the robot
used, a novel end eector for the le pickup was designed for
temporary use and implementaon with 3D prinng and low
cost manufacturing. Although the system used here is based
on moon capture, the methods in which the robot’s kinemac
frame is found within the space can be transferred to other local-
izaon systems as well as other robots.
While the robots in the DRC used a variety of onboard vision
sensors, this work relies solely on an external localizaon system
and kinemac calculaons. For single-use robots, an external
localizaon system may be ideal, and a network of construcon
robots or a permanent live-in robot would greatly benet from a
data center to coordinate and control robot posions and tasks.
As an inial step, this research focuses on the ability to rely
purely on kinemacs based on an external localizaon system,
making this a visually unassisted on-site robot.
LOCALIZATION METHODOLOGY
System Overview
This research uses the V4 NAO robot from Aldebaran, controlled
using the C++ API and the inverse kinemacs soluon as
detailed in the work of Konas, Orfanoudakis, and Lagoudakis
(2013). The method for controlling the humanoid walk is similar
to the methods aorded through the API with a few variables
modied to compensate for the le weight.
The moon capture system is from Vicon, with (12) T-160
Use of a Low Cost Humanoid for Tiling as a Study in On-Site Tiling Schwartz
airplanes. Recognizing the need for versale lightweight robots
to navigate the structure, AirBus Group has begun research
into implemenng humanoids for their airplane manufacturing
(Stasse et al. 2014). In this type of situaon, a wheeled robot
may not be the most praccal choice, but rather one with some
type of legs. Construcon sites in general are dangerous areas,
regulated by governmental bodies and laws (Weil 1992). The
ulizaon of robots on site, albeit through autonomy or teleop-
eraon, can be lifesaving. Further, the use of humanoid robots
allows for a versale robot to use a variety of tools to accomplish
a variety of tasks within a space developed for humans. This use
of humanoids has implicaons for where the ulizaon of roboc
manufacturing begins and ends, as these robots are not limited
to manufacturing, and may be permanent or semi-permanent
agents in the space.
Another clear disncon with commonly used roboc techniques
in architecture and the proposed on-site methods is within the
generaon of toolpaths. Unlike the precisely pre-planned moons
used in CNC or roboc manufacturing, an on-site robot is given
a task, not a specic moon (such as joint angles over me). This
task is processed by the robot, and what would be considered a
toolpath is generated in real-me. Real-me generaon is needed
for the robot to navigate unstructured environments, where
modicaons to the moon control are based on this new or
changing environment, in order to accomplish the given task. As
an example, the arm moon of picking or placing a le is depen-
dent on the robot’s orientaon to that le. The ability to calculate
the robot’s arm moon based on any orientaon, rather than
forcing the robot’s overall orientaon into an exact match each
me, allows the robot to adapt within the space. This workow is
further detailed in the methodology secon.
At Home
Just as the cost of 3D printed manufacturing dropped drama-
cally as it moved from industry to retail market, so has the cost
of robocs. Robots such as the humanoid NAO (Gouaillier et al.
2009) have begun entering a somewhat aordable price point
while maintaining high levels of funconality, as demonstrated
in the RoboCup soccer league (RoboCup Technical Commiee
2015). As with the use of humanoids for disaster situaons and
specialized manufacturing, the benet of a humanoid robot at
home, beyond the social aspect, is the ability for navigaon and
manipulaon in a built-for-people environment. Beyond the
dicules of a wheeled robot traversing ground with raised
secons, the bipedal form reduces the footprint required of the
robot, allowing for a smaller turning radius and ability to t into
more dicult spaces. Furthermore, the current state of low-cost
humanoids, and robots in general, is comparable to the avail-
ability of 3D prinng in the late 1990s.

217
GENERATIVE ROBOTICS
Once this inial frame is created, the frame of the robot in world
space must be created in order to save the oset between the
two frames, and connuously use the inial frame as a reference.
To do this, the head joints are rotated one aer the other. During
each rotaon on a single axis, any of the three original markers
are stored at three points in me. With these three points, the
axis of the motor can be found by circumscribing a circle. The
center point of the circle, which will be the center of the motor,
can be found using Equaon 2.
The motor axis can be found using the normal of the circum-
scribed circle using Equaon 3.
At this point, one of the head motor axes is known. The secondary
axis can be found the same way, represented in Figure 3.
With both motor axes found, two of the three required vectors
exist. However, due to manufacturing tolerances and the slight but
existent noise in the moon capture system, these axes are not
perfectly perpendicular. To esmate the origin of the neck frame,
the closest point to both vectors are found using Equaon 4.
2 Diagram represenng three markers or tracking points on the robot in any conguraon creang a frame.
cameras. Real-me streaming is accomplished with the Vicon
RealTime SDK through Nexus (Vicon 2014). The moon-capture
system is capable of sub millimeter accuracy, oering a viable
tesng plaorm for localizaon.
Localization
Two methods are used for localizaon of the robot. While in
perfect condions they would produce the same results, slight
errors in the encoders and manufacturing of the robot result in
a dierence of a few degrees between the calculated frames.
Both methods provide advantages and disadvantages. In the
rst method, the head joint is calibrated within the space. This
can be done with the robot placed directly in the space without
regard for orientaon as the markers are placed on that joint. In
the second method, a predened orientaon within the space is
designated as the inial robot end eector conguraon.
With an external localizaon system, it is necessary to calibrate
the robot posion of a known joint in order to correctly locate
the end eector in the space. In the case of a humanoid, the
head joint is suitable as it can have markers placed on it without
occlusion during moon. If an industrial arm is used, the base
frame would also be a viable candidate.
In order to begin calibraon, three markers must be placed on
the robot. In a dierent localizaon system, these markers would
be replaced with beacons, acve markers, or tracking markers for
camera-based systems. From these three points (P1, P2, P3), an
inial frame can be created, as seen in Equaon 1 and graphically
in Figure 2.
(1)
(2)
(3)

218
When the robot moves throughout the space, the marker posi-
ons are queried, the marker frame is created, and the stored
transformaon matrix is applied, giving the neck frame in world
space. As the Denavit-Hartenberg (DH) parameters from this
stored frame are known, the end eector locaon in world space
can be calculated. In the next secon, a faster method for calcu-
lang the robot frame is described if the end eector is able to
be placed in a known locaon within the localizaon system.
While the use of the base frame as a calibraon joint allows for
the operator to place the robot within the space without regard
for orientaon or posion, it does take more me and calcu-
laons than beginning at a known conguraon. This second
calibraon method requires the robot end eector to be placed
in a known conguraon and calibrates using the DH parameters
and encoders.
With the end eector in a known locaon, the frame of the end
eector {p} is known in world space {w}, and the transformaon
to the base frame {h} can be calculated (Equaon 8).
Where,
As with the head rotaon method, this stored transformaon is
applied during the ling operaon.
TILING METHODOLOGY
End Effector
While many methods for vacuum systems exist, the mobile
nature of a ling robot requires a portable and light-weight
soluon. Therefore, an externally powered vacuum system with
tubing was not appropriate, and a custom end eector for the
robot was developed. As seen in Figure 4, the components are
3 Using one of the marker references in three locaons aer motor rotaon, a circle center and normal can be extracted, corresponding to the motor posion and axis.
To calculate the nal orientaon, the vectors found previously are
used to generate the third vector, and the cross product is used
again to maintain the primary axis as perpendicular (Equaon 5).
(4)
To create the frame, the midpoint found in Equaon 4 is used in
the translaon (last) column of a 4x4 matrix, and the orientaon
vectors found in Equaon 5 are used for the rotaon. Equaon
6 shows the nal homogeneous transformaon matrix from the
world(w) to the calculated(c) frame.
(5)
(6)
This calculated frame exists in world space but must be relave
to the original marker frame. For this, a transformaon matrix
from the head markers to the calculated frame is stored by
Equaon 7.
(7)
(8)
Use of a Low Cost Humanoid for Tiling as a Study in On-Site Tiling Schwartz

219
GENERATIVE ROBOTICS
a 3D printed housing (which can be seen in the Results secon)
with puller, silicon tube, spring, stopper, and sucon cup.
In the inacve state, the vacuum can always be created, as
a normal sucon cup works as the spring pushes against
the stopper and into the tubing. When the robot ngers
are acvated, the stopper is pulled away from the tubing,
compressing the spring, and allowing air-ow, releasing the
vacuum. The second aspect to the design is the use of the ex-
ible tubing. As can be seen in Figure 4, the pulling lever is oset
from the locaon of the sucon cup. The ngers of the robot
are not in the center; as such, using a straight line for the end
eector would create another oset in the kinemac denion.
To simplify calculaons, the exible design is used to align the
sucon cup with the center of the wrist and elbow joints. With
this, the DH parameters can remain the same, with the nal
frame distance being extended by the length of the end eector.
This system can be applied to a variety of robots in order to
simplify the kinemacs. Addionally, the passive vacuum allows
for a reliable and low energy soluon, a necessity in mobile
robocs where baery life is vital and closely related to the
voltage demand of the motors.
The spring used in the system must be light enough for the robot
to compress, while being strong enough to overcome fricon of
the silicon tube in order to complete the seal. In this congura-
on, a spring was used which takes approximately 9 Newtons to
compress 6 mm, the distance needed to release the vacuum. With
a clean press of the end eector, a solid vacuum is formed, lasng
at least one hour during tesng. However, an o angle press of
the le drascally reduces the vacuum formed and can result in a
drop within a few minutes. Hence, it is vital for the end eector to
be suciently pressed against the le in this passive system.
Tile Planning
Once the robot is calibrated and reaches the le locaon, the end
eector must press against the le to pick it up. The le frame is
in world space and is the desired locaon of the end eector. For
simplicity, the le frame is then used as the end eector frame
for the inverse kinemacs soluon. As shown in the next secon,
State Machine, the frame is calculated before the robot picks up
or drops the le. If no soluon is found, the robot re-aligns to
the le. Once crouched in place for picking/dropping the le, the
frame from the le is copied vercally by a threshold under 30
mm. Mulple posions are queried by rotang the frame in order
to nd an inverse kinemacs soluon such that the path planning
is approached vercally.
State Machine
The state machine diagram shows the logic behind the robot
ling. As seen in Figure 6, calibraon occurs once at the
beginning. The methods for picking or dropping a le are the
same. Throughout the system, the robot queries the current
locaon of the markers in order to calculate the robot frame in
the world, using the oset described in the Methodology of
this paper.
4
5
4 Diagram of passive sucon end eector for robot.
5 Frame conguraon for pick/drop posion of robot arm. The le locaon is
coincident with the target posion of the end eector. The base frame is the
calibrated frame discussed in the localizaon secon, Joint 1 is the shoulder,
and Joint 2 is the elbow frame.

220
A main dierence between much of the work done in pre-
fab architectural/design research and this paper is the
implementaon of a state machine as the controlling factor. In
many cases, a CAD le or G-code is sent to the robot control
system to be interpreted as coordinates to move to. In this
system, the robot receives only a le lisng the colored les to be
used, and autonomously navigates to the goal posions. All steps
are handled within the state machine, including communicaon
delays and errors.
RESULTS
Although the main goal of implemenng robots on the construc-
on site is for safety and reducon of human injuries, both
accuracy and length of me for achieving the goals must be
considered. At the same me, implemenng this system in a
humanoid robot for at-home assistance may present dierent
challenges, such as a case where a person is unaware how to le,
or there is no me constraints on their project.
As this project is not focused on the development of a commer-
cial robot, but rather research and experimentaon for future
methods, accuracy and me-scale are not solved, but instead are
calculated to demonstrate the ecacy of the system, and what
would be required to solve them.
Experiment
The experimental results can be seen through mulple images
during a ling acon. The robot was set in the inial cong-
uraon on a wooden plaorm. The robot was commanded
to autonomously navigate to the le pickup locaon, as seen
in Figure 7. Black tape was used to repeatedly place the le
in the same posion within the moon capture space during
experiments. Aer pickup of the rst le (Red), the robot was to
move 1.5 meters to the back le and drop the le.
Aer picking the le, the robot recongures the arm posi-
on based on the state, ie. turning or walking. As the robot
approaches the goal posion, the le frame is calculated as the
end eector frame in the inverse kinemacs equaon. If a solu-
on is found, the robot will crouch down in order to pick/drop
the le as seen in Figure 8.
When dropping the nal le, the robot accurately places it. As
seen in Figure 8, the ngers grasp the lever, which releases the
vacuum on the le. However, retracon of the arm due to motor
limitaons is not directly vercal, shiing the orientaon of the
le by a few degrees.
Finally, the robot returns to the le pickup area to retrieve the
next le (Green), as seen in Figure 9. Due to baery limitaons
and accuracy of the end eector, the demonstraon experiment
concluded when the robot returned to the le pickup locaon.
Accuracy
While the localizaon system without the use of onboard vision
is shown to be eecve, the accuracy of picking a le prevents
a full mosaic from being created. The encoders on the robot are
within a reasonable tolerance, while the motors are either lacking
torque, or lack the appropriate low level drivers for incremental
moons. When commanding a joint a specic rotaon value, the
joint does not always achieve the desired locaon, and while
the encoder value acknowledges this, compensang for this
error is extremely dicult. A technique for overcompensang
in the opposite direcon of the motors failed aempt was
6 State machine diagram of the robot during the enre experiment. This state machine is independent of robot type and can be implemented with an industrial arm as well.
Use of a Low Cost Humanoid for Tiling as a Study in On-Site Tiling Schwartz

221
GENERATIVE ROBOTICS
implemented, but did not reduce the error to a useful degree.
For this reason, mulple aempts to pick up and drop a le are
done, increasing the me needed for a le cycle. While the robot
would be adequate as a placement system, a higher quality motor
is required for completely accurate ling.
However, applicaons of a limited accuracy robot may sll be
realisc. As recycled or damaged les can be used in mosaic
paerns, a combinaon of computer vision and the roboc
ling system can produce results similar to that of a non-spe-
cic mosaic le placement. Furthermore, input of the damaged
le’s shape and color can be processed by the robot to generate
numerous designs that t within the accuracy limits of the robot.
Times
The me to move from the calibraon locaon to a le, then a
drop locaon and back to the next le is broken down in Figure
10. By comparing this to the state diagram, it is clear that most of
the me spent is approaching the goal posion. While the robot
itself can walk at a moderate rate, as seen in both Tile Approach
secons, the inclusion of a le changes the mass locaon. To
compensate for this, the le is held in the front of the robot
during a forward walk, and low to the side for rotaon and minor
adjustments.
The second most me consuming aspect occurred during the
Drop Aempt. As the motors lack the torque to command small
movements, the nal end eector posion is not exact. Using
the encoders, the achieved posion can be calculated, and if
it is outside of the valid zone, the arm retracts and aempts
again, causing mulple Drop Aempts. As a conservave
esmate using the failed aempts in the meline, it takes 494
seconds to pickup, drop, and return to the pickup locaon. If
this was sustained, it would take 13 hours to le 100 les, a
1x1 meter square of 10 cm les—clearly not a reasonable length
of me for a construcon site, but sll achievable at a house
in order to make a mosaic oor paern autonomously over-
night. Furthermore, these mes can be drascally reduced by
a slight increase of torque in the motors, reducing the required
Drop Aempt mes. Second, further research into the mass
compensaon while carrying a le can reduce the approach
and adjustment mes. Assuming a 50% reducon in approach
me, and a single Drop Aempt, the single cycle would take 339
seconds, and 100 les 9.4 hours.
CONCLUSION
While this paper demonstrates the ability for a low cost
humanoid robot to place les in the correct locaons, it is neither
ecient nor accurate enough to be considered a viable opon
at this point. However, by either increasing the joint accuracy or
7
8
9
7 Snapshots of the robot in the moon capture space performing the experiment.
The black tape on the head secures the three markers described in the method-
ology. The end eector is 3D printed and secured around the hand with a metal
clasp.
8 Robot crouches down and places le in the desired locaon. The rst frame
shows the robot ngers relaxed and spread out, while the second frame shows
the ngers clenched, liing the end eector lever, and the nal image showing
the le dropped.
9 The robot returns to the le pickup locaon aer dropping the Red le at the
goal posion.

222
including an onboard vision feedback system, this work shows
that a soluon is within the near future. Addionally, the method
for calibrang the robot’s posion using an external localizaon
system can be used in the future. The end eector design is a
low cost and low power consuming opon for use with any type
of robot. Furthermore, integraon as a human-robot collabora-
on may increase eciency for aiding in drop placement while
providing a safe locaon for the laborer.
As buildings have autonomy built into them, the use of external
localizaon systems become common place. Either through
built in LIDAR or UWB for use in waynding, the cost of on-site
robocs decreases dramacally. When the robots need limited or
no vision capabilies due to control from a data center, the cost
per robot decreases while the cost per building increases, at the
same me aording exponenally more possibilies during the
life cycle of the building.
ACKNOWLEDGEMENTS
This research was funded in part by grant #AICT-2014-0011. The author
would like to thank the many interns of the Digital Human Research
Center over the last two years that have contributed their knowledge and
helped in the code development.
REFERENCES
Ahamed Khan, M. K. A., K. I. Saharuddin, I. Elamvazuthi, and P. Vasant.
2011. “A Semi-Automated Floor Tiling Roboc System.” In 2011 IEEE
Conference on Sustainable Ulizaon and Development in Engineering and
Technology. Semenyih, Malaysia: STUDENT. 156–159.
Apostolopoulos, Dimitrios, Hagen Schempf, and Jay West. 1996.
“Mobile Robot for Automac Installaon of Floor Tiles.” In Proceedings
of 1996 IEEE Internaonal Conference on Robocs and Automaon, vol. 4.
Minneapolis, MN: ICRA. 3652–3657.
Ariza, Inés, and Merav Gazit. 2015. “On-Site Roboc Assembly of
Double-Curved Self-Supporng Structures.” In Proceedings of the 19th
Conference of the Iberoamerican Society of Digital Graphic Design, edited by
Alice Theresinha Cybis Pereira and Regiane Trevisan Pupo. Florianópolis,
Brazil: SIGRADI. 746–753.
Bock, T., T. Linner, and W. Ikeda. 2012. “Exoskeleton and Humanoid
Roboc Technology in Construcon and Built Environment.” The Future of
Humanoid Robots: Research and Applicaons, edited by Riadh Zaier. InTech
Open Access Publisher. 111–146.
Con, François, Jaeheung Park, and Oussama Khab. 2014. “Interface
Design and Control Strategies for a Robot Assisted Ultrasonic
Examinaon System.” In Experimental Robocs: The 12th Internaonal
Symposium on Experimental Robocs, edited by Oussama Khab, Vijay
Kumar, and Gaurav Sukhatme. Berlin: Springer. 97–113.
Feng, Chen, Yong Xiao, Aaron Willee, Wes McGee, and Vineet R. Kamat.
2015. “Vision Guided Autonomous Roboc Assembly and as-Built
Scanning on Unstructured Construcon Sites.” Automaon in Construcon
59: 128–138.
Gouaillier, David, Vincent Hugel, Pierre Blazevic, Chris Kilner, Jérome
Monceaux, Pascal Lafourcade, Brice Marnier, Julien Serre, and Bruno
Maisonnier. 2009. “Mechatronic Design of NAO Humanoid.” In
Proceedings of 2009 IEEE Internaonal Conference on Robocs and
Automaon. Kobe, Japan: ICRA. 769–774.
Haslam, Roger A., Sophie A. Hide, Alistair G. F. Gibb, Diane E. Gyi, Trevor
Pavi, Sarah Atkinson, and A. R. Du. 2005. “Contribung Factors in
Construcon Accidents.” Applied Ergonomics 36 (4): 401–415.
Hasunuma, Hitoshi, Masami Kobayashi, Hisashi Moriyama, Toshiyuki
Itoko, Yoshitaka Yanagihara, Takao Ueno, Kazuhisa Ohya, and Kazuhito
Yokoi. 2002. “A Tele-Operated Humanoid Robot Drives a Li Truck.” In
In Proceedings of 2002 IEEE Internaonal Conference on Robocs and
Automaon, vol. 3. Washington, DC: ICRA. 2246–2252.
Jeong, Byung Yong. 1998. “Occupaonal Deaths and Injuries in the
Construcon Industry.” Applied Ergonomics 29 (5): 355–360.
Johnson, Mahew, Brandon Shrewsbury, Sylvain Bertrand, Tingfan Wu,
Daniel Duran, Marshall Floyd, Peter Abeles, Douglas Stephen, Nathan
Merns, Alex Lesman, John Car, William Rifenburgh, Pushyami Kave,
Wessel Straatman, Jesper Smith, Maarten Grioen, Brooke Layton,
Tomas de Boer, Twan Koolen, Peter Neuhaus, and Jerry Pra. “Team
IHMC’s Lessons Learned from the DARPA Robocs Challenge Trials.”
Journal of Field Robocs 32 (2): 192–208.
Konas, Nikos, Emmanouil Orfanoudakis, and Michail G. Lagoudakis.
2013. “Complete Analycal Inverse Kinemacs for NAO.” In Proceedings
of the 13th Internaonal Conference on Autonomous Robot Systems. Lisbon,
Portugal: Roboca. 1–6.
10 Breakdown of the mings of the experimental results.
Use of a Low Cost Humanoid for Tiling as a Study in On-Site Tiling Schwartz

223
GENERATIVE ROBOTICS
Marohn, Col. Michael R., and Capt. Eric J. Hanly. 2004. “Twenty-First
Century Surgery Using Twenty-First Century Technology: Surgical
Robocs.” Current Surgery 61 (5): 466–473.
Navon, R. 2000. “Process and Quality Control with a Video Camera, for a
Floor-Tilling Robot.” Automaon in Construcon 10 (1): 113–125.
OECD. 2007. “Work Accidents.” In Society at a Glance 2006: OECD Social
Indicators 2006 Edion. Organisaon for Economic Co-operaon and
Development.
Osumi, Hisashi. 2014. “Applicaon of Robot Technologies to the Disaster
Sites.” In Report of JSME Research Commiee on the Great East Japan
Earthquake Disaster. Tokyo: The Japan Society of Mechanical Engineers.
58—74.
RoboCup Technical Commiee. 2015. “Robocup Standard Plaorm
League (NAO) Rule Book.” Last modied July 1, 2015. hp://www.infor-
mak.uni-bremen.de/spl/pub/Website/Downloads/Rules2015.pdf
Stasse, Olivier, F. Morsillo, M. Geisert, Nicolas Mansard, Maximilien
Naveau, and Chrisan Vassallo. 2014. “Airbus/future of Aircra Factory
HRP-2 as Universal Worker Proof of Concept.” In 2014 IEEE-RAS 14th
Internaonal Conference on Humanoid Robots. Madrid, Spain: RAS.
1014–1015.
Vicon. 2014. “Nexus 1.8.5.” hps://www.vicon.com/downloads/
core-soware/nexus/nexus-185-installer.
Warszawski, Abraham, and Dwight A. Sangrey. 1985. “Robocs
in Building Construcon.” Journal of Construcon Engineering and
Management 111 (3): 260–280.
Weil, David. 1992. “Building Safety: The Role of Construcon Unions in
the Enforcement of OSHA.” Journal of Labor Research 13 (1): 121–132.
IMAGE CREDITS
All Photography: Schwartz, 2016
Mathew Schwartz has a BFA and a MSc. in Architecture with a focus
on digital technology from the University of Michigan. He is currently
at the Advanced Instutes of Convergence Technology in South Korea,
where he is a research scienst who focuses on human factors. His work,
which bridges science and engineering with art and design, makes use of
cung-edge robocs and moon capture technology to mimic human
characteriscs which he then incorporates into commercial applicaons,
architecture, and models used in scienc research.