Edited by: Keum-Shik Hong, Pusan National University, South Korea
Reviewed by: Aiguo Song, Southeast University, China; James Sulzer, University of Texas at Austin, United States
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Motor disabilities limiting the mobility of limbs affect the quality of lives of people with neural injuries. Among various types of motor disabilities, abnormal intermuscular coordination is commonly observed from people with severe impairment. The concept of muscle synergy, defined as characteristic muscle co-activation patterns activated to produce complex motor behavior, has been applied to assess the alteration in intermuscular coordination in pathological populations. This study presents the development of a robotic system named KAIST upper limb synergy investigation system (KULSIS), for accurate measurement of intermuscular synergies while providing the convenient experimental setup. It provides full force/moment measurements for isometric force generation tasks at various upper limb postures and reaching tasks in a three-dimensional workspace. It is composed of: a three-degree-of-freedom gimbaled handle to adjust the orientation of the handle to accommodate potential hand-wrist deformity, a linear actuator that moves the handle for reaching tasks; a five-degree-of-freedom mechanism for positioning and adjusting the orientation of the linear actuator. The design was evaluated in terms of the workspace of the handle, mechanical stiffness and force/moment measurement accuracy. The position/force measurement is synchronized with electromyographic measurements. Muscle synergy patterns, activated during four isokinetic reaching motions, were also assessed as preliminary data using KULSIS from ten healthy subjects.
Neurological injuries result in limiting the activities of daily living and the quality of life. For example, hemiparetic stroke often manifests major motor issues such as spasticity, muscle weakness and stereotypical, abnormal motor coordination (
The previous studies of the human upper extremity focused on characterizing intermuscular coordination patterns of each individual task, for example, reaching (
In order to examine the generalizability and specificity of intermuscular coordination patterns, a mechanical device can be used to implement both static and dynamic motor tasks of the human upper limb. However, a commercial device which is suitable for the assessment of motor coordination both under an isometric condition and in motion is relatively rare and often involves its own constraints to examine intermuscular coordination patterns in varying biomechanical conditions. For example, two different versions of KINARM (BKIN Technologies Ltd., ON, Canada;
The device, named KAIST upper limb synergy investigation system (KULSIS), provides a unique environment where the reaching path is matched between subject populations in comparison. It adopts a linear track along which subjects produce reaching movement. This constraint ensures a comparably similar end-point trajectory of reaching in both neurologically intact and pathological participants in comparison with minimal interference to the subject. While constraining the motion of the end-point, the pathological subject can still utilize impaired intermuscular coordination since alterations in joint kinematics of elbow and shoulder are allowed. Thus, potential alterations in intermuscular coordination patterns and their activation profile observed in a pathological group can be interpreted as the effects of the pathological condition during reaching movement in a well-controlled way.
This study proposes an experimental setup to examine coordination of upper limb muscles, which features a combination of single active DOF and five passive DOFs. The proposed structure of KULSIS supports a consistent examination of intermuscular coordination for both isometric force generation and constrained reach conditions in a large workspace of the human upper extremity. KULSIS can move a force/moment-measuring handle that a subject hold during measurement along a line in the workspace of upper limb to implement a variety of upper limb postures and the directional variances of reaching motion. During the subject performs given upper limb tasks implemented by KULSIS, force/moment at hand are measured as well as the movement of the hand. Activation of upper limb muscles and detailed kinematics of upper limb joints can be measured using commercial EMG and motion capture devices. Then, intermuscular coordination is analyzed by applying a non-negative matrix factorization (NNMF) algorithm (
KAIST upper limb synergy investigation system constrains the trajectory of subjects’ hands during reaching tasks since how similar motions they perform is important for fair comparison of intermuscular coordination across subjects. Specifically, KULSIS allows the subject to move their hand only along a linear actuator to control the direction of the reaching motion. Even though the location of the hand is constrained to the linear trajectory, the subjects still can adopt own motor strategies (i.e., different joint kinematics) to compensate for the difference in their limb sizes or weakness of specific muscles. The subjects’ hands are fastened to a handle connected to a sliding block of the linear actuator during the experiment. Despite restricting upper limb motions to the linear motion, the position and orientation of the linear actuator should be freely adjustable in a three-dimensional space to implement reaching motions toward varying directions and a variety of initial postures. In the case of isometric force generation tasks, the position of the linear actuator should be still adjustable to locate the handle freely in accordance with various upper limb postures. Among in total five degrees of freedom (DOFs) required in the system, three of them determine the position of the linear actuator, and the other two adjust its orientation.
A three-DOF gimbal structure was adopted to the handle to adjust its orientation depending on the task and the subject since the hand orientation would affect the activation of upper limb muscles. Besides, subjects with neural injuries usually have twisted hands and flexed wrist due to abnormal contraction of the upper limb muscles. A six-DOF load cell is used to measure the force and moment applied to the handle by the subjects.
Target specifications of KULSIS were set in three aspects; the maximum allowable force, workspace of the handle, and mechanical stiffness. First, the maximum allowable force was set to 30 kgf. Except the extreme posture of raising an arm straight up, neurologically intact young men could generate the maximum force of 260 N to push or pull a handle (
The position of the linear actuator in the sagittal plane and its three-dimensional orientation was determined by a four-DOF serial link (RRRR) mechanism (
The mediolateral position of the linear actuator was determined by sliding the chair along a linear track (
A ball-screw type linear motion module (RS-075N-Z05PR, Robostar, South Korea) was combined to a 100 W AC servo motor (APM-SA01ACN2, LS Mecapion, South Korea) to build the linear actuator. The motor includes an electromagnetic brake to fix the position of the handle. The maximum stroke of the module was 400 mm, which was a sufficient length to test the reach of the upper limb. The motor was controlled by a commercial motor driver (XSJ-230-06, Copley Controls, United States). The gross weight of the linear actuator including the load cell and the gimbaled handle was 10.3 kg. To reduce the gravitational load on the linear actuator positioner mechanism due to the weight of the links and the linear actuator, a gravity compensation mechanism based on passive springs was added to the RRRR mechanism. For the detail of gravity compensation, we recommend referring to Kim and Song’s study (
The three-DOF gimbal structure was adopted to adjust the orientation of the handle bar where subjects’ hands were fastened during the measurements (
Rotation (absolute value) of the linear actuator according to the directions of the loading (magnitude of 290 N) and the initial positions of the linear actuator.
Post. (X) | 0.01 | 0.66 | 0.02 | 0.01 | 1.16 | 0.03 | 0.04 | 1.28 | 0.01 |
Lat. (Y) | 0.77 | 0.06 | 1.60 | 0.91 | 0.29 | 2.35 | 1.26 | 0.33 | 2.47 |
Inf. (Z) | 0.10 | 1.57 | 0.17 | 0.02 | 1.78 | 0.06 | 0.09 | 1.62 | 0.04 |
To measure the force and torque generated by the subjects, a six-DOF load cell (Delta SI-660-60, ATI Industrial Automation, United States) was attached between the gimbaled handle and the sliding block of the linear actuator. The force and torque applied to the handle bar (
where
KULSIS implemented three types of reaching motions; reaching in a constant speed (isokinetic reaching), reaching by generating a constant force (isotonic reaching), and reaching with a minimal interaction force between the subject and the KULSIS at the handle (free reaching). First, the isokinetic reaching is to generate the maximal force while reaching to a target distance at constant speed. The load cell beneath the handle measures the force applied to the handle. The actuator moves the handle at speed proportional to the magnitude of the force along the target direction (
Second, during the isotonic reaching, the subject exerts a constant force toward the target force direction regardless of the reaching direction (for example, keeping exerting a force anteriorly even the hand moves posteriorly). The speed of the handle is determined from the difference between the force applied by the subject and the target force magnitude set by the experimenter (
The last condition, free reaching, can be implemented based on the same controller by setting the target force magnitude as zero.
A large workspace of the handle guarantees that KULSIS as a testbed to examine intermuscular coordination at a diversity of more upper limb postures. The two factors that determine the size the workspace include the area in the sagittal plane and the length in the mediolateral direction. The sagittal work area (the green contours in the sagittal view of
Workspace of the handle. The three green contours represent the workspace of the handle projected in each of three planes. In the sagittal view, the black dotted contour represents workspace of the linear actuator positioner. Gray lines represent the posture of the linear actuator positioner at each position. At the three positions marked as red circles, mechanical stiffness of the link structure was tested (see section “Mechanical Stiffness of the Linear Actuator Positioner”). Workspace of HapticMaster is represented as blue dotted contours for comparison in each of the three planes.
For reliable measurement, it is essential to maintain the position and orientation of the linear actuator. Deformation of the linear actuator positioner subject to an external force was measured by using a VICON motion capture system (VICON motion systems, Oxford, United Kingdom). According to the literature on accuracy of VICON system which used similar motion capture camera model and placement of the cameras compared to this study, mean position error of the stationary optical marker was 0.15 mm and variability of the position was lower than 0.025 mm (
The mechanical stiffness was calculated as the magnitude of the loading (i.e., 290 N) divided by the deviation in the position of the linear actuator along the direction of the loading. The position of the linear actuator always most deviated in the direction of the loading. The deviation was quantified as position change of the center point of the two markers attached to the third axes of the RRRR mechanism. In addition to the mechanical stiffness, the deviation of the orientation was calculated as the rotation of the coordinate frame for the linear actuator in the form of
The stiffness was in the range of 91.0∼1130 kN/m for the posterior direction, 32.6∼86.7 kN/m for the lateral direction and 102∼302 kN/m for the inferior direction depending on the initial position of the linear actuator. For the orientation of the linear actuator (see
RMS error values of the calibrated force and moment at different gimbal postures and different directions of loading.
1 | Posterior(X) | 6.95 | 1.81 | 34.57 | 0.17 | 1.21 | 0.13 |
Lateral(Y) | 1.31 | 4.09 | 3.96 | 0.68 | 0.10 | 0.16 | |
Inferior(Z) | 5.49 | 2.44 | 6.53 | 0.32 | 0.61 | 0.20 | |
2 | Posterior(X) | 6.29 | 3.78 | 28.18 | 0.39 | 1.21 | 0.17 |
Lateral(Y) | 4.71 | 4.68 | 2.83 | 0.66 | 0.48 | 0.12 | |
Inferior(Z) | 4.29 | 2.35 | 6.99 | 0.27 | 0.33 | 0.15 | |
3 | Posterior(X) | 12.08 | 1.38 | 12.49 | 0.15 | 1.30 | 0.39 |
Lateral(Y) | 6.16 | 3.81 | 3.85 | 0.46 | 0.66 | 0.27 | |
Inferior(Z) | 11.25 | 0.96 | 5.71 | 0.15 | 1.20 | 0.35 | |
Calibration (Gimbal posture 1∼3, loading up to 232 N) | 5.31 | 2.05 | 3.85 | 0.27 | 0.55 | 0.19 |
The accuracy of force/moment measurement through the gimbaled handle was evaluated. When a subject applies a force and/or a moment to the gimbaled handle, the force and moment are transmitted to the load cell through the handle. The force and moment applied at the handle are then reconstructed from the force and moment measured by the load cell using Eq. 1. While forces of known magnitudes and directions were applied to the handle bar at a given position, load cell signals were collected. The measured force and moment were compared to their actual values. Specifically, we applied a loading ranging from 29 N (i.e., one 3 kg weight) to 290 N (i.e., ten 3 kg weights) along the three directions (i.e., posteriorly, laterally and inferiorly) at the point of the handle bar corresponding to the rotational center of the gimbal mechanism. The weight(s) and the handle bar were connected by a cable which was routed by a pulley to control the direction of the loading. Three different postures of the gimbal were tested; (θ1, θ2) = (0 °, 0°), (θ1, θ2) = (45 °, 0°) and (θ1, θ2) = (45 °, 90°). To compensate the sources of measurement error that could be generated while mounting it on the device, the load cell was calibrated with respect to the actual loading and the actual moment based on the least square method. It was assumed that the loading was applied exactly along the posterior, lateral or inferior directions as well as that magnitude of the force applied to the handle was equal to the weight of the weights. Then, the actual moment was estimated by multiplying the magnitude of the loading and the distance between the load cell and the point of action of the loading. Note that, for the calibration, we excluded the data collected when more than eight weights (i.e., a loading over 232 N) were applied since we observed the distortion of the gimbal structure and rapid change of the load cell signals if the posterior loading exceeded 232 N. After the calibration, the root-mean-square (RMS) error between the calibrated force/moment and the actual loading/moment was quantified. In addition, how accurately the moment at the handle can be reconstructed using Eq. 1 was also evaluated. Since we applied only a pure force to the handle, the desired value of the reconstructed moment was zero. The accuracy of the reconstruction was quantified as the magnitude of the reconstructed moment.
Evaluation of force/moment measurement accuracy. Result at the first posture of the gimbaled handle is presented. Absolute force/moment values in each of three directions are shown.
Ten healthy subjects (five men and five women; age = 46.7 ± 3.97 years old; height = 169 ± 7.23 cm; and weight = 61.4 ± 10.5 kg) participated in this experiment, and their muscular activation was measured while performing isokinetic reaching tasks. EMG of twelve shoulder and elbow muscles (upper trapezius, lower trapezius, teres major, serratus anterior, clavicular fiber of pectoralis major, anterior deltoid, middle deltoid, posterior deltoid, triceps long head, triceps lateral head, biceps, and brachioradialis) was recorded using surface electrodes. For this experiment, we used a custom EMG measurement system (amplification gain, 1000; common mode rejection ratio, 120 dB; and bandwidth, 4000 Hz) based on the commercial differential amplifier (INA-128, Texas Instrument, United States). The subjects performed four gross motions of the upper limb (
Initial postures of the four upper limb motions.
The upper limb motions were selected from Wolf Motor Function Test (WMFT) (
Muscle synergies were analyzed from the EMG measurements. The surface EMG data were collected at 1000 Hz and processed in the following order; (1) low-pass filter with a cut-off frequency of 450 Hz, (2) band-rejection filter with cut-off frequencies of 55 Hz and 65 Hz to eliminate 60 Hz power noise, (3) high-pass filter with a cut-off frequency of 30 Hz, (4) rectification, and (5) low-pass filter with cut-off frequency of 0.3 Hz to obtain the amplitude of the EMG data. Based on the force and speed of the handle, the EMG intensity data were segmented into three phases; forward motion (i.e., from the initial posture to the final posture), rest at the final posture and reverse motion. The segmented EMG data were then resampled to have a uniform length across subjects and trials. Baseline EMG amplitude was subtracted. For each trial, the baseline amplitude was determined to be the smaller one between the average amplitude of the 2-s interval before the onset of the motion and the average amplitude of the 2-s interval of resting at the final posture. The processed EMG data were concatenated per subject. A NNMF algorithm (
Per subject, four to seven muscle synergies were identified. A total of 55 synergies were obtained from ten subjects, and the synergies were classified into nine clusters (
Nine clusters of upper limb muscle synergies. White bars represent muscle synergies identified from the data of individual subjects. Black bars represent the average of the muscle synergies (i.e., cluster centers). UT, upper trapezius; LT, lower trapezius; TM, teres major; SA, serratus anterior; PM, pectoralis major; MD, middle deltoid; AD, anterior deltoid; PD, posterior deltoid; Tr long, triceps long head; Tr lat, triceps lateral head; Bic, biceps; BRad, brachioradialis.
We developed KULSIS, a novel experimental setup for intermuscular coordination assessment of the upper extremity. It allows alignment of the force/moment-measuring handle to the hands in varying upper limb postures and straight reaching motions along a variety of three-dimensional directions. The specification of KULSIS was compared to that of the four comparable existing end-point type setups (
Specification of various experimental setups for upper limb intermuscular coordination evaluation.
Degrees of freedom | 2(active) | 2(active) | 3(active) | 6(active) | 6(active) | 1(active) + 5(passive) |
workspace | 400 mm, 760 mm (ellipse, 2D) | 381 mm, 457 mm | 400 mm (anteroposterior), 640 mm (mediolateral), 400 mm superoinferior)∗ | A sphere with a diameter of 2000 mm | 1600 mm (anteroposterior) 1400 mm (mediolateral), 2000 mm (superoinferior) | 740 mm (anteroposterior), 1000 mm (mediolateral), 1230 mm (superoinferior) |
Allowable maximum force | 58 N (Peak) | 45 N | 100 N (Nominal), 250 N (Peak) | 45 N | 178 N∗∗ | 290 N (linear actuator) 232 N (gimbaled handle) |
Mechanical stiffness | 16∼40 kN/m | Not reported | 10∼50 kN/m (depending on position of end-effector) | 1500 kN/m | 67∼80 kN/m | 91∼1130 kN/m (anteroposterior) 32.6∼86.7 kN/m (mediolateral) 102∼302 kN/m (superoinferior) (Linear actuator) |
There are three essential requirements in mechanical characteristics of the experimental setups for investigating intermuscular coordination patterns of human upper limb – the workspace, maximum allowable force, and mechanical stiffness. First, the workspace of KULSIS (i.e., 740 mm × 1200 mm × 1230 mm, in order of anteroposterior, mediolateral and superoinferior direction) covers most workspace of human upper limb in the sitting posture considering the length from the shoulder to the wrist, which is about 33% of the stature. The workspace of KULSIS is larger than that of HapticMaster (400 mm × 640 mm × 400 mm), which was reported as insufficient to cover workspace of the human upper limb (
Second, KULSIS has the largest maximum allowable force. KULSIS features a simple structure combining one active DOF (i.e., the linear actuator) to implement the straight reaching and five passive DOFs to set the position and the orientation of the linear actuator. The passive DOFs are fixed mechanically while the linear actuator implements various experimental tasks. The structural design allows KULSIS to maintain the position and the orientation of the linear actuator against the continuous external force up to 290 N, which can be generated by a healthy subject. Among the existing setups, MACARM can withstand the largest external force (i.e., 178 N) continuously. It was, however, reported that the end-point position of MACARM became unstable because of slackness of the cable mechanism at the larger portion of its workspace as the larger external force was exerted (
Last, the mechanical stiffness of KULSIS was similar or larger than those of the most existing setups. For the loading smaller than 290 N, KULSIS would allow much smaller deflection than the other devices. Among the three directions of loading, KULSIS is the most vulnerable to the lateral loading. However, KULSIS can still be utilized to implement upper limb motor tasks in the lateral direction, since the human also has the smallest force generation capacity in that direction.
This study also verified that the accuracy of force/moment measurement by the gimbaled handle of KULSIS was not affected by varying postures of the gimbal mechanism and the extreme loading condition up to 232 N. If the loading exceeded 232 N, however, the force/moment measurement was affected presumably by the deformation of the gimbal structure. Among the existing devices, only MACARM reported its force measurement performance under an external loading condition (
In addition to the evaluation results above, KULSIS is also capable of implementing various types of upper limb motor tasks under various biomechanical constraints such as isometric force generation, isokinetic reaching, isotonic reaching, and free reaching. This feature will allow identification of comprehensive intermuscular coordination patterns rather than that specific to limited tasks and will contribute to investigation on generalizability of the intermuscular coordination of human upper limb.
The averaged muscle synergies from clustering analysis were the elbow flexor, the elbow extensor, the shoulder flexor/adductor, the shoulder extensor/abductor and the combinations of the scapular muscles. The identified synergies verified that the neurologically intact subjects could control shoulder flexion/extension, shoulder abduction/adduction, and elbow flexion/extension separately. The composition of these patterns was similar to those reported in the previous literature which examined intermuscular coordination for isometric force generation tasks (
Limiting the upper limb motions to a straight translation of the hand along the linear actuator can be considered as a limitation of KULSIS. To compare intermuscular coordination of different subject populations, such as participants with stroke versus neurologically unimpaired individuals, it is inevitable to restrict the trajectory of the hand to minimize the discrepancy of the end-point motion between the groups in comparison. For the current version of KULSIS, a linear actuator was used. Upper limb tasks in daily living can be tested with KULSIS by approximating the tasks to the closest straight motions as we tested four upper limb tasks. To implement more natural upper limb tasks without the approximation to the linear trajectory, a two-DOF planar mechanism may be adopted for the advanced version of KULSIS. If so, the trajectory of the hand can be controlled by designing a force field to suppress the movement of the handle off the desired path.
The position and orientation of the linear actuator were maintained against the external loading up to 290 N, but the force/moment measurement was affected by the distortion of the gimbaled handle when the loading exceeded 232 N. Mechanical stiffness of the gimbaled handle will be improved for more accurate force measurement under the extreme loading condition. To suppress bending of the gimbal links, a harder material such as steel can be added to the gimbal links to increase its structural stiffness. Gimbal joint structures will be re-designed to improve their sturdiness. These improvements will cause an increase of inertia of the handle. The heavier inertia can affect the control performance of the linear actuator, especially in a free reaching task. If the control performance is degraded despite optimizing control gains, we can apply an additional sensor to detect the subject’s intention.
We propose KULSIS as a novel experimental setup for intermuscular coordination assessment in both isometric force generation and reaching tasks in the human upper extremity. Based on the quantitative, simultaneous measurement of motion and force as well as EMG, one can investigate how the alterations of the intermuscular coordination would affect their motor performance in the individuals with neural injuries. This paper mainly presents the design of KULSIS. Prior to the human study, its design was evaluated, and the parts that can be improved were identified. KULSIS features a large workspace which can cover that of the human upper limb and stable positioning of the linear actuator against the maximum loading of 290 N. The gimbaled handle structure will be modified to improve the accuracy of force/moment. Three different types of reaching tasks such as isokinetic, isotonic and free reaching as well as isometric force generation were developed for testing intermuscular coordination in diverse upper limb motor tasks. Overall, KULSIS can be used as an experimental setup for studying intermuscular coordination of human upper limb.
The datasets generated for this study are available on request to the corresponding author.
The studies involving human participants were reviewed and approved by the Institutional Review Board, Korea Advanced Institute of Science and Technology. The patients/participants provided their written informed consent to participate in this study.
J-HP, J-HS, JR, and H-SP conceived the concept of KULSIS. J-HP, HL, CP, and H-SP developed and evaluated the device, and carried out the experiment to identify muscle synergies of the healthy participants. J-HP and JR drafted the manuscript with inputs from all other authors. J-HS, JR, and H-SP contributed to the critical revision of the manuscript. HP supervised the study.
J-HP, HL, CP, J-HS, and H-SP are inventors of the patent application (KR10-2019-0022360, pending), filed on February 26, 2019, for the proposed design of the force measurement device. The remaining author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.