Learning an EMG Controlled Game: Task-Specific Adaptations and Transfer
Learning an EMG Controlled Game: Task- Specific Adaptations and Transfer
Ludger van Dijk 0 1
Corry K. van der Sluis 1
Hylke W. van Dijk 1
Raoul M. Bongers 0 1
0 University of Groningen, University Medical Center Groningen, Center for Human Movement Science , Groningen , The Netherlands , 2 University of Groningen, University Medical Center Groningen, Department of Rehabilitation Medicine , Groningen , The Netherlands , 3 NHL University of Applied Sciences,Faculty of Engineering, Serious Gaming Group , Leeuwarden , The Netherlands
1 Editor: Dingguo Zhang, Shanghai Jiao Tong University , CHINA
Video games that aim to improve myoelectric control (myogames) are gaining popularity and are often part of the rehabilitation process following an upper limb amputation. However, direct evidence for their effect on prosthetic skill is limited. This study aimed to determine whether and how myogaming improves EMG control and whether performance improvements transfer to a prosthesis-simulator task. Able-bodied right-handed participants (N = 28) were randomly assigned to 1 of 2 groups. The intervention group was trained to control a video game (Breakout-EMG) using the myosignals of wrist flexors and extensors. Controls played a regular Mario computer game. Both groups trained 20 minutes a day for 4 consecutive days. Before and after training, two tests were conducted: one level of the Breakout-EMG game, and grasping objects with a prosthesis-simulator. Results showed a larger increase of in-game accuracy for the Breakout-EMG group than for controls. The Breakout-EMG group moreover showed increased adaptation of the EMG signal to the game. No differences were found in using a prosthesis-simulator. This study demonstrated that myogames lead to task-specific myocontrol skills. Transfer to a prosthesis task is therefore far from easy. We discuss several implications for future myogame designs.
Data Availability Statement: All relevant data are
within the paper and its Supporting Information files.
Funding: This work was supported by the Northern
Netherlands Provinces Alliance (SNN).
Competing Interests: The authors have declared
that no competing interests exist.
Although video games that aim to improve myoelectric control are becoming an important
part of the rehabilitation process following an upper limb amputation [
], little is known
about the benefits of training myoelectric control by video gaming (i.e. using a myogame).
Many studies so far limit their research to the development of the myogame, and do not
include an evaluation of training effects after using the game [
]. Studies that did include the
training of the myogame often did not provide statistical support for apparent improvement in
performance and, with the exception of one study [
], none have used a control group [
Most importantly, there is, to our knowledge, only one study that determined whether training
effects will subsequently transfer to other myoelectric tasks, such as the use of a prosthesis .
That study showed a task-specific learning effect that transferred only on a few highly
taskspecific outcome measures. The study thus raised the concern that the way myogames are
currently adopted in clinical situations might not promote any transfer of skill. The current paper
aims to evaluate this implication.
By focusing research on motivational aspects and playability rather than on explicitly
designing for transfer to activities of daily living (ADL) (see e.g. [
]), myogame development
has been able to proceed without paying much attention to aspects that could constrain
transfer to actual prosthesis use. For example, studies often do not attempt to simulate the way the
amplitude of the myosignal is related to the opening and closing of the prosthesis hand [
Other technical constraints of the prosthesis are also not taken into account (e.g. motor delays,
EMG response curves)–nor can they be, as detailed technical specifications are often not
supplied by prosthesis manufacturers. Finally, the consensus in motor learning literature is that
training is task-specific–that is, in order to transfer a skill between tasks, the goal of these tasks
should be as similar as possible [
]. So far however, myogames deliberately deviate from
real ADL tasks in order to remain entertaining and motivating [
These concerns led Van Dijk et al. [
] to create an experimental set-up that controlled both
technical constraints and the amount of task similarity. After providing evidence that the used
myogames were actually learned, the study showed to what extent the learning of these games
affected the performance of a prosthesis task. The results showed that neither the technical
similarity in EMG interfacing nor the goal of the gaming task will ensure transfer. Rather, only
the training conditions in which very specific feedback was added to the game elicited transfer
to the use of a prosthetic device. Crucially, this feedback was not only relevant to attaining the
gaming task, but the feedback was also important to the prosthetics grasping task that the
participants needed to perform to assess transfer.
Since the current generation of myogames typically has little similarity with the activities
in daily life they set out to promote, the question therefore becomes to what extent the set-up
currently adopted in serious gaming research will elicit transfer to a basic prosthesis task. The
main aim of this study is to determine what learning and transfer effects can be expected
from training with the current generation of myogames. We aimed to stay as close as possible
to currently established practices: we developed a basic but motivational myogame that is
comparable to those currently used in research, and we used a prosthesis task similar to the
previous transfer study that reflects the basic settings and function used by patients in ADL
To reach our aim this study answers three questions. A prerequisite for showing transfer is
showing learning during training. Therefore, the first question is whether our serious game
that incorporates a myoelectric control interface can be learned. If the myogame is learned we
expect an increase in accuracy of in-game performance after training in comparison to the
sham training. Finding a learning effect, the second question is what change in the gaming task
might account for this. Although the above mentioned study [
] does not report on this issue,
it suggests highly task-specific adaptations of the myosignal. We evaluate this prediction by
looking into the relation between the myosignal and the goal of our game (i.e. intercepting a
ball). The third and final question is whether learning effects of this myogame transfer to the
actual use of a prosthesis during a grasping task. If so, we expect participants to get more skilled
at using a myoelectrically controlled prosthesis. This skill improvement will be reflected in (a)
the participants’ ability to adapt the aperture of the grasping hand to the size of an object [
This adjustment has been found in experienced prosthesis users [
] and is also typical for
grasping with an intact hand [
]. Skill improvement will also be reflected in (b) the time
that the myoelectric hand remains maximally opened: this is expected to be shorter in skilled
prosthetic users .
2 / 14
Twenty-eight able bodied adults participated (mean age 21.39 (SD 1.95) y); 21 men and 7
women. The participants played video games for 4.32 (SD 4.38) hours a week. All participants
(1) were right handed, (2) had corrected to normal vision, (3) were free of any (history of)
disorders of the arms or upper body, and (4) had no prior experience in the use of myoelectric
devices. The study was approved by the local ethics committee (Ethics Committee for Human
Movement Sciences, University of Groningen, the Netherlands) and a signed informed consent
was obtained from all participants prior to the start of the experiment. Upon completion of the
experiment all participants received a gift voucher.
In order to train the use of myoelectric control in a serious game, a customized version of the
game “Breakout” was created (originally created by Atari Inc.). This game, called
“BreakoutEMG” was run on a laptop computer. Two active socket 13E200 MyoBock electrodes (Otto
Bock Healthcare products, Austria) were used. The electrodes used a bandwidth of 90-480Hz
and a notch filter at 50Hz. After that the signal was rectified and low pass filtered (2nd-order).
The amplification of the signal could be controlled linearly with a gain controller. These signals
were fed into the laptop computer, via a NI-USB 6009 data acquisition device (National
Instruments Corporation, USA) that sampled the signals at 125 Hz. Custom LabView software
(National Instruments Corporation, USA) digitally filtered the signals (low pass filter, cutoff
frequency 150 Hz). The game sampled these digitally filtered EMG signals at 50 Hz. To log all
the gaming data for analyses, in a separate process the (x and y) positions of the elements of the
game were written to a text file at 90 Hz.
As a sham training, a standard platform game called “Super Mario Bros” was run on a
Nintendo Entertainment System (Nintendo Co. Ltd, Japan). This game was connected to a
To resemble a myoelectric upper-extremity prosthesis for a transradial amputation level as
closely as possible, a myoelectric simulator was developed (Fig 1) [
This simulator consisted of a myoelectric hand attached to an open cast in which the hand
could be placed, and a splint that was adjustable in length and attached the simulator to the
forearm with a Velcro sleeve. The myoelectric hand was a MyoHand VariPlus Speed (Otto
Bock Healthcare products, Austria) with proportional speed (15–300mm/s) and grip force
During the myoelectric simulator task three wooden cylinders were grasped. These cylinders
were 10 cm in height and were either 2 cm (small), 4 cm (medium) or 6 cm (large) in diameter.
In order to measure the aperture of the myoelectric hand during the grasp, a goniometer
(Cermet PC300 potentiometer, Contelec, Switzerland) was attached to the thumb and index finger
of the hand. The goniometer sampled the angle of the hand at 2000Hz and sent this data to the
laptop computer. Because of a technical problem the angular data on the trials grabbing the
largest cylinder could not be established. Therefore, only the data on grabbing the small and
medium cylinders are presented.
Participants were randomly assigned to either the Breakout-EMG group (n = 16) or to the
Control group (n = 12) as they signed up. The Breakout-EMG group trained the game
“Breakout-EMG” (see Fig 2). Breakout-EMG was a videogame in which the objective was to
3 / 14
Fig 1. The myoelectric simulator. Top view of the myoelectric simulator while grasping the medium
cylinder. The goniometer is attached to the thumb and index finger.
intercept a bouncing ball so that it did not hit the ground. By bouncing the ball with the
paddle, a wall of blocks could be hit. The overall objective of the game was to clear the screen of
these blocks. The movement of the paddle to the left and right was controlled using the
myoelectric signals from the flexor or extensor muscles of the wrist, respectively. The speed of the
paddle was proportional to the amplitude of the EMG signals. During testing and training
with Breakout-EMG the participants were free to hold their arm in any position they felt
comfortable with as long as the electrodes were not perturbed (e.g. by hitting the table). The
Control group trained in playing Super Mario Bros. In this game the objective was to control
an avatar and safely guide him through a world by jumping platforms and avoid enemies.
The game was played using a standard hand held Nintendo controller, which was held in the
palms of both hands and typically operated using both thumbs (i.e. pressing down with the
left thumb for moving the avatar left and right, and pressing down with the right thumb for
jumping). The experiment was conducted in 5 days and consisted of 4 training sessions. On
the first day a pretest was performed, after which 4 days of training followed. On the fifth day
a posttest was performed. For practical reasons, participants were randomly assigned to
either have the first training session after the pretest on day 1, or have their fourth training
session prior to the posttest on day 5.
Fig 2. Breakout-EMG. Screenshot of Breakout-EMG showing an example of a terminal ball drop. The
distance that the ball needed to move was calculated by determining the interval from the point at which the
ball began to drop down towards the ground (t0) and the point at which the ball got to the height (y-position) of
the paddle (t1). The required distance was the difference in position of the paddle at t0 and the position of the
ball at t1. The required distance was correlated to the observed net EMG signal (see text for details).
4 / 14
Fitting of the electrodes. Prior to playing Breakout-EMG, the electrodes were fitted
by palpating for the most prominent muscle bellies of the extensors and flexors of the wrist
during contraction. The electrodes were subsequently placed at those sites and held in
place by a flexible wristband. The signals were filtered and sent to the game computer. In
the game environment both signals were calibrated by determining the minimum and
maximum value of each electrode independently and scaling each signal to a standard range before
the game began. The signal was amplified so that reaching the maximum movement speed
in the game required 20% of the maximum voluntary contraction (MVC) of the muscles.
This was necessary to allow for comfortable game-play and prevented muscle fatigue during
For controlling the hand of the prosthesis simulator at the pre- and posttest, the sites for
fitting the electrodes were similarly determined. The electrodes were subsequently placed by
attaching the prosthesis simulator to the participant's arm. The sensitivity of the electrodes was
adjusted to the upper threshold for each participant individually, so that the maximum EMG
signal that could be sustained for 2 seconds of each participant corresponded to the maximum
opening and closing speed of the myoelectric hand.
Pretest and posttest. The pretest was equal to the posttest and these tests were used to
determine the improvement in skill in playing Breakout-EMG and in using the prosthesis
simulator. To determine whether the myogame Breakout-EMG could be learned, during the
pretest and the posttest, participants from both the Breakout-EMG group and from the Control
group were asked to play one level of the game. This level (level 1) consisted of a screen with 45
blocks that needed to be hit by intercepting and bouncing a ball (see Fig 2). The level started
when the experimenter pressed start and finished when the last block was hit. The participants
did not receive specific instructions other than to play the game.
In order to find out whether any improvement of skill in Breakout-EMG transferred to
using the prosthesis simulator, the change in performance during a simple grasping task was
measured. In this task participants sat at a comfortable position in front of a table wearing the
prosthesis simulator. Prior to the start of the task participants were instructed to maximally
open and close the myoelectric hand to establish the minimum and maximum aperture for
each participant. Starting with a closed myoelectric hand, they were then asked to grasp one of
three wooden cylinders that were placed at 35 cm from the starting position of the myoelectric
hand, lift the cylinder slightly, and then place it back at its original position. Each cylinder
needed to be grasped five times. The order in which the cylinders were presented was
randomized. The participants were instructed to be as accurately as possible in grasping, emphasizing
not to focus on speed of performance but rather to focus on not dropping the cylinder while
Training sessions. In each session the Breakout-EMG group trained by playing
BreakoutEMG for 20 minutes. To keep the participants challenged during training, the game consisted
of three levels. These levels differed in the amount of blocks to be hit (increasing the difficulty
of attaining a high accuracy–i.e. a perfect score). After completion of each level, the participants
received feedback on their performance: on their accuracy in intercepting the ball, on the
number of points scored (with each block hit points were added) and the duration of the level. After
playing all three levels, the participants started again at level 1. There were no negative
consequences to a bad performance. The game had no sound.
The Control group played Super Mario Bros for 20 minutes per session. The participants
were instructed to only play the first four levels of the game (i.e. level 1–1 to 1–4) and then start
over. To match Breakout-EMG training, this game was muted so that it too had no sound.
5 / 14
Using customized Matlab (The Mathworks Inc., USA) scripts, all dependent variables used to
determine in-game performance were calculated from the output file provided by
BreakoutEMG (all dependent variables are listed in Table 1). As playing Breakout-EMG proficiently
required a high degree of accuracy in intercepting the ball, we looked at accuracy for in-game
learning effects. The accuracy with which the ball was intercepted was determined as the
number of balls intercepted divided by the total number of balls that dropped to the ground level.
To look into specific adaptations of the EMG signal to the goal of the game, we defined the
“EMG-ball coupling” as a measure of adaption. The strength of the EMG-ball coupling was
determined by calculating the correlation between the distance the paddle needed to move and
the observed net EMG signal during the terminal drop of the ball. We calculated the required
distance (and direction) that the paddle needed to travel from the start to the end of each
terminal ball drop (see Fig 2). The start of the terminal ball drop was defined as the point in time
where the last change in direction of the ball occurred before reaching the height of the paddle.
A change in direction less than 1 cm from the height of the paddle was disregarded, as this
change hardly influenced the required position of the paddle to intercept. The net EMG signal
was the integral of the difference between the calibrated EMG signal of the flexor and extensor
muscle, over the duration of the drop of the ball. The net EMG signal thus had both a
magnitude and a direction, which corresponded to the speed the EMG signals gave to the paddle.
Changes in the use of the myoelectric simulator were determined from the angular data
from the goniometer using customized Matlab scripts. The angular data was filtered using a
low pass filter (cutoff frequency 20 Hz). Subsequently, the start and end of the opening as well
as of the closing of the myoelectric hand were determined from the data. If participants were
better able to control the prosthesis simulator due to a more controlled use of EMG signal in
the game, we expect participants to better adjust the hand opening to the size of the cylinder;
requiring a smaller maximum hand opening (MHO) during the plateau phase as they learned
to use the prosthesis [
]. The plateau phase was defined as the time from the end of the
opening of the hand to the start of the closing, and by definition contained the maximum hand
opening. Based on previous research , we also expect that increased prosthetic skills would
show as a shorter plateau phase. As the goniometer was sometimes repositioned between
participants and sessions, we normalized the angular data to a value between 0 and 1 based on the
measured minimum and maximum value of each participant prior to analysis. The maximum
hand aperture corresponds to a distance between the thumb and index finger of about 10 cm.
So a change in aperture of 0.1 corresponds to ~1 cm in change in distance between thumb and
To determine learning effects, several repeated measures ANOVA's were conducted on
accuracy and on the strength of the EMG-ball coupling, with test (pretest, posttest) as a within
Maximum hand opening
Standard deviation of the maximum hand opening
Within subjects factor(s)
Test and Cylinder
Test and Cylinder
Test and Cylinder
Between subjects factor
6 / 14
subjects factor and group (Breakout-EMG, Control) as a between subjects factor. To determine
transfer, repeated measures ANOVA's were conducted on the maximum hand opening, on the
duration of the plateau phase, and on the standard deviation of the maximum hand opening,
with test (pretest, posttest) and cylinder (small, medium) as a within subjects factor and group
(Breakout-EMG, Control) as a between subjects factor. A summary of all planned analyses can
be found in Table 1.
Based on their skewness and on a Shapiro-Wilk test for normality, we checked the normality
of the dependent variables. All variables were judged to be normally distributed, with the
exception of the MHO and the SD-MHO. We therefore transformed the data on these variables
using a square root transformation (xtrans = p((xmax + 1) + x). As a precautionary measure we
repeated this procedure for the EMG-ball coupling. To check the effects of the distribution we
then repeated our analyses of the MHO, the SD-MHO and the EMG-ball coupling on the
preand posttest with these transformed data. None of the analyses on the transformed data
differed from the analyses on the non-transformed data. We therefore present only the results on
the non-transformed data here.
Effect sizes were calculated using generalized eta-squared (η2G) [
]. For the in-game
learning effects, the Breakout-EMG group is expected to improve relative to controls.
Therefore follow up comparisons were done using one-tailed independent t-tests (with Bonferroni
correction for multiple comparisons). All analyses used a significance level of α = .05.
The accuracy of the Breakout-EMG group across all sessions, and the accuracy of the Control
group on the pre- and posttest can be found in Fig 3. The increase in accuracy appeared to
have been greatest at the start of the training. The improvement in accuracy after all training
sessions was compared to the Control group. Accuracy of both the Breakout-EMG group and
the Control group improved from pretest to posttest.
Fig 3. In-game accuracy. Mean accuracy (and standard error of the mean) on both the pretest and the posttest of the Breakout-EMG group (black points
with grey error bars) and of the Control group (white points with black error bars), as well as the accuracy on all sessions for the Breakout-EMG group only.
Each point denotes one trial of playing level 1 of the game. During the training sessions, after completing level 1 the participants played a trial at level 2 and a
trial at level 3 before having another trial playing level 1. The accuracy on levels 2 and 3 are not shown. The number of trials participants played during a
session depended on the time participants required to complete each trial. Therefore, not all participants managed to play three trials at level 1. The data on
the first trial of each session is based on all 16 participants. The data on the second trial of each session is based on 15–16 participants. The third trial of each
session is based on the data of 11–14 participants. Note that the biggest improvement occurred from pretest to the first session. There was a significant test
effect from pre- to posttest. However, the improvement in the Breakout-EMG group was significantly greater than Controls (see text for details).
7 / 14
Fig 4. EMG-ball coupling. Representative example of the relation between the net EMG signal (x-axis) and
the required distance (cm) to intercept the ball (y-axis) during a full trial of Breakout-EMG (Note that distance
is actually expressed in units specific to the program used to design the game. However, on the monitor we
used, these units are approximately equivalent to centimeters.). The net EMG signal has no unit of
measurement but is the integral of the difference between the flexor and extensor EMG signal over the
duration of the ball drop. Each point represents one interception attempt. To the left an example of a pretest,
to the right an example of the posttest (of the Breakout-EMG group).
Importantly, the increase in accuracy after all training sessions was significantly greater for
the Breakout-EMG group. A repeated measures ANOVA revealed a strong main effect for Test
(F(1,26) = 58.25, p < .001, η2G = .55), and a significant interaction effect Test x Group: F(1,26) =
21.39, p < .001, η2G = .20. A follow up analysis revealed this improvement was explained by a
difference between groups on the posttest (t(26) = -3.42, p = .002).
To better understand the changes in performance of the game, we examined the goal
specific adaptation of the EMG signal from pretest to posttest. For this we used the strength of the
EMG-ball coupling. An example of the EMG-ball coupling on a typical pretest and posttest is
shown in Fig 4. The effect of the training on the EMG-ball coupling can be found in Fig 5.
Fig 5. Change in EMG-coupling. Mean strength of the EMG-ball coupling (and standard error of the mean) on the pretest and the posttest for both
groups. Both groups showed a significant test effect. However, the improvement in the Breakout-EMG group was significantly greater than Controls (see
text for further details).
8 / 14
Fig 6. Prosthesis hand aperture. Representative example of the hand aperture during the pretest (left
figure) and posttest (right figure) during grasping a medium size cylinder. Both examples are from the same
participant. This participant was assigned to the Breakout-EMG group, however the example is equally
representative for the Control group. The time (s) is shown on the x-axis, the normalized aperture on the
yaxis. A change in aperture of 0.1 corresponds to a change of ~1 cm in distance between thumb and index
finger. The lines represent five trials of grasping a (medium) cylinder. The four round markers on each line
represent (from left to right) the start of the opening, the end of the opening, the start of the closing and the
end of the closing of the hand.
A repeated measures ANOVA on the strength of the EMG-ball coupling revealed a main
effect for Test (F(1,26) = 10.09, p = .004, η2G = .25), and a significant interaction effect Test x
Group: F(1,26) = 4.76, p = .038, η2G = .12. A follow up analysis revealed this improvement was
due to a difference between groups on the posttest (t(14.46) = -2.42, p = .029).
Transfer to prosthesis use
A typical example of the aperture of the myoelectric over time is shown in Fig 6. Due to a
technical problem one of the participants in the control group could not complete the pretest
prosthesis task and was excluded from further analysis.
First, we looked at the maximum hand opening. The (normalized) maximum hand opening
is shown in Table 2. Statistical analysis revealed a small significant effect for cylinder (F(1,25) =
9.67, p < .001, η2G = .05). There was no significant main effect for Test, and there was no
significant interaction effect Test x Group.
Adaptation of the hand aperture to the size of the cylinder could also be expected through a
three-way interaction of Test x Group x Cylinder. That is, the difference in the hand aperture
between cylinders is expected to increase over time for the Breakout-EMG groups more than
for the Controls. Analysis however revealed no significant three-way interaction effect.
The duration of the plateau phase can also be seen in Table 2. Analysis revealed a small
significant effect for cylinder (F(1,25) = 6.74, p = .016, η2G = .04). There was no significant main
effect for Test, nor were there any significant interaction effects.
The standard deviation of the normalized maximum hand opening is also shown in Table 2.
As one of the participants in the control group had only one correct grasp of the small cylinder
during the pretest, the standard deviation in MHO could not be established in this case. We
excluded this participant from analysis. Analysis revealed a significant effect for cylinder
(F(1,24) = 9.52, p < .001, η2G = .14). There were no other significant effects.
In this study we tested whether a simple myogame that conforms to the specifications of the
current generation of myogames (i) can be learned, and if so (ii) what changes in the myosignal
9 / 14
may account for such in-game learning. Finally, we tested (iii) whether in-game improvement
would transfer to a prosthesis task conforming to the settings typically used in clinical practice.
Our results showed that performance on playing the Breakout-EMG game improved
significantly in comparison to controls. Moreover, we showed that the increase in in-game
performance was associated with an increase in the EMG-ball coupling. When compared to the
control group however, we found no indications of transfer of this skill to a prosthesis task.
That is, the participants learned to adjust the EMG signal they generated specifically to the
requirements of the gaming task.
The main aim of this study was to determine what learning and transfer effects can be
expected from training with the current generation of myogames. Therefore we aimed to
maximize our chances of finding transfer. To do so, we chose a “sham” control group to control for
testing-effects, for the amount of training, and for motivational aspects such as novelty effects
or the effect of being part of an experiment (a Hawthorne effect). We did not choose a
comparable myoelectric interface with comparable muscular involvement for the controls. With
respect to transfer, this meant that all positive effects of training the game on prosthesis use
should have shown up in the post test performance. Thus, in our opinion, the current set-up
maximized chances of finding transfer to our prosthesis-task. Consequently however, if we had
found transfer, we would not have been able to pin-point its likely origin. Combining the
current results with previous transfer effects [
] however, creating more subtle control conditions
will be an interesting next step in order to tease out how different aspects of a game can
As a step in improving the design of myogames for prosthesis use, the current study aimed
to provide an evaluation of current practices: it stayed close to both the settings and designs of
myogaming research and to settings clinically used in prosthesis fitting. We thus designed a
basic game much like those currently used, a game that was fun to play and easy to control by
10 / 14
myosignals. We trained participants to play this game using the same muscles as they had to
use for handling a prosthesis simulator. The EMG signals were furthermore proportionally
related to the speed of the end-effector, just as in a prosthesis task. Thus we followed the same
logic as earlier studies using myogames [
], but extended this to include a transfer test. If
improvement in prosthetic control was, for example, based on isolating muscular activity, the
repetitive generation of EMG signals or on re-calibrating the acquired EMG control to a new
range (see also ), we should have found transfer to our prosthesis task. Our results however,
corroborate earlier findings [
] and imply that creating a game that transfers effectively to
ADL, may not be that easy.
In the end, myogame training should make the transfer to starting to practice with an actual
prosthesis easier. It might therefore have its biggest role early in the rehabilitation process (i.e.,
in the pre prosthetic phase), when for example neural plasticity is high but wound healing
prohibits the use of a prosthetic device (see [2;11]). To facilitate transfer, our study points to
several design features that deserve scrutiny in future myogame development. First,
BreakoutEMG required less activation to play than did the myoelectric hand (i.e. 20% MVC, which is
~80% “comfortable contraction”, see [
]). In as far as the calibration to MVC is reported,
this is a common design choice that is aimed at preventing fatigue during training [
In accordance with clinical practice [
], the myoelectric hand was however calibrated so that
the maximum opening speed required the MVC sustainable for 2 seconds. Although it was
recently shown that aligning the EMG intensity required for in-game performance with actual
prosthesis use is insufficient for allowing transfer [
], this does not preclude the possibility
that it could create favorable conditions for transfer to occur. Thus it seems that future
myogames should aim to determine the effects of these settings.
Second, as any myoelectric prosthesis, our prosthesis simulator had a time delay between
generating the myosignal and the change in aperture. Such a delay was not present in the game
as this would have made our game unplayable. As our grasping task was self-paced, timing the
EMG signal was much less critical than in Breakout-EMG. It has been shown in a controlled
pre-posttest design that simulating this delay is not sufficient to allow for transfer [
Nonetheless, it may still be beneficial to accommodate for a delay parameter in a future game design.
To do so however, we need better estimates of the movement characteristics of currently
available prostheses in relation to the generated EMG signals. To our knowledge, such estimates are
not currently available. Future research should aim to establish these estimates and determine
their exact effects on transfer.
An interesting aspect of our current study is our finding on in-game learning. This may
help to guide ideas to improve myogaming for prosthesis use. The development of a
strengthened coupling between the generated EMG signal and the game implies that during the game a
very task-specific adaptation of the myosignal occurred; participants coupled their EMG
directly to the required distance to make the paddle move in order to intercept the ball. This
may indicate that when learning a myoelectric skill, it is not the myosignal that is being
controlled as such, nor is control limited to the relation between the signal and the movement of
the end-effector. What is being controlled might be the myosignal relative to goal-relevant
information in the task (see also e.g. [
]). Our in-game learning effects thus add to
the previous effect study  by suggesting that transfer is enabled in so far as the myosignal
can be coordinated to the similar goal-relevant information across tasks.
An important limitation of our current set-up was our use of able bodied participants
controlling a prosthesis simulator. So far, myogame research has not shown much empirical
evidence for their benefit (see e.g. [
], but see ). The current designs therefore do not
yet warrant testing for motor learning effects on patient groups. Simulators have been used
before to approximate prosthesis use and it appeared that kinematic performances is
11 / 14
comparable to performances with real upper limb prostheses [
]. The advantage of using
these methods is that the small population of persons with upper limb amputations will not be
bothered with research that does not lead to clinically useful results. Generalization to clinical
populations that are already using myogames should however be handled with
caution–especially since myogames might also be used for rehabilitation goals other than motor learning. It
should also be noted that we cannot rule out that our training period was too short for transfer
to occur. Although previous research that used the same amount of training was successful in
this respect [
], and although we did find in-game learning effects, transfer of this particular
game might require more extensive training time.
Although myogames are becoming an integral part of rehabilitation, designing a game that
actually transfers to ADL is far from trivial. The marked improvement in myogame
performance does not transfer to a prosthesis task. We have thus shown the need to explicitly design
myogames for transfer to daily life and hope to have put some new design considerations on
the map. By providing an evaluation of the transfer effect of the current generation of
myogames, this study can provide a starting point for developing myogames that successfully
transfer to activities of daily life.
S1 File. Data file.
We gratefully acknowledge Wim Kaan and Emyl Smid for their technical assistance, and
Michael Bosschert and Jorn Postma without whom the game could not have been developed.
We thank Marleen Post, Bastiaan Sporrel, and Theo Zondag for their assistance when
collecting the data. We also thank Ramón de la Rosa and two anonymous reviewers for their
insightful commentary. This study has been performed on behalf of research center SPRINT.
Conceived and designed the experiments: LD CS RB.
Performed the experiments: LD.
Analyzed the data: LD CS RB.
Contributed reagents/materials/analysis tools: LD CS HD RB.
Wrote the paper: LD CS HD RB.
12 / 14
13 / 14
1. Anderson F , Bischof WF . Augmented reality improves myoelectric prosthesis training . In: Proceedings of the 9th International Conference on Disabilities Virtual Reality Associated Technologies , Laval, France, 2012 ; p. 1 - 8 .
2. Smurr LM , Gulick K , Yancosek K , Ganz O. Managing the upper extremity amputee: a protocol for success . J Hand Ther 2008 ; 21 ( 2 ): 160 - 175 . doi: 10 .1197/j.jht. 2007 . 09 .006 PMID: 18436138
3. Davoodi R , Loeb G . Development of a physics-based target shooting game to train amputee users of multijoint upper limb prostheses . Presence 2012 ; 21 ( 1 ): 85 - 95 .
4. De la Rosa R , Alonso A , de la Rosa S , Abásolo D . Myo-Pong: a neuromuscular game for the UVa-Neuromuscular training system platform . In: IEEE Virtual Rehabilitation 2008 , 2008 ; p. 61 .
5. Ison M , Artemiadis P. Enhancing Practical Multifunctional Myoelectric Applications through Implicit Motor Control Training Systems . In: Engineering in Medicine and Biology Society (EMBC), International Conference of the IEEE , 2014 ; p. 3525 - 3528 .
6. Lovely DF , Stocker D , Scott RN . A computer-aided myoelectric training system for young upper limb amputees . J Microcomput Appl 1990 ; 13 ( 3 ): 245 - 259 .
7. Armiger RS , Vogelstein RJ . Air-Guitar Hero: a real-time video game interface for training and evaluation of dexterous upper-extremity neuroprosthetic control algorithms . In: Biomedical Circuits and Systems Conference, BioCAS 2008 IEEE , 2008 ; p. 121 - 124 .
8. Ma S , Varley M , Shark L- K , Richards J . EMG biofeedback based VR system for hand rotation and grasping rehabilitation . In: Information Visualisation (IV) ;2010 14th International Conference . IEEE, 2010 ; p. 479 - 484 .
9. Oppenheim H , Armiger RS , Vogelstein RJ . WiiEMG: A real-time environment for control of the Wii with surface electromyography . In: Proceedings of 2010 IEEE International Symposium on Circuits and Systems (ISCAS) . IEEE; 2010 ; p. 957 - 960 .
10. Pistohl T , Cipriani C , Jackson A , Nazarpour K. Abstract and Proportional Myoelectric Control for MultiFingered Hand Prostheses . Ann Biomed Eng 2013 ; 41 ( 12 ): 2687 - 2698 . doi: 10 .1007/s10439-013 - 0876-5 PMID: 23934195
11. Terlaak B , Bouwsema H , van der Sluis CK , Bongers RM . Virtual training of the myosignal . PloS ONE 2015 ; 10 ( 9 ): e0137161. doi: 10.1371/journal.pone.0137161 PMID: 26351838
12. Van Dijk L, van der Sluis CK , van Dijk HW , Bongers RM . Task-oriented gaming for transfer to prosthesis use . IEEE Trans Neural Syst Rehabil Eng 2015 ; published ahead of print.
13. Newell KM . Change in movement and skill: Learning, retention, and transfer . In: Latash ML , Turvey MT (eds) Dexterity and its Development . Mahwah, NJ: Lawrence Erlbaum, 1996 ; p. 393 - 429 .
14. Reed ES . Applying the theory of action systems to the study of motor skills . In: Meijer OG , Roth K (eds ) Complex Movement Behavior 'The' Motor-Action Controversy . Amsterdam: Elsevier, 1988 ; p. 45 - 86 .
15. Schmidt RA. A schema theory of discrete motor skill learning . Psychol Rev 1975 ; 82 ( 4 ): 225 - 260 .
16. Woodworth RS , Thorndike E. The influence of improvement in one mental function upon the efficiency of other functions.(I) . Psychol Rev 1901 ; 8 ( 3 ): 247 - 261 .
17. Bouwsema H , van der Sluis CK , Bongers RM . Movement characteristics of upper extremity prostheses during basic goal-directed tasks . Clin Biomech 2010 ; 25 ( 6 ): 523 - 529 .
18. Bootsma RJ , Marteniuk RG , MacKenzie CL , Zaal FT . The speed-accuracy trade-off in manual prehension: effects of movement amplitude, object size and object width on kinematic characteristics . Exp Brain Res 1994 ; 98 ( 3 ): 535 - 541 . PMID: 8056073
19. Castiello U. The neuroscience of grasping . Nat Rev Neurosci 2005 ; 6 ( 9 ): 726 - 736 . PMID: 16100518
20. Meulenbroek RG , Rosenbaum DA , Jansen C , Vaughan J , Vogt S. Multijoint grasping movements . Exp Brain Res 2001 ; 138 ( 2 ): 219 - 234 . PMID: 11417463
21. Smeets J , Brenner E. A new view on grasping . Motor Control 1999 ; 3 ( 3 ): 298 - 301 .
22. Bouwsema H , Kyberd PJ , Hill W , van der Sluis CK , Bongers RM . Determining skill level in myoelectric prosthesis use with multiple outcome measures . J Rehabil Res Dev 2012 ; 49 ( 9 ): 1331 - 1348 . PMID: 23408215
23. Bouwsema H , van der Sluis CK , Bongers RM . Learning to control opening and closing a myoelectric hand . Arch Phys Med Rehabil 2010 ; 91 ( 9 ): 1442 - 1446 . doi: 10 .1016/j.apmr. 2010 . 06 .025 PMID: 20801265
24. Bouwsema H , van der Sluis CK , Bongers RM . The role of order of practice in learning to handle an upper-limb prosthesis . Arch Phys Med Rehabil 2008 ; 89 ( 9 ): 1759 - 1764 . doi: 10 .1016/j.apmr. 2007 . 12 . 046 PMID: 18675393
25. Bakeman R . Recommended effect size statistics for repeated measures designs . Behav Res Methods 2005 ; 37 ( 3 ): 379 - 384 . PMID: 16405133
26. Olejnik S , Algina J . Generalized eta and omega squared statistics: measures of effect size for some common research designs . Psychol Methods 2003 ; 8 ( 4 ): 434 . PMID: 14664681
27. Liu X , Mosier KM , Mussa-Ivaldi FA , Casadio M , Scheidt RA . Reorganization of finger coordination patterns during adaptation to rotation and scaling of a newly learned sensorimotor transformation . J Neurophysiol 2011 ; 105 ( 1 ): 454 - 473 . doi: 10 .1152/jn.00247. 2010 PMID: 20980541
28. Radhakrishnan SM , Baker SN , Jackson A . Learning a novel myoelectric-controlled interface task . J Neurophysiol 2008 ; 100 ( 4 ): 2397 - 2408 . doi: 10 .1152/jn.90614. 2008 PMID: 18667540
29. Ortiz-Catalan M , Håkansson B , Brånemark R. An osseointegrated human-machine gateway for longterm sensory feedback and motor control of artificial limbs . Sci Trans Med 2014 ; 6 ( 257 ): 257re6 .
30. Bootsma RJ . Ecological movement principles and how much information matters . In: Post AA , Pijpers JR , Bosch P , Boschker MSJ (eds) Models in human movement science: Proceedings of the second symposium of the institute for fundamental and clinical human movement sciences , Print Partners Ipskamp, Enschede , 1998 ; p. 51 - 63 .
31. Warren WH . The dynamics of perception and action . Psychol Rev 2006 ; 113 ( 2 ): 359 - 389 .