1

Simple and Fast Compensation of sEMG Interface

Rotation for Robust Hand Motion Recognition
Minjae Kim, Keehoon Kim, and Wan Kyun Chung

Ch 3 Ch 1 Ch 8
Abstract—Surface electromyography (sEMG) measurements Ch 2
Ch 4
have demonstrated the potential to recognize complex hand Ch 7
Ch 2
motions. Additionally, sEMG enables natural recognition without Ch 5 Ch 1 Rotation
disturbing movements, and thus, can be used in various fields Ch 6
Sources Initial Sources Rotated
Ch 3
such as teleoperation, assistant robots, and prosthetic hands. Ch 6 Ch 8 electrode electrode
placement Ch 4 Ch 5 placement
However, sEMG signals highly depend on electrode placements Ch 7

due to the complex muscle structures. A shift of the electrode

can lead to inconsistent signal measurement. Thus, sEMG-based
recognition is not practical for applications that require long- Signals after
Signals rotation Recognized motion
term and repeated usage. This paper proposes compensation of Motion
recognizer
sEMG interface rotation for robust motion recognition. Once
Motions for training
the relationship between sEMG signals and motions is trained, Motion
recognizer
additional training for different electrode configuration is not
necessary for a band-type interface. The proposed process is Compensated
signals
simple and fast. The interface rotation can be compensated for by
performing only a single motion for approximately two seconds.
Fig. 1. Overview of motion recognition with compensation.
The single motion for compensation is dependent on the muscle
properties of the user. Generally, ulnar deviation may work.
To demonstrate the proposed compensation, recognition of five
hand motions is conducted. The experimental results indicate the skin surface as sEMG signals. sEMG signals represent
that the proposed compensation can cover the overall range of muscle activations for each motion. Thus, these signals contain
rotation. Additionally, the proposed compensation is validated information related to motions and their intention.
with a transradial amputee.
sEMG electrodes are generally placed on the forearm be-
cause many finger motion-related muscles are located in the
I. I NTRODUCTION forearm. Therefore, sEMG measurement facilitates natural mo-

H AND gestures and grasping motions are widely used in

everyday activities. Therefore, the recognition of hand
motions has many potential applications, including teleopera-
tions [1], [2] and assistant robotics [3], [4].
Various hand motion recognition methods that use vision
[5], [6] or strain sensors [7], [8] have been proposed. However,
these vision- and strain sensor-based methods have limitations.
Vision-based methods are weak with respect to occlusion
and involve a limited workspace. Strain sensor-based methods
hinder natural hand movement and interactions with objects.
One of the well-known techniques for hand motion recog-
nition is surface electromyography (sEMG) signal measure-
ment [9]–[11]. A muscle contraction is triggered by action
potentials, which are induced by motor neurons and propa-
gated along the muscle fibers; they can be measured from
This work was supported in part by the National Research Foundation of
Korea grant funded by the Korea government (MSIT) (No. 2011-0030075),
and in part by the Global Frontier Research and Development Program on
Human-Centered Interaction for Coexistence funded by the National Research
Foundation of Korea grant funded by the Korean Government under Grant
2012M3A6A3056423. (Corresponding author: Wan Kyun Chung.)
M. Kim was with the Department of Mechanical Engineering, Pohang
University of Science and Technology, Pohang 37673, South Korea.
M. Kim and K. Kim are with the Korea Institute of Science and Technol-
ogy, Seoul 02792, South Korea (e-mail: minjaekim@kist.re.kr;
khk@kist.re.kr)
W. K. Chung is with the Department of Mechanical Engineering, Pohang
University of Science and Technology, Pohang 37673, South Korea (e-mail:
wkchung@postech.ac.kr).
tion recognition without disturbing movements. Additionally,
it is possible to recognize motion intentions for amputees who
cannot perform functions using their hands. Therefore, various
sEMG-based prosthetic control systems [12]–[14] have been
proposed.
However, there are significant drawbacks to sEMG-based
motion recognition. Hand motion-related muscles have com-
plex structures and are located in a small region of the
forearm. Therefore, it is difficult to extract a particular muscle
activity. Additionally, the sEMG signal is highly dependent on
electrode parameters [15]–[17] including electrode placement,
inter-electrode distance, and electrode size. Therefore, even
a short electrode displacement can significantly increase the
recognition error [18]. Although increasing the size of an
electrode makes recognition more robust against electrode
displacement [19], a large electrode decreases the motion
selectivity by reducing the spatial resolution.
For this reason, conventional sEMG-based motion recog-
nition methods are not practical for applications that require
long-term and repeated usage because electrode displacement
frequently occurs during donning and doffing. In the case
of prosthetic hands, users require an operation time of over
8 h [20]. Moreover, during use, the electrode can easily slip.
To maintain reliable recognition performance, the relationship
between the motions and sEMG signals should be trained
every time the user wears an sEMG interface.

Rotation
Source
Initial
electrode
Known parameters placement
Unknown parameters
Identified parameters
Source
parameter
identification
sEMG signals
Electrode parameters Electrode parameters
Source parameters
Fig. 2. Scheme for proposed model-based rotation compensation.
Several methods have been proposed for robust motion
recognition against electrode displacement. Methods of en-
suring robustness by analyzing high-density sEMG signals
[21], [22] have been proposed. However, these methods require
many sEMG channels. Additionally, Muceli et al. [23] demon-
strated that activation signals extracted using non-negative
matrix factorization (NMF) [24] are robust against electrode
displacement.
However, these conventional methods were demonstrated
for small displacements of only about 1 or 2 cm. Although
in the case of a prosthetic hand, compensation of this level of
displacement might be sufficient due to a user-specific socket
[19], the user should carefully wear the interface for consistent
performance. In the case of a teleoperation, the effect of
electrode displacement is dominant.
In this study, simple and fast rotation compensation for
a wide range is proposed. The overview of the motion
recognition with compensation is described in Fig. 1. This
research is based on our previous work on sEMG signal
decomposition as muscle activation sources [25]. We assumed
that the contraction of a muscle involves activation sources.
The sEMG signal measured from an electrode is a combina-
tion of source activations attenuated by the conductivity of
each source and the distance between each source and the
electrode. If the electrode parameters, including the three-
dimensional placement, size, and shape, are known, then the
source parameters, including the three-dimensional placement,
conductivity, and source amplitude, can be identified using the
measured root-mean-square (RMS) sEMG signals.
In our previous study, rotation of the sEMG interface
was recognized by comparing the source positions identified
before and after rotation. However, this is time-consuming and
impractical.
In this study, our proposed method is improved for simple
and fast compensation. If either of the electrode parameters
or source parameters are known, the remaining parameters
can be identified using sEMG signals. Electrode parameters
are known before the interface rotation; thus, the source pa-
rameters can be identified. Theoretically, even if the electrode
2
Initial configuration
Compensation
Source Source
Rotated Compensated
electrode electrode
placement placement
Electrode
sEMG signal
parameter
compensation
identification
sEMG signals sEMG signals
Compensated
Electrode parameters
sEMG signals
Source parameters Source parameters
parameters are changed according to the interface rotation,
the source parameters remain unchanged because they depend
on anatomical factors. Therefore, once source parameters
are identified before rotation, the electrode parameters after
rotation can be identified via a simple procedure. The scheme
for the model-based rotation compensation is shown in Fig. 2.
Heuristically, it is sufficient to recognize the rotation by per-
forming only a single motion for approximately two seconds.
The motion for compensation is dependent on the anatomical
factors of the user, which is selected as the ulnar deviation for
able-bodied subjects in this study.
After rotation, when the electrode parameters are identified,
the sEMG signals can be converted into those with respect to
the initial electrode parameters. Therefore, this compensation
enables sEMG-based motion recognition without additional
training to determine the relationship between motions and
sEMG signals after rotation.
Additionally, the compensation process is separated from
the motion recognition process. Therefore, any recognition
method, such as linear discriminant analysis classifier [26],
support vector machine classifier [27], or deep learning [28],
can be integrated.
In conclusion, the proposed method provides robust sEMG-
based motion recognition, which is useful in assistant robots
and prosthetic hands that require long-term and repeated usage
in everyday activities. The users do not require additional
motion training each time they put on and take off the
interface.
The remainder of the paper describes the details of the
model and compensation method. Then, hand motion recog-
nition will be described to demonstrate the rotation compen-
sation.
II. M ETHODS
A. Subjects
Ten able-bodied subjects and one transradial amputee partic-
ipated in the study. The conducted experiments were approved
by the institutional review board (PIRB-2018-E079) of the

Pohang University of Science and Technology (POSTECH),
Pohang, Korea.
B. Experimental Setup and Data Acquisition
For the able-bodied subjects, the motions selected included
wrist extension/flexion, hand close/open, and ulnar deviation.
A transradial amputee has a very limited muscle contraction
pattern, and thus, was asked to contract muscles while intend-
ing four motions (wrist extension/flexion, ulnar deviation, and
hand close).
An MYO armband (Thalmic Labs Inc., Kitchener, Canada)
was used to measure the sEMG signals from the forearm. The
interface consists of eight bipolar channels, which are arranged
at constant intervals. Each channel consists of two active
electrodes and one reference electrode between the active
electrodes. Additionally, an inertial measurement unit (IMU)
was built into the MYO armband. sEMG signals and IMU
signals were simultaneously measured at sampling frequencies
of 200 Hz and 50 Hz, respectively.
C. Experimental Procedures and Signal Analysis
The subjects were comfortably seated on a chair and in-
structed to perform muscle contractions following a specific
motion displayed on a monitor. For each motion, the subjects
were asked to maintain the pose with moderate and constant
force according to their convenience.
Each motion was contracted for 3 s and repeated three times.
Between contractions, the subjects rested for at least 1 s. In
the first contraction, the middle two seconds of contractions
were used for model identification, reconstruction, and motion
training. The remaining contractions were used for the tests.
After all motions were performed, the subjects removed the
interface, rotated it, and put it back on again.
The non-negative sEMG signals are required to identify
the model parameters and to compensate for the rotation.
Therefore, both parameter identification and motion training
were based on the RMS sEMG signals. The signals were
extracted from a window of 200 ms every 100 ms. From
the contraction of 2 s, 19 samples for each motion can be
extracted. A large number of samples might provide a more
convincing performance; however, the number of samples
used in this study might guarantee at least the minimum
performance.
The relationship between the RMS sEMG signal and motion
was trained via a shallow neural network that contained 10
neurons. The network was trained using the scaled conjugate
gradient backpropagation method by using MATLAB 2017a
(MathWorks, Inc., Natick, MA, USA).
To investigate statistical significance, a one-sample and
paired-sample t-test, and a two-sample t-test were performed
according to the data. All analyses were performed using
MATLAB 2017a.
D. Model Configuration
Generally, an sEMG channel consists as a bipolar configu-
ration in which the sEMG signal represents the difference in
3
signals measured at the two active electrodes relative to that
of the reference electrode. Active electrodes are placed on the
skin surface near the target muscle, and a common reference
electrode is placed on the bony region or between the two
active electrodes.
The proposed model assumes that the signal measured
from each electrode is a summation of the attenuated source
amplitudes. The corresponding relationship is expressed as
follows:
X
Vit = dit k Vsk ,
k
i = 1, ..., m, k = 1, ...n, t = 1, 2, r (1)
where Vit denotes the summation of the attenuated source
amplitudes measured at electrode t consisting the ith channel;
t denotes the index of the electrodes, where 1 and 2 denote
two active electrodes and r denotes the reference electrode; m
denotes the number of overall channels; n denotes the number
of sources; Vsk denotes the kth non-negative source amplitude;
and dit k denotes the function value that decreases with the
conductivities and distance between the source position and
electrode position as follows:
s
(xit − xk )2 (yit − yk )2 (zi − zk )2
dit k = 1/ + + t (2)
σxk σyk σzk
where xit , yit , and zit denote the three-dimensional po-
sition of the electrode, σxk , σyk , and σzk denote the three-
dimensional conductivity of the kth source, and xk , yk , and
zk denote the three-dimensional position of the kth source.
For the bipolar configuration of the sEMG channel, the
measured sEMG signal can be represented as follows:
X
Vi = ||di1 k − dir k | − |di2 k − dir k ||Vsk
k
X
= Dik Vsk (3)
k
where Vi denotes the RMS sEMG signal measured from the
ith channel.
The electrode parameters, including the electrode placement
and the channel configuration, are known according to the
sEMG interface specification. Therefore, the source parameters
for the motion can be identified using the sEMG signals
measured during static muscle contractions.
The source parameters for each motion were independently
identified. There were 19 RMS sEMG signal samples from a
static contraction of 2 s, 8 equations for a number of channels,
and 7 unknown source parameters (x, y, z, σx , σy , σz , Vs ) per
each source; all 19 samples were used to find best-fit pa-
rameters. After identification, the first 6 source parameters
were used for further analysis, and the last parameter, Vs ,
only represented the identified source amplitude during the
model identification. To guarantee a unique solution, a number
of channels should be larger than the unknown parameters.
To identify the parameters, the lsqcurvefit function in the
MATLAB 2017a Optimization Toolbox was used.
 4

Identification of model parameters Extraction of source amplitudes

on the position. For all measurements, the interface cannot be
Measured sEMG signals, 𝐕𝐌 Measured sEMG signals, 𝐕𝐌
guaranteed to be placed at the same position; thus, a fixed
diameter of 90 mm, which is the largest value among the
subjects, was used.
Additionally, skin deformation due to the pressing of the
Motion 1 Motion 2 Motion 3
electrodes or muscle contraction was not considered. Then, the
𝑉 𝐷𝑉 𝑉 𝐷 𝑉 𝑉 𝐷 𝑉
eight sEMG channels were placed on the cylindrical surface
Source extraction
at every 45◦ .
Extracted source amplitudes, 𝐕𝐒

Source 2 F. Source Configuration

The source parameters include the three-dimensional source
Source 1
positions, directional conductivities, and the number of sources
Source 3
comprising each motion. The MYO armband has a longitudi-
nal limitation. In this configuration, heuristically, fixed x and
z directional conductivities and varying y directional conduc-
Fig. 3. Scheme for source parameter extraction. Source parameters for tivity provided better performance than the other conditions.
each motion were separately identified using sEMG signals during static
0 , were obtained
contractions. Afterward, the overall source amplitudes, VS
Therefore, the x and z directional conductivities were set to
by solving the sEMG signals, as described in (5). a constant of 1, and only the y directional conductivity was
considered as a variable.
Experimentally, a single source for each motion was suffi-
The scheme for the source parameter extraction is shown cient, as described in our previous paper [25]. However, several
in Fig. 3. After the model parameter identification, when motions such as the closing of the hand exhibit complex
new sEMG signals are measured, the relationship between the patterns that are difficult to be represented using a single
sEMG signals and source amplitudes can be expressed via the source. For these motions, the use of two sources for a single
following equation: motion exhibits improved signal reconstruction and rotation
compensation. To identify the unique source amplitudes, the
V10
 
  0  overall number of sources should be the same or lower than the
 V20  D11 ··· D1n Vs1
   .. ..   ..  number of sEMG channels. Otherwise, multiple solutions are
 ..  =  . ..
 .  . .  .  possible. Therefore, careful selection of the number of sources
Dm1 ··· Dmn Vs0n is required.
Vm0
0
VM = DVS0 (4)
G. Rotation Recognition
where each element of D is an identified function value In the initial electrode configuration, the relationship be-
0
according to (3), VM represents the newly measured RMS tween the measured sEMG signals and source amplitudes is
sEMG signals, and VS0 is the unknown non-negative source expressed as follows:
amplitudes that can be estimated by solving the following
equation:
0
VM = DVS0
1 = f (PS , CS , PM )VS0 (7)
VS0 = minVS ||DVS − VM 0
||22 , subject to VS ≥ 0 (5)
2 where VM 0
represents the measured sEMG signals, D
In this study, NMF was applied to obtain VS0 . represents the identified source parameters, VS0 represents the
unknown source amplitudes that can be estimated by solving
E. Forearm and Electrode Configuration (5), PS denotes the source positions, CS denotes the source
The electrode occupies a specific area; therefore, dit k was conductivities, and PM denotes the electrode positions.
considered as an average of several values that include the The relationship between the sEMG signals and source
electrode boundaries and the center as follows: amplitudes after rotation is expressed as follows:

∗ ∗
)VS0
PNp
p dpit k VM = f (PS , CS , PM
dit k = (6)
Np = f (PS , CS , T PM )VS0 (8)
where dpit k denotes the function value between the kth where VM ∗
and PM∗
denote the measured sEMG signals and
source and the pth point, which is comprised of electrode t electrode positions after rotation, respectively, and T denotes
∗
consisting of the ith channel, and Np denotes the number of the rotation matrix considering only roll rotation. PM can be
points selected for the electrode construction. represented as T PM because the electrodes comprising the
The target limb, which is the forearm, was considered to interface are simultaneously rotated. Therefore, the rotation
be a rigid cylinder. The forearm thickness varies depending can be identified by estimating T .
 5

After the rotation identification, the compensated sEMG A. Signal Reconstruction

signals can be obtained as follows:
1 motions and no motion (rest condition) are described in Fig. 4.
VS0 = minVS ||f (PS , CS , T PM )VS − VM ∗ ||22 ,
2
subject to VS ≥ 0
V˜M = f (PS , CS , PM )VS0
= DVS0
where V˜M denotes the converted sEMG signals with ref-
erence to the initial electrode parameters.
Although the source parameters depend on the motions,
it is sufficient to estimate the rotation by performing only a
single motion that represents the rotation well. In other words,
performing only a single motion is enough to find a proper
T that can be applied to other motions. The experimental
results indicate that ulnar deviation exhibits a pattern that is
sensitive to rotation for the able-bodied subjects. This will be
demonstrated in section III-A and III-B.
III. R ESULTS
In sections III-A and III-B, model-based signal reconstruc-
tion and rotation compensation are demonstrated. Additionally,
a best single motion for the model update is presented. The
figures in these sections were obtained from a representative
subject.
The normalized RMS sEMG signal distributions, identified
source positions, and reconstructed signal distributions for five
In our previous work [25], experiments were performed with
six motions, including radial deviation. However, in contrast to
other motions, radial deviation was uncomfortable and caused
fatigue in several subjects. Therefore, in this study, only rest
(9) five motions were considered.
To quantify the reconstruction ratio, R2 [29] was calculated
as follows:
Pm 2
i=1 (VRi − VM )
R2 = Pm 2
(10)
i=1 (VMi − VM )
where VMi denotes the signal measured from the ith chan-
nel, VRi denotes the reconstructed sEMG signal for the ith
channel, VM denotes the mean of the measured sEMG signals,
and m denotes the number of sEMG channels.
The calculated R2 for different conditions is shown in
Fig. 5. The hand close represented poor reconstruction com-
pared to other motions; thus, for this case, the motion required
an additional source for good reconstruction. The recon-
1
0.9
Reconstruction ratio

0.8

In sections III-C and III-D, compensations for both short 0.7

and large displacements are presented. Ten able-bodied sub-
0.6
jects participated in the experiments.
As described in section III-E, long-term and repeated use 0.5

are investigated. Three able-bodied subjects participated in the 0.4

experiment.
0.3
In section III-F, an experiment performed with a single No Wrist Wrist Hand Hand Ulnar
motion extension flexion close open deviation
transradial amputee is presented. Motion

Fig. 5. The mean and standard deviation of reconstruction ratio according

to motion from a representative subject. Hand close produced a poor recon-
struction compared to other motions.

(a) No motion (b) Wrist extension (c) Wrist flexion

(d) Hand close (e) Hand open (f) Ulnar deviation Fig. 6. Signal reconstruction using two sources for hand close from a
representative subject: individual source positions (blue and red points) and
Fig. 4. Normalized sEMG signal distribution and its reconstruction from a reconstructed signal distributions according to each source (blue and red lines).
representative subject: The measured signal distribution for each electrode Additionally, the merged reconstructed signal distribution (green line) and
arrangement (red line), source position (black point), and reconstructed signal measured signal distribution (black line) are described. This indicated that
distribution (blue line). two sources are sufficient for reconstructing the signal.
 6

structed signal considering two sources for the hand close is

shown in Fig. 6.
Additionally, although hand open represented a good recon-
struction, heuristically, hand open also required two sources
for good motion recognition performance. Therefore, both
hand close and open were reconstructed using two sources.
The ulnar deviation achieved an R2 of almost 1 with
No motion Extension Flexion Hand close Hand open Ulnar
a single source and represented good reconstruction. This Trained motion

indicates that sEMG interface rotation can be recognized by (a)

only considering ulnar deviation. In this study, ulnar deviation-
based compensation was conducted.

B. Rotation Recognition and Compensation

Upon initial electrode placement, five motions were consec-
utively performed. Subsequently, the interface was randomly
rotated and the same procedure was conducted.
The average sEMG signal distributions corresponding to (b)
static muscle contractions before and after rotation and their
compensated distributions are shown in Fig. 7. The compen-
sated sEMG signals represented distributions that were similar
to the signals corresponding to the initial electrode configura-
tion. Although the sEMG signal during resting showed poor
reconstruction, this signal can be ignored because no muscle
was activated during the condition. Thus, the motion can be
recognized using the compensated signal after rotation. (c)
The measured sEMG signals during the five static muscle
contractions for model parameter identification are shown
in Fig. 8 (a). The sEMG signals for the five consecutively
performed motions before and after rotation are shown in
Fig. 8 (b)–(c). They represented different patterns due to
change in electrode placement. The signals compensated by
the model are described in Fig. 8 (d). The compensated signals
represented similar patterns to those for the initial electrode
(d)
configuration. Although the rotation induced different distri-
butions, the proposed method enabled signal conversion to the Fig. 8. Rotation compensation from a representative subject: (a) Signals
initial electrode placement. measured during static contractions for model parameter identification. Signals
measured during the five consecutively performed motions (b) before and (c)
after rotation. (d) Compensated signals upon initial electrode placement.

Ulnar Actual motion

Hand open Recognized motion
Hand close
(a) No motion (b) Wrist extension (c) Wrist flexion Flexion
Extension
No motion

Ulnar Actual motion

Hand open Recognized motion
Hand close
Flexion
Extension
No motion

Fig. 9. sEMG signals after rotation and corresponding motion recognition

(d) Hand close (e) Hand open (f) Ulnar deviation with and without compensation from a representative subject. Rotation
compensation significantly (p < 0.005) improved the motion recognition
Fig. 7. Normalized distribution of the RMS signals from a representative performance from 37.5% to 99.4%
subject. The red and blue lines indicate the signals measured from the initial
electrode placement and after rotation, respectively. The black line indicates
compensated signals after rotation, which exhibited similar patterns to that
from the initial electrode placement. The result of motion recognition after the rotation is shown
 7

100
With compensation 350

80 Without compensation
300

Angular difference [°]

25%—75%
250
60
min—max
200
40

150
20
100
45°
0 50
Small Large Measured
rotation rotation Recognized
0
1 2 3 4 5 6 7 8 9
Fig. 10. Motion recognition accuracy according to level of rotation. There was Session
no significant difference (p = 0.17) between the small and large rotations.
Fig. 11. Measured and recognized angular differences based on the first
session from a representative subject. The shape of the actual forearm is not
in Fig. 9. The case in which compensation was not performed cylindrical. Therefore, the angular increment in each trial was not exact 45◦ .
Furthermore, there were differences between the measured and recognized
demonstrated poor recognition performance (37.5%). How- angular differences due to the model assumption.
ever, the proposed model-based compensation provided con-
sistent motion recognition performance after rotation (99.4%). 360

315

Recognized angular difference [°]

270
C. Rotation Compensation for Small/Large Rotation
225
To demonstrate that the proposed method works for both
180
small and large rotations, subjects were asked to rotate the
interface, and then, the compensation was performed. Due to 135

the anatomical variance of the subjects, it is difficult to fix the 90

level of displacement for all subjects. Therefore, the subjects 45

were only instructed that the rotation should not exceed half of
0
the mark left on the skin due to the pressure of the electrodes. 0 45 90 135 180 225 270 315 360
Acutal angular difference [°]
Subsequently, they were asked to rotate large enough without
considering the level of displacement, and then, the same Fig. 12. Differences between the measured and recognized angles from all
procedure was performed. subjects.
The result is shown in Fig. 10. Without compensation,
100
recognition with a small rotation produced relatively higher With compensation
90
performance than that with a large rotation. With compensa-
80 Without compensation
tion, both cases showed consistent performance (p = 0.17).
70 25%—75%
60
min—max
50
D. Rotation Compensation for Wide Ranges
40
Five motions were performed nine times each. In each 30

session, the interface was rotated according to the mark left on 20

the skin as a result of pressing the electrode in which the ideal 10

angular increment was 45◦ . Additionally, the rotation angle 0

0 45 90 135 180 225 270 315 360
was measured using the sEMG interface-integrated IMU. Intended angular difference [°]

The measured and recognized angular differences based on

the first session are shown in Fig. 11 for a representative
subject. In the model, the forearm was considered to be a
rigid cylinder. However, the actual forearm does not have a
cylindrical shape, and the electrode pressed against the skin;
therefore, the measured angular increment in each session was
not exact 45◦ .
Fig. 12 shows the differences between the measured and
recognized angles from all subjects. The results showed a
reliable recognition.
Additionally, average motion recognition accuracy with and
without compensation for all subjects are shown in Fig. 13.
The results indicated that the compensation significantly im-
proved the performance (p < 0.005) with the average accuracy
in all cases increasing from 43.1% to 68.1%.
Fig. 13. Motion recognition accuracy with and without compensation ac-
cording to angular differences from all subjects. The overall accuracy was
increased from 43.1% to 68.1%
E. Investigation of Long-term and Repeated Usage
sEMG signals for five motions were measured at intervals
of 30 min. All 11 sessions were conducted. In each session,
the subject put on and took off the interface, disregarding the
sEMG interface rotation. In other words, the subject did not
consider the electrode configuration in the previous session.
In the first session, the model parameters were identified,
and the relationship between the motion and sEMG signal was
trained. In other sessions, the compensation was performed.
The motion recognition performance with and without com-
 8

100

90

Accuracy of recognition [%]

80
With compensation
70
Without compensation
60

50

40

30

20
(a) Recognition for session 2.
10

0
0 50 100 150 200 250 300
Time [min]

Fig. 14. Motion recognition accuracy with and without compensation for
long-term and repeated usage from a representative subject. Rotation com-
pensation produced a consistently high performance.

Session 2
(b) Recognition for session 3.

Difference of 𝟒𝟓° Fig. 17. Motion recognition with and without compensation. The performance
Session 1 was significantly increased from 51.1% to 79.5% and from 41.6% to 83.6%
for sessions 2 and 3, respectively.
Wrist Wrist Ulnar Hand
extensions flexions deviations close

Session 3
F. Validation with a Transradial Amputee
Wrist Wrist Ulnar Hand
extensions flexions deviations close
The measured sEMG signals during muscle contractions are
Difference of 𝟏𝟐°
described in Fig. 15. In each session, the subject was asked
Wrist Wrist Ulnar Hand
to take off the interface, randomly rotate it, and put it on
extensions flexions deviations close
again. Therefore, very different sEMG patterns were obtained
Fig. 15. Measured sEMG signals for different sessions. Due to the rotation, depending on the session.
different patterns were obtained. The average patterns during static muscle contractions
are described in Fig. 16. Unlike the able-bodied subjects,
wrist extension produced a consistent pattern. Therefore, wrist
extension-based compensation was performed. This indicates
that the optimal motion for rotation compensation depends on
the muscle properties of the user.
From the first session, model parameters were identified and
the relationship between the motion and sEMG signal was
trained. The recognition with and without compensation are
(a) Wrist extension (b) Wrist flexion presented in Fig. 17. The compensation significantly increased
the recognition accuracy from 51.1% to 79.5% and from
41.6% to 83.6% for sessions 2 and 3, respectively.

(c) Ulnar deviation
Session 1
Fig. 16. Normalized RMS sEMG signal distributions from all sessions. Wrist
extension produced a consistent distribution compared to other motions.
pensation for different sessions with a representative subject
are shown in Fig. 14. The results confirmed that the proposed
model-based compensation guarantees long-term and repeated
usage without additional training (p < 0.005). The overall
accuracy was increased from 40.8% to 97.1%.
IV. D ISCUSSION
A. Summary of Experiments
(d) Hand close With known electrode parameters, the source parameters can
Session 2 Session 3 be identified for an individual motion. Each source contains
information about the anatomical factors that are related to
the corresponding motion. Once the source parameters are
identified, the rotation can be compensated for by performing
a single motion for two seconds.
To demonstrate both rotation recognition and its compen-
sation, motion recognition was conducted by rotating the
sEMG interface. Ten able-bodied subjects and one transradial
amputee participated in the experiments. The results indicated
that the proposed model guarantees the recognition of any
rotation.

B. Muscle Property Variation
In this paper, ulnar deviation was consistently selected for
the able-bodied subjects to compensate rotation. Additionally,
the number of sources for each motion was heuristically
chosen by observing the sEMG signal distribution and recon-
struction ratio. However, the sEMG spatial patterns depend
on the involved anatomical factors. There might be many
variations depending on the subject. Especially, that might be
dominant for a person who has an amputation or a muscular
disorder. As an example, in the case of the transradial amputee
who participated in this experiment, wrist extension was the
best motion for compensation. Therefore, to achieve accurate
compensation, optimal selection of subject-specific parameters
is required by analyzing the muscle properties.
C. Inter-Subject Variability
The proposed model facilitates long-term and repeated
usage by identifying the parameters for the muscle activa-
tion source that depend on the anatomical factors. However,
anatomical factors vary from user to user. Thus, the source
parameters determined for one user cannot be simply applied
to a different user. We conducted several experiments in which
the model parameters obtained for one user were applied
to another user (not described in this paper). However, the
performance was inconsistent and even significantly degraded
in several cases. Therefore, the proposed model-based com-
pensation is suitable for applications that are for only a single
user.
D. Scope for Improvement
Fig. 18 describes a comparison of the performance between
ulnar deviation-based and actual angle-based compensations
for a representative subject. The actual angle was measured
using IMU. The accuracy of actual angle-based compensation
(94.8%) was slightly lower (p = 0.14) than that of ul-
nar deviation-based compensation (98.6%). The performance
100
95
Accuracy of recognition [%]
90
85 Ulnar deviation-based compensation
Actual angle-based compensation
80
75
0 45 90 135 180 225 270
Actual angular difference [°]
Fig. 18. Performance comparison between ulnar deviation-based and actual
angle-based compensation from a representative subject. Due to the differ-
ence between the model assumption and actual forearm, actual angle-based
compensation produced a slightly lower (p = 0.14) accuracy (94.8%) than
ulnar deviation-based compensation (98.6%). The result indicated that a more
accurate model can achieve an accurate compensation without performing
even a single motion.
9
degradation for the actual angle-based compensation was
caused by the difference between the actual forearm and the
model assumption. As mentioned above, the current version of
the proposed model does not accurately represent the elliptical
shape of an actual forearm. Furthermore, skin deformation can
occur due to the muscle contraction and the electrode pressing
against the skin. Enhanced compensation based only on the
angle measured using an IMU could be achieved using a more
accurate model that considers the actual forearm shape and
skin deformation.
Additionally, during donning and doffing, not only rotation
but also translation can occur. By including the translation
matrix in (8), the translation could be compensated. However,
a band-type interface has a longitudinal limitation; thus, a large
translation cannot be compensated for. A sleeve-type high-
density sEMG interface could help to provide a more reliable
compensation that considers translation due to a high spatial
resolution and a large skin contact area.
One of the advantages of the proposed compensation is that
the process is separate from the motion recognition process.
Therefore, any motion recognition method can be integrated.
In this study, a simple shallow neural network was integrated.
Thus, the motion recognition performance can be improved
using a state-of-the-art algorithm.
E. Limitation of Source Extraction
The proposed model requires non-negative sEMG signals
to identify the model parameters. Therefore, other types
of features widely used in sEMG signal analysis, such as
wavelength, slope sign change, and autoregressive coefficients,
cannot be applied. However, many deep learning-based motion
estimations have been proposed. By integrating a state-of-
the-art motion-estimating method, the proposed compensation
might guarantee high performance using only RMS features.
F. sEMG Signal Reconstruction
The proposed model-based source extraction is a dimension
reduction technique. In this study, only eight channels were
used due to the specification of the MYO armband. Therefore,
there might be no significant differences between source-
based recognition [25] and reconstructed sEMG signal-based
recognition. However, many high-density interfaces use more
than a hundred channels [30], [31]. Due to rich spatial infor-
mation, motion recognition using a large number of features
from sEMG channels might show better performance than
that using dimension-reduced features (sources). sEMG signal
reconstruction might aid precise motion recognition while
maintaining a number of feature dimensions.
315 360
V. C ONCLUSION
This paper proposes model-based rotation recognition and
signal compensation. This concept is based on the assump-
tion that the contraction of a muscle involves activation
sources. Several experiments were conducted to confirm this
assumption and to demonstrate the compensation. The re-
sults indicated that simple and fast rotation compensation

can be accomplished by performing only a single motion.
The proposed method can be particularly useful in assistant
robots and prosthetic hands. These applications require long-
term and repeated usage in everyday activities. By applying
rotation compensation, their reliability and functionality might
be improved.
R EFERENCES
[1] M. T. Wolf, C. Assad, M. T. Vernacchia, J. Fromm, and H. L. Jethani,
“Gesture-based robot control with variable autonomy from the jpl
biosleeve,” in Robotics and Automation (ICRA), 2013 IEEE International
Conference on. IEEE, 2013, pp. 1160–1165.
[2] J. Vogel, C. Castellini, and P. van der Smagt, “Emg-based teleoperation
and manipulation with the dlr lwr-iii,” in Intelligent Robots and Systems
(IROS), 2011 IEEE/RSJ International Conference on. IEEE, 2011, pp.
672–678.
[3] Z. O. Khokhar, Z. G. Xiao, and C. Menon, “Surface emg pattern
recognition for real-time control of a wrist exoskeleton,” Biomedical
engineering online, vol. 9, no. 1, p. 41, 2010.
[4] Y. Yun, S. Dancausse, P. Esmatloo, A. Serrato, C. A. Merring, P. Agar-
wal, and A. D. Deshpande, “Maestro: an emg-driven assistive hand
exoskeleton for spinal cord injury patients,” in Robotics and Automation
(ICRA), 2017 IEEE International Conference on. IEEE, 2017, pp.
2904–2910.
[5] D. Kim, O. Hilliges, S. Izadi, A. D. Butler, J. Chen, I. Oikonomidis, and
P. Olivier, “Digits: freehand 3d interactions anywhere using a wrist-worn
gloveless sensor,” in Proceedings of the 25th annual ACM symposium
on User interface software and technology. ACM, 2012, pp. 167–176.
[6] G. Marin, F. Dominio, and P. Zanuttigh, “Hand gesture recognition with
leap motion and kinect devices,” in Image Processing (ICIP), 2014 IEEE
International Conference on. IEEE, 2014, pp. 1565–1569.
[7] M. Amjadi, A. Pichitpajongkit, S. Lee, S. Ryu, and I. Park, “Highly
stretchable and sensitive strain sensor based on silver nanowire–
elastomer nanocomposite,” ACS nano, vol. 8, no. 5, pp. 5154–5163,
2014.
[8] J. Kim, M. Lee, H. J. Shim, R. Ghaffari, H. R. Cho, D. Son, Y. H.
Jung, M. Soh, C. Choi, S. Jung et al., “Stretchable silicon nanoribbon
electronics for skin prosthesis,” Nature communications, vol. 5, p. 5747,
2014.
[9] J. G. Ngeo, T. Tamei, and T. Shibata, “Continuous and simultaneous
estimation of finger kinematics using inputs from an emg-to-muscle ac-
tivation model,” Journal of neuroengineering and rehabilitation, vol. 11,
no. 1, p. 122, 2014.
[10] M. V. Liarokapis, P. K. Artemiadis, P. T. Katsiaris, K. J. Kyriakopou-
los, and E. S. Manolakos, “Learning human reach-to-grasp strategies:
Towards emg-based control of robotic arm-hand systems,” in Robotics
and Automation (ICRA), 2012 IEEE International Conference on. IEEE,
2012, pp. 2287–2292.
[11] W. Geng, Y. Du, W. Jin, W. Wei, Y. Hu, and J. Li, “Gesture recognition
by instantaneous surface emg images,” Scientific reports, vol. 6, p.
36571, 2016.
[12] M. Carozza, G. Cappiello, G. Stellin, F. Zaccone, F. Vecchi, S. Micera,
and P. Dario, “On the development of a novel adaptive prosthetic
hand with compliant joints: experimental platform and emg control,”
in Intelligent Robots and Systems, 2005.(IROS 2005). 2005 IEEE/RSJ
International Conference on. IEEE, 2005, pp. 1271–1276.
[13] A. Fougner, Ø. Stavdahl, P. J. Kyberd, Y. G. Losier, and P. A. Parker,
“Control of upper limb prostheses: terminology and proportional myo-
electric control—a review,” IEEE Transactions on neural systems and
rehabilitation engineering, vol. 20, no. 5, pp. 663–677, 2012.
[14] A. D. Roche, H. Rehbaum, D. Farina, and O. C. Aszmann, “Prosthetic
myoelectric control strategies: a clinical perspective,” Current Surgery
Reports, vol. 2, no. 3, p. 44, 2014.
[15] A. J. Young, L. J. Hargrove, and T. A. Kuiken, “Improving myo-
electric pattern recognition robustness to electrode shift by changing
interelectrode distance and electrode configuration,” IEEE Transactions
on Biomedical Engineering, vol. 59, no. 3, pp. 645–652, 2012.
[16] L. Hargrove, K. Englehart, and B. Hudgins, “A training strategy to re-
duce classification degradation due to electrode displacements in pattern
recognition based myoelectric control,” Biomedical signal processing
and control, vol. 3, no. 2, pp. 175–180, 2008.
10
[17] H. J. Hermens, B. Freriks, C. Disselhorst-Klug, and G. Rau, “Devel-
opment of recommendations for semg sensors and sensor placement
procedures,” Journal of electromyography and Kinesiology, vol. 10,
no. 5, pp. 361–374, 2000.
[18] L. Hargrove, K. Englehart, and B. Hudgins, “The effect of electrode
displacements on pattern recognition based myoelectric control,” in
Engineering in Medicine and Biology Society, 2006. EMBS’06. 28th
Annual International Conference of the IEEE. IEEE, 2006, pp. 2203–
2206.
[19] A. J. Young, L. J. Hargrove, and T. A. Kuiken, “The effects of electrode
size and orientation on the sensitivity of myoelectric pattern recognition
systems to electrode shift,” IEEE Transactions on Biomedical Engineer-
ing, vol. 58, no. 9, pp. 2537–2544, 2011.
[20] C. Pylatiuk, S. Schulz, and L. Döderlein, “Results of an internet
survey of myoelectric prosthetic hand users,” Prosthetics and orthotics
international, vol. 31, no. 4, pp. 362–370, 2007.
[21] A. Boschmann and M. Platzner, “Towards robust hd emg pattern
recognition: Reducing electrode displacement effect using structural
similarity,” in Engineering in Medicine and Biology Society (EMBC),
2014 36th Annual International Conference of the IEEE. IEEE, 2014,
pp. 4547–4550.
[22] L. Pan, D. Zhang, N. Jiang, X. Sheng, and X. Zhu, “Improving
robustness against electrode shift of high density emg for myoelectric
control through common spatial patterns,” Journal of neuroengineering
and rehabilitation, vol. 12, no. 1, p. 110, 2015.
[23] S. Muceli, N. Jiang, and D. Farina, “Extracting signals robust to
electrode number and shift for online simultaneous and proportional
myoelectric control by factorization algorithms,” IEEE Transactions on
Neural Systems and Rehabilitation Engineering, vol. 22, no. 3, pp. 623–
633, 2014.
[24] D. D. Lee and H. S. Seung, “Algorithms for non-negative matrix
factorization,” in Advances in neural information processing systems,
2001, pp. 556–562.
[25] M. Kim and W. K. Chung, “Muscle activation source model-based semg
signal decomposition and recognition of interface rotation,” in Intelli-
gent Robots and Systems, 2018. IROS 2018. IEEE/RSJ International
Conference on. IEEE, 2018, pp. 2780–2786.
[26] H. Zhang, Y. Zhao, F. Yao, L. Xu, P. Shang, and G. Li, “An adaptation
strategy of using lda classifier for emg pattern recognition,” in Engi-
neering in Medicine and Biology Society (EMBC), 2013 35th Annual
International Conference of the IEEE. IEEE, 2013, pp. 4267–4270.
[27] M. Yoshikawa, M. Mikawa, and K. Tanaka, “A myoelectric interface for
robotic hand control using support vector machine,” in Intelligent Robots
and Systems, 2007. IROS 2007. IEEE/RSJ International Conference on.
IEEE, 2007, pp. 2723–2728.
[28] M. Atzori, M. Cognolato, and H. Müller, “Deep learning with convo-
lutional neural networks applied to electromyography data: a resource
for the classification of movements for prosthetic hands,” Frontiers in
neurorobotics, vol. 10, p. 9, 2016.
[29] T. O. Kvålseth, “Cautionary note about r 2,” The American Statistician,
vol. 39, no. 4, pp. 279–285, 1985.
[30] M. Gazzoni, N. Celadon, D. Mastrapasqua, M. Paleari, V. Margaria, and
P. Ariano, “Quantifying forearm muscle activity during wrist and finger
movements by means of multi-channel electromyography,” PloS one,
vol. 9, no. 10, p. e109943, 2014.
[31] M. Rojas-Martı́nez, M. A. Mañanas, and J. F. Alonso, “High-density
surface emg maps from upper-arm and forearm muscles,” Journal of
neuroengineering and rehabilitation, vol. 9, no. 1, p. 85, 2012.