The Supernumerary Robotic 3rd Thumb for Skilled Music Tasks

James Cunningham, Anita Hapsari, Pierre Guilleminot, Ali Shafti, and A. Aldo Faisal

Abstract— Wearable robotics bring the opportunity to aug-

ment human capability and performance, be it through pros-
thetics, exoskeletons, or supernumerary robotic limbs. The lat-
ter concept allows enhancing human performance and assisting
them in daily tasks. An important research question is, however,
whether the use of such devices can lead to their eventual
cognitive embodiment, allowing the user to adapt to them and
arXiv:1807.08274v1 [cs.RO] 22 Jul 2018

use them seamlessly as any other limb of their own. This paper
describes the creation of a platform to investigate this. Our
supernumerary robotic 3rd thumb was created to augment piano
playing, allowing a pianist to press piano keys beyond their Fig. 1: Our supernumerary robotic 3rd thumb (SR3T) in use.
natural hand-span; thus leading to functional augmentation of
their skills and the technical feasibility to play with 11 fingers. hand-span, and creating unprecedented complex music that
The robotic finger employs sensors, motors, and a human is impossible to play with 10 fingers.
interfacing algorithm to control its movement in real-time. A
proof of concept validation experiment has been conducted to II. BACKGROUND
show the effectiveness of the robotic finger in playing musical
pieces on a grand piano, showing that naive users were able to Wearable robotics can be seen as a very promising medium
use it for 11 finger play within a few hours. through which the human body and various robotic forms can
be integrated for the purpose of synergistic performance of
I. I NTRODUCTION tasks. Wearable robotics come to place, for the most part, in
the form of replacements for lost limbs, using prosthetics
Human augmentation can be achieved through wearable [13], or as a means to enhance the human body’s force
robotics, these devices enhance human capability by replac- generation mechanics and its ability to carry out precise
ing missing limbs [1], [2] giving more power to existing motions using, for example, exoskeletons [14].
human body parts in the form of exoskeletons [3]–[5] or
A larger body of research exists related to supernumerary
by providing additional limbs [6]–[11]. This latter category
arms as they provide abilities that would be useful in a
of wearable robotics is referred to as supernumerary robotic
number of industrial settings. In [15] two robotic arms that
limbs (SRL). These are devices that attach to the human
are mounted on a user’s body, aiding them in performance
body, giving it more degrees of freedom and thus more
of tasks. The SR arms have actuators which allow them to
capabilities. An additional robotic limbs can lead to unprece-
mimic the rotation and movement of the human arm. They
dented opportunities in augmenting human capability and
are mounted to the waist of the user, so that minimum torque
performance, including reduction of the number of workers
is exerted on the spine. The device can help in balancing
required to do a particular task [7], [9], [11], or improvement
or reduction of weight being handled by the user [9]. A
of the manipulation capabilities of a single hand in grasping
demonstration control algorithm is employed by this design,
an object [10], [12]. Currently most of the research on
a task is repeated a number of times by a worker, the
additional robotic limbs for healthy humans are focused on
identifier data of that task collected, and used to create a
grasping of objects [6], [10], [12] and reducing the workload
dynamic model for how the robotic follower arm should
on industrial workers [7], [9].
react to the user [11]. Because the human body will not be
This paper presents efforts to expand the focus of super- normally stationary, be it through involuntary or voluntary
numerary robotic limbs to the realm of music technology movements, SR systems must be able to adapt in real-time.
and more functional tasks within the augmentation of the As an example, in [9], a discrete Kalman filter is used to
hand. The goal is thus to develop a supernumerary robotic 3rd evaluate error covariance due to motion, and then use a
thumb (SR3T) finger that can augment musicians in playing bracing technique to counteract and suppress user induced
piano, including expanding their reach to larger chords, disturbances. Bracing is carried out through attaching one
playing keys located beyond the natural reach of a single of the two SR limbs to a structure in the environment, for
support. In the setting of collaborating so closely with robots,
Research supported by eNHANCE (http://www.enhance-motion.eu) un-
der the European Union’s Horizon 2020 research and innovation programme the coordination between the human and the SR device is
grant agreement No 644000. key. It has recently been shown that if algorithms can produce
J. Cunninghum, A. Hapsari, P. Guilleminot, A. Shafti and A. A. close mirroring of movements between humans and robots,
Faisal are with the Brain and Behaviour Lab, Dept. of Computing and
Dept. of Bioengineering, Imperial College London, SW7 2AZ, London, the extra limbs can be perceived as part of the user’s body
[email protected] [16]. This would allow the user to interact with the SR
 it allows for creative behavior from the user while using
the SR3T in a typical functional human activity. The SR3T
will thus have two degrees of freedom, allowing it to move
horizontally (to reach a particular piano key) and vertically
(to press the key). The user will control the two degrees
of freedom of the SR3T using movements from other body
parts, namely their thumb for horizontal and their foot for
vertical alignment.

A. Design specifications
To achieve the embodiment goals of the project, the
following criteria are of high importance;
Human integration: The entire device must take a form
and behave in a manner that would seem natural to the user
as an addition to their existing fingers. This includes the
physical appearance of the supernumerary finger, its control
and its reaction to user input. Latency between user input
and the movement of the finger/striking of a key is thus a
Fig. 2: The SR3T presented in this paper, (a) CAD drawing of the critical metric to consider. Therefore, a total processing time
design in SolidWorks, (b) 3D printed and assembled device with threshold, from sensor input to motor actuation, is set at
labeling of the different parts. 80ms. This will allow a time window which will prevent a
peripheral more effectively and with less effort. noticeable delay in hearing a sound/seeing movement [17].
Research efforts have focused more on assistance with Interference: As this supernumerary finger is to be used in
grasping or robotic limbs that help you while you work, with the setting of playing a piano, it is key that the device does
no strong body of work directed at creative applications of not interfere with the user’s natural abilities. This includes
this technology, involving finer movements and force level the manner in which the device is mounted and the way it is
control such as those with fingers. In [10] a robotic aid in the controlled. The robotic finger should not cause limitations to
grasping and handling of objects was developed using . They the movements of the user’s own fingers, with workspaces
have proposed a system that synergizes grasp movements clearly separate from each other. Device weight: The weight
between human fingers and supernumerary robotics. This is of the device is a very important metric in the situation where
done by completing grasps while wearing a high fidelity data delicacy and agility are an essential part of the end-goal
glove which tracks wrist, fingertip and joint movements. In activity. Here, the user must be able to comfortably wear
real-time use, a data glove is employed in order to relay the device for a sustained period, suggested as between 45-
hand movement information to the SR system so that it can 60 minutes, continuously. Looking at pre-existing handheld
react accordingly. Similar to this approach, [8] adds a single devices for guidance on acceptable weight limits, it was
robotic finger to the user’s hand that is positioned on the found that modern Apple smartphones weigh between 148-
underside of the wrist forming a claw with the hand. In this 202 grams (iPhone 8/Plus). As a device that users would
research the robotic device reacts to movements of the user’s bear the weight of for shorter periods of time than a phone,
hand which is mapped, again, using a data glove. The robotic it was aimed to not exceed twice the average weight of a
finger, however, reacts in real-time to finger motions rather smartphone ( 350g). Accuracy: The SR3T must firstly be
than preset configurations. able to fit in the width of a standard piano key so that it can
Research regarding supernumerary robotic fingers deals press a single key accurately, and secondly, must be able to
with the concept purely from a grasping perspective, and move with a level of accuracy in rotation that adjacent black
does not visit the need for applying forces, and variable and white keys can be pressed in succession. The standard
force exertion in functional tasks. This is limiting when width of a white key on a piano is 23.5mm with a black
considering integration into uses other than picking up ob- key width of 13.7mm . Torque: The key of a standard piano
jects. Furthermore, the dexterity with which these systems requires 0.5 N of force acting on its face to press it. The
can operate is limited and needs to be expanded to closer motors used in the design should allow adequate torque to
resemble human movement. be generated such that this force requirement is met, and
be adjustable as needed for different volume levels to be
III. M ETHODOLOGY achieved.
We propose a supernumerary robotic 3rd thumb (SR3T)
which can augment the functionality of the human hand in B. Mechanical implementation
dexterous tasks such as the playing of a piano, and to serve as The SR3T consists of 2 degrees of freedom, with a
a test rig for human augmentation through wearable robotics, finger-shaped body, moving vertically and horizontally as
allowing research into questions on embodiment and learn- commanded by the user. Movements are performed by
ability of their use. A piano playing task is considered, as two 9V DC motors from Maxon Motors (Switzerland),
customized with a planetary gearhead with reduction ratio
16:1, increasing the output torque 16 times higher than the
nominal, by decreasing the velocity of the motor. The motors
are equipped with the ENX10 EASY 256IMP encoders
to read position and velocity. This is then used with the
EPOS2 24/2 DC motor controllers, allowing control of motor
current, position and velocity. These motor controllers also
include analogue inputs, which are made use of for reading
movement commands from the user.
The main motivation when designing the shape of the
finger attachment was that the user could naturally relate
to its form. A large part of attempting to add to the basic
human form is the user feeling that they may ”embody” the
robotic device that they are interacting with [17]. Following
this idea, different design ideas were considered regarding
the size, shape and mechanical mechanisms that would be
employed in creating the SR3T. From a functional stand
point, the design must be able to carry out the task of pressing
both white and black keys interchangeably, as the user sees Fig. 3: Block diagram for the complete SR3T system.
fit. The finger design employed resembles that of a fixed
human finger. It consists of a knuckle, around which rotation by typical mechanical engineering criteria, ensuring proper
takes place, a horizontal section, similar to a human proximal mating of relevant parts and translating the rotations of the
phalange, and a hinged section, which represents a combined DC motors into the movements of the robotic finger. All
medial and distal phalange. The robotic finger was modeled necessary parts were designed in SolidWorks, 3D printed
on an average adult male’s middle finger dimensions. The using PLA material, and assembled together.
horizontal section has a length of 58mm, which joins the The complete system can be seen in Figure 2. The horizon-
tip section of length 48.5mm, at an angle of 60 degrees. tal motor is free to create a rotation of the finger at over 360
The knuckle section adds an additional 41mm of length to degrees, which is far beyond the needs of the piano player.
the finger, so that the positioning difference, back on the The vertical motor is also free to create rotations up to about
hypothenar eminence and the ulnar aspect of the hand, is 120 degrees, allowing for the user to press piano keyboards
offset. This allows the SR3T to be in line with the player’s with different levels of pressure with precision. The linear
fingers. The player is then able to decide whether black or drive section of the wrist support contains two aluminum
white keys are targeted through moving the whole hand in couplings, which were manufactured to specification using
or out depth wise, on the piano keys. This is something that universal drives, and a plastic linear drive carriage and
players augment inherently and so it is natural for them to aluminum carriage rail to keep the different sections in place.
implement in this scenario.
The design mounts flush to the hypothenar eminence, C. Sensors and human interfacing
allowing the user to move their wrist freely and minimizing The SR3T’s two degrees of freedom are controlled through
the distance between the device and the keys of the piano. two movements from the human user. The horizontal move-
The top of this mounting system is designed to sit over ment is picked up from the user’s thumb on the hand being
the back of the user’s hand, with the large curvature on augmented. The vertical movement is picked up from the
top and the smaller lip below following the curvature of lifting movement of the user’s left foot (dorsiflexion). The
a human hand for better fit and greater comfort. This in left foot is chosen due to the right foot being of use in
turn helps reduce any turning moments on the player’s hand, pressing the pedals of the piano.
generated by the weight of the device. The device sits across To sense the flexion of the thumb and control the hori-
a 5cm long area, on the length of the hypothenar eminence. zontal movement of the SR3T, a generic, off-the-shelf flex
This dimension was used so that when the user attempts sensor is used. This is a variable resistor, which changes
to press a piano key down, the finger mount is not pushed its resistance based on the curvature applied to it during
up by the resistance. For vertical rotation of the SR3T, a flexion. When flat, the resistance value was measured at
rotating top plate was designed. It was very important that the around 13 kΩ and when at 180 degree pinch bending, the
space taken up by this part, both horizontally and vertically, resistance value was measured at around 26 kΩ. A voltage
was minimized to achieve an overall design that kept the divider and unity buffer setup is used to interrogate the flex
SR3T as close to the hand height as possible, while not sensor, translating the changes in sensor resistance into buffer
adding a device with a cumbersome width. The rest of the output voltage changes. This is then connected to the analog
device was recessed at different heights below this to ensure input of the EPOS2 24/2 motor controller for the horizontal
that the SR3T would eventually sit as close as possible to movement. The flex sensor is placed on the thumb of a golf
the user’s fingers. The remainder of the design was driven glove. The golf glove is made of leather material which
provides flexibility and comfort to move the natural fingers. and minimum respectively. The highest vertical position
It also ensures that the flex sensor stays in place as the thumb is chosen as the position when the robotic finger hovers
moves, yet bends with respect to the motion of the thumb. above the piano notes while the lowest vertical position is
To sense the dorsiflexion of the foot, the ADXL335 (Ana- chosen at the height at which the SR3T has fully pressed a
log Devices, Massachusetts, USA) analog accelerometer is piano key. Y-axis acceleration data determines the position
used. The device is mounted on the user’s foot and its outputs of the foot above the ground. This data corresponds to the
connected directly to the analog inputs of the EPOS2 24/2 direction of the SR3T’s vertical motion. Z-axis acceleration
motor controller for the vertical movement. The values for data describes the instantaneous acceleration of the foot
two of the accelerometer axes are used. The Y axis provides when it starts moving. This data is related to the velocity of
information on the direction of the vertical movement and the the vertical motion of the SR3T. The Y-axis maximum and
Z axis provides information on the velocity of the movement. minimum parameters are saved by moving the foot up and by
An arm mount for all the required printed circuit boards placing the foot on the ground respectively. Z-axis maximum
(PCBs) was designed and 3D printed, in order to make parameter is the acceleration data when the foot is being
the system as portable as possible. The mount houses the lifted and Z-axis minimum parameter is the acceleration data
two EPOS2 24/2 motor controllers and a smaller PCB for when the foot is stationary.
handling of sensor outputs; refer to Figure 5.a. The calibration processes above maps the sensor input and
D. Control software motor outputs in a linear fashion, allowing the user to control
The control of the robotic system runs under a software the position and velocity of the SR3T’s movement, in the two
algorithm that was built using C# code in Visual Studio. defined degrees of freedom.
The aim of the algorithm is to process data from sensors and
IV. E XPERIMENTS AND D ISCUSSION
convert them into robotic finger motion command. In order to
speed up the computation time, the vertical motion algorithm A. Robot design
and the horizontal motion algorithm run on separate threads Experiments were performed to evlauate the SR3T’s
in the C# code. This method allows the calculation for workspace. An optical motion capture system, the OptiTrack
vertical motions and the horizontal motions to run in parallel. Prime 13W (NaturalPoint, Inc. DBA OptiTrack, Oregon,
The sensors are connected to the motor controllers, the USA), was used to capture the 3D motion of the thumb and
PC reads sensor data via the analog inputs of the motor the SR3T in real time. The cameras captured the change
controllers, computes all input information from sensors and in body position provided by the specific optical markers
sends control commands back to the motor controllers. The attached. During the experiment, three OptiTrack cameras
motor controllers are connected to the PC via USB, refer to were used to track the motion of the thumb in x, y, and z
Figure 3. The software presents a graphical user interface planes. Before the motion tracking experiment started, the
(GUI) for the user to configure and calibrate the SR3T as OptiTrack cameras were calibrated to ensure the accuracy
needed. and precision of the recorded data positions. Markers were
When the motor controller is enabled, the position value placed at the base and tip of the SR3T, to measure its max-
from the encoder will be restarted to zero. Consequently, imum workspace, sweeping the SR3T along its maximum
the position of the SR3T at the beginning of the experiment range of movement. The same experiment repeated, with
determines the start of the position counter of the encoder. the subject moving their thumb along its maximum range of
Using this method, the motion control calculations remain movement, and its motion tracked. Results are visualised in
unaffected by the starting position of the device. In the Figure 4, with the angular end-point surface mapped onto an
horizontal movement calibration, the furthest piano note from sphere. Overall, the SR3T has an angular end-point surface
the user’s hand that the SR3T can reach will be selected 4 times that of the human thumb, effectively augmenting
as the maximum, and the closest piano note to the user’s human reach and capability.
hand that the SR3T may press, as the minimum. In the Using the same setup, delay experiments were also per-
horizontal motion calibration, the analog flex sensor values formed, to measure the delay between the command given by
that correspond to the maximum and minimum flexion of the user and the respective motion by the SR3T, i.e. thumb
the user’s thumb are also recorded. The vertical and hori- extension/flexion for horizontal and foot lift and drop for the
zontal motion parameters are saved by moving the robotic vertical motions. Optical markers were placed on the SR3T,
finger to the preferred positions and triggering a button the user’s thumb and their foot. Results showed a mean delay
on the software GUI. The analog maximum and minimum between motor intention and motor action of ≈ 85msec.
parameters for the flex sensor are saved by straightening and
flexing the thumb to the preferred posture. The velocity of B. Human augmentation
the horizontal movement is defined through a proportional Our proof-of-concept test is designed to evaluate a naive
control approach, i.e. a larger distance from the SR3T’s users performance when using the SR3T in a functional
initial position to its target position, will result in a larger manner and to explore whether the SR3T is used so as to
velocity of motion. substitute the thumb on the ipsilateral hand (i.e. using the
In the vertical motion calibration, the highest vertical right hand as if it were a left hand) or whether the user
position and the lowest vertical position are set as maximum can embrace the ability to operate a hand with six fingers.
To this end we focused on the task of piano playing which
allows straightforward dynamic mapping and remapping of
fingers to skilled task performance. In music, the choice of
which fingers and hand positions to use when playing certain
musical instruments is called fingering. Fingering typically is
dynamically adjusted throughout a musical piece by skilled
performers so as so optimally use the available fingers. The
challenge of choosing a good fingering is to make the hand
movements as comfortable as possible in the flow of the
music without changing hand position too often.
The test aims to evaluate whether an end-user would
adapt and learn to use the device as an extension to their
own fingers. A musician who had no previous interaction Fig. 4: Angular end-point workspace of the SR3T compared to that
of a human thumb, based on data collected through optical motion
with the SR3T was asked to play piano with the device trackers, and mapped onto unit sphere.
mounted on their forearm. The system control was explained
to them and, over a period of two hours, the musician was
allowed to improvise and try to use the finger in different the dynamics of how hard or soft they press piano keys with
musical contexts. At the beginning of the test, the user was the SR3T. This was achieved by lifting their foot, which
struggling to control the SR3T to press even a single note, had the accelerometer attached to it, at different speeds,
while doing nothing else with either hand, due to having to which were reflected on the SR3T’s movement. Accuracy:
focus on moving their thumb and foot in correct coordination The musician was able to play both black and white keys
to control the SR3T. Over the course of the two hours, the accurately using the finger. The finger was also capable
musician went from this stage, to being able to use the SR3T of moving between adjacent keys in succession. The user
to play melodies and chords, while using all of their other pressing a black key can be seen in Figure 5.a. Comfort:
natural fingers simultaneously. The pianist demonstrated the The test went on for over two hours, with the pianist wearing
ability to use the added finger, and the extra reach it provided, the finger and attachments at all times. When asked during,
in a musical fashion, adding higher notes into the chords of and after the test, the user reported that they suffered little
the left hand and playing additional notes with melodies. to no discomfort throughout. Delay: The user was able to
In the first few trials it was observed that the SR3T presses play in time with the additional finger, after the earlier stage
the piano notes at considerable delay. But as the experiment of practice and learning. The initial delay is attributable to
went on, the delay decreased. This result suggests that the the user learning to rely on the movements of their foot to
delays occurred as the pianist hesitated, focusing and move play a key, rather than due to delays in the robotic system.
their foot to press the piano note using the SR3T. After a Interference: Although the user initially had to think a lot
while, the cognitive load decreases and the pianist is able to about how to position their hand, the device did not interfere
move their thumb and foot naturally to control the robotic with their natural ability to play the instrument after the
finger in real time. The results of the experiment confirmed initial adjustment to its presence and was used seamlessly
our proposal that using the SR3T, a pianist is able to reach afterwards.
piano notes beyond their natural hand-span. The SR3T was
able to press as far as 4 whole notes beyond the hand-span
of the pianist. The results also showed that the pianist was
able to play improvised musical pieces with 11 fingers (10
natural fingers and 1 robotic finger). The pianist was also
able to press the black piano key notes using the SR3T,
although it was more difficult to press those compared to
the white piano note keys. Range Increase: The main aim
was to increase the player’s range. During the course of the
test it was found that the SR3T could play notes, from a
single semitone to three and a half semitones to the right
of the player’s pinkie finger. This is in effect a four whole-
note increase in range to the musician’s natural hand span.
This can be seen in Figure 5.b, where the player uses their
thumb to play, whilst controlling the SR3T. Note that the
SR3T extends the reach while still using the 5 fingers of
the hand, adding an extra note to the reach of the user.
If the hand is fully extended already, however, the extent
Fig. 5: The test subject playing piano using the SR3T. (a) view of
of note reach augmentation will be reduced, due to the the full system mounted on the user’s arm as they play piano, (b)
mounting style. Dynamics: The pianist was able to control user’s point of view; note the 4-note increase in reach using the
SR3T.
 V. D ISCUSSION & C ONCLUSIONS [4] A. M. Dollar and H. Herr, “Lower extremity exoskeletons and ac-
This paper presents the development of a supernumerary tive orthoses: challenges and state-of-the-art,” IEEE Transactions on
robotics, vol. 24, no. 1, pp. 144–158, 2008.
robotic 3rd thumb for the purpose of functional human [5] J. F. Veneman, R. Kruidhof, E. E. G. Hekman, R. Ekkelenkamp,
augmentation in a skilled motor task: piano playing. The E. H. F. Van Asseldonk, and H. Van Der Kooij, “Design and evaluation
mechanical structure, electronic components and control of the LOPES exoskeleton robot for interactive gait rehabilitation,”
IEEE Transactions on Neural Systems and Rehabilitation Engineering,
strategy to be used in the execution of the SR3T were vol. 15, no. 3, pp. 379–386, 2007.
described in detail. The SR3T’s angular end-point surface [6] I. Hussain, G. Salvietti, M. Malvezzi, and D. Prattichizzo, “Design
was 4 times that of a human thumb, and mean motor guidelines for a wearable robotic extra-finger,” in Research and
Technologies for Society and Industry Leveraging a better tomorrow
intention to action delay was measured as ≈ 85msec. A (RTSI), 2015 IEEE 1st International Forum on, pp. 54–60, IEEE, 2015.
functional user test with a naive pianist showed that the [7] B. Llorens-Bonilla and H. H. Asada, “A robot on the shoulder:
SR3T was in fact effective, and that the user could adapt Coordinated human-wearable robot control using coloured petri nets
and partial least squares predictions,” in Robotics and Automation
to incorporate it into their natural playing of the musical (ICRA), 2014 IEEE International Conference on, pp. 119–125, IEEE,
instrument. The piano player transitioned from an initial use 2014.
of the SR3T on the right hand as if it were a left hand, using [8] D. Prattichizzo, M. Malvezzi, I. Hussain, and G. Salvietti, “The sixth-
finger: a modular extra-finger to enhance human hand capabilities,” in
5-finger based fingering, and managed after 1 hour to use 6- Robot and Human Interactive Communication, 2014 RO-MAN: The
finger based fingering, i.e. using the right hand as a truly 23rd IEEE International Symposium on, pp. 993–998, IEEE, 2014.
augmented hand. [9] F. Parietti, K. Chan, and H. H. Asada, “Bracing the human body with
supernumerary robotic limbs for physical assistance and load reduc-
Given the results of our preliminary functionality test we tion,” in Robotics and Automation (ICRA), 2014 IEEE International
conclude that there is a positive outlook for future work. Conference on, pp. 141–148, IEEE, 2014.
The user demonstrated the ability to learn and control the [10] F. Y. Wu and H. H. Asada, “Bio-artificial synergies for grasp posture
control of supernumerary robotic fingers,” MIT Press, 2014.
SR3T quickly, and was able to produce musical results in [11] B. Llorens-Bonilla, F. Parietti, and H. H. Asada, “Demonstration-
hours which is little time compared to rehabilitation times for based control of supernumerary robotic limbs,” in Intelligent Robots
bionic prosthetics users of months. Furthermore, exploring and Systems (IROS), 2012 IEEE/RSJ International Conference on,
pp. 3936–3942, IEEE, 2012.
the idea of low-dimensional control synergies , such as those [12] I. Hussain, G. Salvietti, L. Meli, C. Pacchierotti, D. Cioncoloni,
employed in [10], [18], might result in a better performance S. Rossi, and D. Prattichizzo, “Using the robotic sixth finger and
of the robotic finger in action. While our human-robot inter- vibrotactile feedback for grasp compensation in chronic stroke pa-
tients,” in Rehabilitation Robotics (ICORR), 2015 IEEE International
face is substitutive in nature (using other limb movements Conference on, pp. 67–72, IEEE, 2015.
to control artificial movements[19], [20]), true augmentation [13] L. Zollo, S. Roccella, E. Guglielmelli, M. C. Carrozza, and P. Dario,
may require more sophisticated human-robot interfaces that “Biomechatronic design and control of an anthropomorphic artificial
hand for prosthetic and robotic applications,” IEEE/ASME Transac-
are not just neurobionic [21] but cognitive in nature [22]. tions On Mechatronics, vol. 12, no. 4, pp. 418–429, 2007.
The SR3T setup provides a unique test rig for experiments [14] A. B. Zoss, H. Kazerooni, and A. Chu, “Biomechanical design of
on learn-ability and embodiment potential of supernumerary the Berkeley lower extremity exoskeleton (BLEEX),” IEEE/ASME
Transactions On Mechatronics, vol. 11, no. 2, pp. 128–138, 2006.
robotic limbs and augmentative devices. The piano applica- [15] C. Davenport, F. Parietti, and H. H. Asada, “Design and biomechanical
tion presents a creative environment where the user is free analysis of supernumerary robotic limbs,” in ASME 2012 5th annual
to improvise and learn along the way how to (or not to) dynamic systems and control conference joint with the JSME 2012 11th
motion and vibration conference, pp. 787–793, American Society of
incorporate the SR3T into their work, while focusing on the Mechanical Engineers, 2012.
control of two extra degrees of freedom.Future work will [16] A. Guterstam, V. I. Petkova, and H. H. Ehrsson, “The illusion of
focus on improving the setup and running extended exper- owning a third arm,” PloS one, vol. 6, no. 2, p. e17208, 2011.
[17] T. R. Makin, F. De Vignemont, and A. A. Faisal, “Neurocognitive
iments to better understand the implications and limitations barriers to the embodiment of technology,” Nature Biomedical Engi-
of robotic augmentation. Our work here can be considered neering, vol. 1, no. 1, pp. 1–3, 2017.
an ”existence proof” that naive, but hand-craft skilled users [18] C. Konnaris, A. A. Thomik, and A. A. Faisal, “Sparse eigenmotions
derived from daily life kinematics implemented on a dextrous robotic
can indeed learn in under a few hours to perform highly hand,” in Biomedical Robotics and Biomechatronics (BioRob), 2016
dexterous tasks such as piano playing, suggesting that our 6th IEEE Intl. Conf. on, pp. 1358–1363, IEEE, 2016.
brain has sufficient plasticity and capacity to support true [19] A. A. Thomik, D. Haber, and A. A. Faisal, “Real-time movement
prediction for improved control of neuroprosthetic devices,” in Neural
human augmentation through supernumerary robotic limbs. Engineering (NER), 2013 6th Intl. IEEE/EMBS Conf. on, pp. 625–628,
R EFERENCES IEEE, 2013.
[20] J. J. Belić and A. A. Faisal, “Decoding of human hand actions to
[1] E. J. Rouse, L. M. Mooney, E. C. Martinez-Villalpando, and H. M. handle missing limbs in neuroprosthetics,” Frontiers in computational
Herr, “Clutchable series-elastic actuator: Design of a robotic knee neuroscience, vol. 9, p. 27, 2015.
prosthesis for minimum energy consumption,” in Rehabilitation [21] S. Fara, C. Gavriel, C. S. Vikram, and A. A. Faisal, “Prediction of
Robotics (ICORR), 2013 IEEE International Conference on, pp. 1– arm end-point force using multi-channel MMG,” Proceedings - 11th
6, IEEE, 2013. International Conference on Wearable and Implantable Body Sensor
[2] M. Xiloyannis, C. Gavriel, A. A. Thomik, and A. A. Faisal, “Gaus- Networks, BSN 2014, pp. 27–32, 2014.
sian process autoregression for simultaneous proportional multi-modal [22] R. O. Maimon-Dror, J. Fernandez-Quesada, G. A. Zito, C. Konnaris,
prosthetic control with natural hand kinematics,” IEEE Transactions S. Dziemian, and A. A. Faisal, “Towards free 3d end-point control
on Neural Systems and Rehabilitation Engineering, vol. 25, no. 10, for robotic-assisted human reaching using binocular eye tracking,” in
pp. 1785–1801, 2017. Rehabilitation Robotics (ICORR), 2017 Intl. Conf. on, pp. 1049–1054,
[3] R. A. R. C. Gopura, K. Kiguchi, and Y. Li, “SUEFUL-7: A 7DOF IEEE, 2017.
upper-limb exoskeleton robot with muscle-model-oriented EMG-
based control,” in Intelligent Robots and Systems, 2009. IROS 2009.
IEEE/RSJ International Conference on, pp. 1126–1131, IEEE, 2009.