EMG – Force Correlation considering Fitts’ Law

Jangwoo Park, Woongjin Bae, Hyunkyu Kim, Shinsuk Park

Abstract— In recent years, many dangerous tasks are to learn the motion activity of humans.
performed by robots. In most contact tasks, however, humans First, the extraction of the central nerve signal is
are far superior to their robotic counterparts. To control robot achievable via the electroencephalogram (EEG)
manipulators in contact tasks, the characteristics of human measurement. In using the EEG measurement, a
movements should be investigated through biological signals, non-invasive method cannot accurately acquire data due to
such as EMG, which reflects the dexterity of human noise, and an invasive method has the possibility of
manipulation. In this study, EMG signal is analyzed through
damaging neural tissues. Second, the extraction of the
direct and numerical comparison with force estimation. The
EMG and force signals were analyzed in the perspective of peripheral nerve signal is possible through the
Fitts' law, which is a general estimation method on speed and electroneurogram (ENG) and the electromyogram (EMG)
accuracy. The results of this study show that both EMG and measurements. The ENG measurement can damage the
force controls follow Fitts’ law, which suggests that both neural tissues due to an electrode transplant. The EMG
controls can be applicable as a control modality for robot measurement is more suitable since it can be done in a more
control. non-invasive way without the danger of damaging neural
tissues.
I. INTRODUCTION The EMG signal has been actively studied in the past
ranging from the on/off control method [2] to the latest
I n many industrial applications, robots are replacing
humans in various tasks. Tasks performed in dangerous
surroundings, such as space, underwater and high/low
Human Computer Interface (HCI) development [3]. In a
recent study, Choi et al. developed a computer device for
temperature environments, are all unsuitable for humans, as disabled people using a pattern analysis of the EMG signal
well as simple and repetitive tasks. Industrial robots can [4]. In Artemiadis et al.’s research, they controlled the
perfectly carry out highly repetitive actions requiring precise robot’s arm by using an EMG signal [5]. B.D. Kang et al.
placement over long periods of time. Those conventional predicted the human’s multi-joint stiffness by utilizing a
robot, however, are poor at contact tasks, such as opening neural network and the EMG signal [6]. W. Herzog et al.
doors, inserting keys, and fastening bolts, compared to estimated the magnitude of a force with the EMG signal by
humans. In performing particular tasks, humans are able to dynamically contracting the plantaris muscle of a cat [7].
adjust their physical characteristics in order to proficiently These studies were on the analysis of EMG signal patterns,
perform the required task. The physical advantage of the rather than on the quantification or applications of EMG
human body lies in one’s ability to skillfully cope with signals.
dynamic and complex tasks. This means that, for robots to be By using Fitts' law, Fimbel et al. applied the EMG signal
used in more diverse fields, characteristics of the human from the finger and eyebrows to motor control [8]. In their
body must be studied. In order to effectively imitate the study, for the wider application of the EMG signal, the
human body, the movement process of humans must be control ability of the EMG signal was analyzed through the
carefully analyzed. Currently, much research is being done direct and numerical comparison with force measurement
in various fields of human movement studies [1]. and the final result of human action. Fitts' law is a general
The institution that controls the human body exists in the estimation method of speed and accuracy [9]. Fitts’ law
brain. Human action can be simply defined as a signal that states that the movement time (MT) is linearly proportional
comes out of the brain, passes the central nerve and the to index of difficulty (ID) (ID = log2 (A/W+1)), which can be
peripheral nerve, and acts at the end points of the body. Thus, formulated as follows:
the extraction and analysis of this signal process is necessary
§A ·
Manuscript received April 1, 2008. This work was supported by the
MT a  b log 2 ¨  1¸ (1)
Korean Institute of Construction & Transportation Technology Evaluation ©W ¹
and Planning (KICTEP). (Program No: 06-Unified & Advanced
Construction Technology Program-D01) where a and b are constants.
Jangwoo Park is with the Department of Mechanical Engineering, Korea
Index of difficulty depends on the distance A between an
University, Seoul, Korea, 136-701. ([email protected])
Woongjin Bae is with the Department of Mechanical Engineering, object and a target and the width W of the target. Fitts’ law
Messachusetts institute of technology, Cambridge, USA, MA02139. can be applied to various motion estimations, including 2-D
([email protected]) [10] and 3-D [11] motion, rotation in virtual reality [12],
Hyunkyu Kim is with Samsung Electronics CO., LTD., Suwon, Korea drawing [13], and grip force [14]. Although some
443-742 ([email protected])
Shinsuk Park is with the Department of Mechanical Engineering, Korea researchers argues against the validity of Fitts’ law [15][16],
University, Seoul, Korea, 136-701. ([email protected]) the accuracy of EMG measurement may be improved by
measuring EMG and force in the isometric condition.
1 n 2
In this study, the EMG signals from biceps and triceps and
the force generated by them were measured simultaneously.
xRMS ¦ xi
ni1
(2)

These two muscles play the most important role in the arm
contraction reaction [17]. The measured data was analyzed
using Fitts’ law, where the distance between the targets was C. Measurements
defined by the visual targets that indicated the magnitude of When the EMG signal is measured, its range is assigned
the EMG signal and the force. The width of the target was between 0 - 1.2 V by LAXTHA’s EMG signal measurement
chosen according to the index of difficulty of the task. instrument (QMEG4). This signal range becomes 0 - 1200 V
The results of this study can be applied to telerobotic by 1000 times and finally changes to 0 - 900 V by passing the
manipulation of contact tasks, where the human operator low-pass filter. The study used the EMG signal between
directly controls the robot from a remote location. The EMG 0-450 V to decrease the fatigue effect from muscles, since
signal as a new modality of teleoperation, can be used to the highest point arrival ability of the EMG differs among
control the compliance of the robot arm, which is one people. The range of the force measured by Force-Moment
important factor of robotic manipulation. Sensor (NITTA Corp.) is 200 N on X axis and Y axis, and
400N on Z axis.
As described in Fig. 1, an experiment position is decided
by the position (90o elbow angle) that can best measure the
II. MATERIAL AND METHOD EMG signal and the forces of biceps and triceps [19].

A. Relationship of EMG and Force

There have been a number of studies on the relationship
between the EMG and force. Also, there has been debate on
the linearity of the relationship between the EMG and force
in the isometric condition of muscles [18]. Recent research
showed that the EMG and force have a linear relationship
regardless of the arm posture and the muscle length [19][20].
Based on this finding, this study purposes a direct
quantification of the relationship between the EMG and the
force.
B. Participants and apparatus
10 Participants with no history of neurological disease
Fig. 1. Experimental Setup
were selected for experiments. The mean age was 26 (SD
To decide Index of difficulty, a distance (A) between an
3.9) and all of them were male and right-handed.
object and a target decided as maximum 800 by adding the
Surface electrodes were placed on the right bicep and the
areas of right biceps and left triceps and it is assigned in the
right triceps, i.e., the points of movement [21]. For the high
order of 800, 700, 600, 500, 450, 400, 350, 300, and 250 as
accuracy, the interval of measurement surface electrodes and
grounding electrodes was determined 1cm. described in Fig. 2. The more precise measurement is
This study used three surface electrodes (measurement, required as a difference between index of difficulty becomes
standard, folding paper) for the EMG signal acquisition and small due to the decrease of the EMG signal value. Because
data was collected by using LAXTHA Co., Ltd's EMG Fitts' law's can not be applied when the width of target (W) is
measuring instrument (QMEG4). Amplification ratio of each too small [16] and the value of highest point is more
channel is 1:1000, and the signal is extracted with 1 kHz. In important, it is decided that a 150 of a target width from
this study, RMS (root mean square) low-pass filters in several times of experiment processes. As a result, the range
equation (2) were used to reduce the variability of signal and of index of difficulty (ID) is 1.4 - 2.7.
to achieve the accurate control. For each screen conversion,
128 data is used and 4 Hz of cutoff frequency gives enough
filtering efficiency.
The Force-Moment Sensor (NITTA Corp.) is used for the
force measurement and a data acquisition board (JR3 PCI,
NITTA Corp.) is used for data collection. Screen conversion
is conducted between 125ms andGa time delay is controlled
by 300ms so that user could not recognize [22].
 experiment participant achieve assignment of 18 times and
total duration of task is about 2 hours.
E. Data capture and analysis
Data collection is achieved when the EMG signal reached
inside target width in an experiment that use the EMG signal.
The EMG signal, force of X, Y, and Z axes, measurement
time were recorded individually. When force reached inside
target width in an experiment that use force by equal basic, is
collected to this data the EMG signal, force of X, Y, Z axis,
time.
Data analysis process is as follows. First, control ability of
the EMG signal can estimate applying experiment data that
use the EMG signal to Fitts’ law. It is expressed in
two-dimension graph, and X axis is index of difficulty and Y
axis is success time. And next step is deciding correlation
Fig. 2. Experimental display setup constant R2 to use Least Square. It can be said that control
ability is good as value of correlation constant R2 is near to 1.
Control ability of force estimated by same method.
Next, we analyzed data that is measured at the same time
doing each experiment. That is, it is the EMG signal data of
an experiment that use force and force data of an experiment
that use the EMG. We applied Fitts' law to these data.
The EMG signal that extract from a person is influenced
Fig. 3. Task progress in length and region of muscle [24]. Therefore, the EMG
signals of triceps and biceps that do not control hard to
compare directly. For comparing the EMG signal of these
muscles directly, normalization processing need.
Normalization process is as follows. First, we measure
maximum value of two the EMG signals that wish to
compare repeatedly and calculate each average of maximum
value. Calculated two average values can compare values
Fig. 4. Task success that are measured in an experiment relatively [24]. As
described in Fig. 5, we used experiment data of most difficult
D. Procedure index of difficulty by maximum value in this paper.
This research provides 10 minutes for which participants
can practice the EMG signal control and a force control
before the experiment begins. The EMG signals of biceps
increases when experiment participants lift a beam that is
connected to Force-Moment Sensor (NITTA Corp.), and the
displacement of a force increases to the positive Z direction.
In the contrary, the EMG signals of triceps increases when
experiment participants put down a beam, and the
displacement of a force increases to the negative Z direction.
As described in Fig. 3, when a signal reaches inside the
width of a target, a progress graph starts to operate and then,
as described in Fig. 4, show that the task is achieved as
changing the color of a target to red (triceps) and to blue
(biceps) when a signal stays more than 1 second within the Fig. 5. Signal Normalization Process
range of a target width. If a signal escapes from the range of
a target width, a progress graph stops operating and returns
to the starting point. Experiment participants were required The data that pass through normalization process is
to escape the target’s width range as fast as possible and the applied to Fitts' law. The data that flow normalization
experiment order by index of difficulty is randomly chosen. process need index of difficulty choice of task to apply Fitts'
Each duration of Index of difficulty was 2 minutes, and law. The index of difficulty of task is decided by the distance
gave rest time of 5 minutes between each task and minimized (A) between targets and the width (W) of target. In this paper,
effect by fatigue occurrence of muscle [23] . All of the we used 95% confidence interval as described in Fig. 6 that
is old and famous method of statistics for deciding the index B. Force Space
of difficulty of data that do not controlled directly.
In the EMG signal data that do not controlled, distance of
target decided by mean value of the EMG signal and width

Fig. 7. Example of Fitts’ law on EMG space

Fig. 6. 95% Confidence Interval The average of Fitts' law correlation constant R2 is 0.912
of target decided using standard deviation with equation (3). in an experiment that control force directly.
Movement time (MT) that is measured by index of difficulty The lowest value and the highest value were recorded
of each task is used. each 0.877, 0.951 in this result. In this case, as was with the
results of the EMG signals, R2 values for the average,
G highest, and lowest values were all close to 1, and in
W 2 u1.96 u (3) comparison to other work based on the existing Fitt’s Law,
n the resulting values are still favorable. This result also shows
that control ability of the EMG signals is credit in occasion
After these processes, the EMG signal that do not of force although it is lower value than result of the EMG
controlled directly can apply to Fitts' law. It is expressed in signal are shown below in Table 1. In addition, force control
two-dimension graph by equal method, and X axis is index is already being widely applied in fields such as control of
of difficulty and Y axis is success time. And next step is robotic arms [6]. Therefore, the results of this study can give
deciding correlation constant R2 to use Least Square. further credibility to previous studies in the area of force
Similarly, It can be said that control ability is good as value control.
of correlation constant R2 is near to 1.
Force is influenced by the EMG signal because it is last C. EMG-Force Correlation
result of the EMG signal. Therefore, apply process of the
EMG signal referred over to analysis of force that do not
control directly.
We can get four Fitts' law correlation constants R2 per
experiment participant of 1 person finally. As a result, we
can estimate control ability of the EMG signal and force
calculating average of correlation constant R2 of all
experiment participants. And we can consider correlation of
force and the EMG signal analyzing an each experiment.

III. RESULTS

A. EMG Space
The average of Fitts' law correlation constant R2 is 0.945 Fig. 8. Example of Fitts’ law on Force space
in an experiment that control the EMG signal directly. The average value of the correlation constant R2, which
The lowest value and the highest value were recorded was calculated by applying the force data obtained from
each 0.884, 0.975 in this result. R2 values for the average, control of the EMG signal to Fitt’s Law, was 0.894. This
highest, and lowest values were all close to 1, and in result was 5.4% less than the average R2 value obtained from
comparison to other work based on the existing Fitt’s Law, the direct control of the EMG signal. The difference when
the resulting values are still favorable. This result talks that compared to the direct control of force was 2.0%. With
control ability of the EMG signal is credit are shown below differences of only 5.4% and 2.0% respectively, the result is
in Table 1. quite good. In addition, because the R2 value is not small
compared to other works that were based on Fitt’s Law, the
results of this study can be considered to be favorable. The However, there may be a difference in the absolute values,
index of difficulty, decided through a normalization process, but control of the EMG signal and control of force can talk
has a range of 1-2. This range is slightly less than the 1.4-2.7 that have almost same principle.
difficulty index of the original experiment, but the difference
was minimal. It should be noted that the width of the target
(W) at each level of difficulty were similar. The results IV. CONCLUSION
suggest that although there is a size difference, EMG The goal of this study was to analyze the control process
signal-based control is almost equivalent to force control. of the EMG signal to evaluate EMG as a novel modality of
The average value of the correlation constant R2, which robot control. The evaluation of the control ability of the
was calculated by applying the EMG signal data obtained EMG signal was performed using Fitts’ Law, which is the
representative method for evaluating speed and precision.
EMG control was directly and numerically compared to the
ability of force control. The comparison with the ability of
force control was based on the linear proportionality
between the two control methods. There were four different
categories for the Fitts’ law correlation constant R2 from the
experiments, and the average, highest, and lowest values
from each experiment are shown in Table 1.
TABLE I
RESULT OF CORRELATION CONSTANT R2
NON- Non-
Control Control
CONTROL Control
EMG Force
EMG Force

MEAN 0.945 0.812 0.912 0.894

Fig. 9. Example of Fitts’ law on Force by EMG Space

from control of force to Fitt’s Law, was 0.812. This result MAX 0.975 0.865 0.951 0.924
was 10.9% less than the average R2 value of 0.912 obtained
from the direct control of force. The difference when MIN 0.884 0.735 0.877 0.823
compared to the direct control of the EMG signal was 14.1%.
Although the difference values of 10.9% and 14.1% are
larger than the difference values shown previously, they are The results of this study indicate that the EMG signal
not large in comparison to other works based on Fitt’s Law control is nearly identical to force control. While previous
and can thus be considered to be good results. The index of studies investigated the movement in task space by
difficulty, which was decided through a normalization analyzing the hand or foot movement, the present study
process, has a range of 1.2-1.8. This is lower than the analyzed the force in task space and the neuromuscular
original experiment’s difficulty index of 1.4-2.7, and the activity in muscle space.
range has also decreased by half. These results suggest that The results of this study show that Fitts’ law applies to the
during force control in the experiment, muscles other than human movement in task space, force in task space and in
the biceps and the triceps came into play. muscle space. This suggests that Fitts’ law is determined in
the level of the nervous system and is not affected by the
dynamics of the limbs.
The EMG can be applied to perform telerobotic contact
tasks by using the EMG signal to modulate compliance of
the robot arm. In addition, EGM signals could be applied to
rehabilitation engineering, such as human-computer
interface for disabled people.

REFERENCES
[1] H. K. Kim, “Strategies for control of neuroprostheses through
brain-machine interfaces,” Ph.D. dissertation, Dept. Mech. Eng., Mass.
Inst. Technol., MA, Cambrige, 2005.
[2] D. S. Childress., “An approach to powered grasp,” in 4th International
Symposium External Control Human Extremities: Advances in
External Control of Human Extremities. Dubrovnik, Yugoslavia,
Fig. 10. Example of Fitts’ law on EMG by Force Space 1972.
[3] Barreto, A. B., Scargle, S. D., & Adjouadi, M., “A practical
EMG-based human-computer interface for users with motor
 disabilities,” Journal of Rehabilitation Research and Development, 37,
1-14, 2000.
[4] Choi, C., Han, H., Ha, S., Kim, J., “Development of an EMG-based
computer interface for the physically handicapped,” HCI2007, pp. 222
– 227, 2007.
[5] Artemiadis, P. K., Kyriakopoulos, K. J., “EMG-based Teleoperation
of a Robot Arm in Planar Catching Movements using ARMAX Model
and Trajectory Monitoring Techniques,” Proceedings of the 2006
IEEE International Conference on Robotics and Automation, pp. 3244
– 3249, 2006.
[6] Kang, B., Kim, B., Park, S. and Kim, H.., “ Modeling of Artificial
Neural Network for the Prediction of the Multi-Joint Stiffness in
Dynamic Condition,” Proceedings of the 2007 IEEE/RSJ International
Conference on Intelligent Robots and Systems, pp. 1840 - 1845, 2007.
[7] Herzog, W., Sokolosky, J., Zhang, Y. T., Guimaraes, C. S.,
“EMG-Force Relation in Dynamically Contracting Cat Plantaris
Muscle,” Journal of Electromyography and Kinesiology 8, pp. 147 –
155, 1998.
[8] Fimbel, E. J., Lemay, M., Arguin M., “Speed-accuracy trade-offs in
myocontrol,” Human Movement Science 25, pp.165 – 180, 2006.
[9] MacKenzie, I. S., “Movement time prediction in human-computer
interfaces,” in Proceedings of Graphics Interface, pp. 140 – 150, 1992.
[10] Bohan, M., LongstaV, M. G., Van Gemmert, A. W. A., Rand, M. K., &
Stelmach, G. E., “Differential effects of target height and width on 2D
pointing movement duration and kinematics,” Motor Control, 7, pp.
278 - 289, 2003.
[11] Murata, A., & Iwase, H., “Extending Fitts˅ law to a
three-dimensional pointing task,” Human Movement Science, 20, pp.
791 – 805, 2001.
[12] Ruddle, R. A., & Jones, D. M., “Manual and virtual rotation of a
three-dimensional object,” Journal of Experimental Psychology:
Applied, 7, pp. 286 – 296, 2001.
[13] Mottet, D., & Bootsma, R. J., “The dynamics of rhythmical aiming in
2D task space: Relation between geometry and kinematics under
examination,” Human Movement Science, 20, pp. 213 – 241, 2001.
[14] Jungling, S., Bock, O., & Girgenrath, M., “Speedˀ accuracy trade-off
of grasping movements during microgravity,” Aviation, Space and
Environmental Medicine, 73, pp. 430 – 435, 2002.
[15] Smits-Engelsman, B. C., Van Galen, G. P., & Duysens, J., “The
breakdown of Fitts˅ law in rapid, reciprocal aiming movements,”
Experimental Brain Research, 145, 222 - 230, 2002.
[16] Danion, F., Duarte, M., & Grosjean, M., “Fitts˅ law in human
standing: The effect of scaling,” Neuroscience Letter, 277, pp. 131 –
133, 1999.
[17] Coggshall, J. C., Bekey, G. A., “EMG-FORCE DYNAMICS IN
HUMAN SKELETAL MUSCLE,” Med. & Biol. Engng. Vol. 8
Pergamon Press Printed in Great Britain, pp. 265 – 270, 1970.
[18] Milner-Brown, H. S., Stein, R. B., “The Relation Between The Surface
Electromyogram And Muscular Force,” J. Physiol. Printed in Great
Britain, pp. 549-569, 1975.
[19] Valle, A. D., Thomas, C. K., “Motor unit firing rates during isometric
voluntary contractions performed at different muscle lengths,” Can. J.
Physiol. Pharmacol. 82, pp. 769 – 776, 2004.
[20] Santo, F. D., Gelli, F., Ginanneschi, F., Popa, T., Rossi, A., “Relation
between isometric muscle force and surface EMG in intrinsic hand
muscles as function of the arm geometry,” Brain Research 1163, pp.
79 – 85, 2007.
[21] Mercer, J. A., Bezodis, N., Delion, D. Zachry, T., Rubley, M. D.,
“EMG sensor location : Does it influence the ability to detect
differences in muscle contraction conditions?,” Journal of
Electromyography and Kinesiology 16, pp. 198 – 204, 2006.
[22] Englehart, K.m Hudgins, B., “A Robust, Real-Time Control Scheme
for Multifunction Myoelectric Control,” IEEE Transactions on
Biomedical Engineering, Vol. 50, No. 7, pp. 848 – 854, 2003.
[23] Oberg, T., “Muscle Fatigue and Calibration of EMG Measurements,”
Journal of Electromyography and Kinesiology, Vol. 5, No. 4, pp. 239 -
243, 1995.
[24] Konrad, P., “The ABC of EMG,” Noraxon INC. USA, pp. 4 - 58,
2005.