Design and kinematical performance analysis of the 7-DOF upper-limb exoskeleton toward improving human-robot interface in active and passive movement training

Abstract

BACKGROUND:

Upper-limb rehabilitation robots have become an important piece of equipment in stroke rehabilitation. The design of exoskeleton mechanisms plays a key role to improve human-robot interface in the upper-limb movements under passive and active rehabilitation training.

OBJECTIVE:

This paper proposes a novel of the 7-DOF (RR-RR-PRR) under-actuated exoskeleton mechanism based on the characteristics of the upper-limb movements in both of active and passive training. This aim of the proposed work is to improve human-robot interface in rehabilitation training with robots.

METHODS:

Firstly, the characteristics of active and passive movement training are analyzed depending on the human upper-limb model. Then, a novel 7-DOF (RR-RR-PRR) exoskeleton mechanism is proposed based on the analyzed characteristics. After that, kinematical performances of the proposed exoskeleton are analyzed on the workspace, manipulability and manipulability ellipsoid by compared with the common exoskeleton configuration of the 7 DOFs (RRR-R-PRR) mechanism. In the end, the prototype is manufactured and tested by undergoing the experiments of single-joint passive movement training and multi-joint active movement training. The human-robot interface of the proposed exoskeleton is demonstrated by root mean square error, Pearson correlation coefficient, and the time-delay difference.

RESULTS:

The results of the kinematical performance show that the effective workspace and the flexibility of the exoskeleton with the proposed configuration are increased by 10.44% and 1.7%. In the single-joint passive movement training experiment, the root mean square errors are 6.986, 7.568, 5.846, and Pearson correlation coefficients are 0.989, 0.984, 0.988 at the shoulder joint and the elbow joint, respectively. The time-delay differences are not beyond 3.1%. In the multi-joint active movement training experiment, the root mean square errors are 9.312 and 7.677, and Pearson correlation coefficients are 0.906 and 0.968 at the shoulder joint and the elbow joint, respectively. The time-delay differences are not beyond 3.28%.

CONCLUSIONS:

The proposed 7 DOFs exoskeleton mechanism shows uniformity with that of the common exoskeleton on the same rehabilitation trajectory which is effective to improve human-robot interface under passive and active rehabilitation training.

Keywords

Upper-limb rehabilitation exoskeleton active and passive movement training kinematical performance workspace manipulability

1. Introduction

Stroke is the leading cause of disability in adults, the number of stroke patients worldwide has increased dramatically in recent decades. Against such a background, more and more attention is paid to the quality of stroke patient’s life, especially the upper-limb dysfunction related to physical activity and daily life. For most of them, physical exercise training can restore their motor function [1]. Long-term intensive rehabilitation training will accelerate the recovery process of patients [2]. Compared with rehabilitation therapists, the upper-limb exoskeleton rehabilitation robots have been widely used in the rehabilitation training with the characteristics of high intensity, good repeatability and strong interaction [3].

The upper-limb exoskeleton rehabilitation robot is mainly divided into series type and parallel type according to the mechanical structure. The upper-limb exoskeleton with parallel mechanisms, such as SPM [4], BONES [5], ExoArm [6], the shoulder complex linkage mechanism [7], the 2-DOF parallel actuated exoskeleton [8] and MAHI EXO-II [9]. Compared with the series exoskeleton, the parallel exoskeletons have the advantages of compact structure and the capability of complex movement with fewer motors, but they are difficult to adjust the structure dimension and they also lack sufficient strength. Although the application of parallel mechanisms in the upper-limb rehabilitation brings new concepts of decreasing motors, it limits the kinematical performances in workspace and manipulability.

There are two main configurations of this series of exoskeleton robots. One type is an exoskeleton configuration (RRR-R, Here, R represents the active joint and the revolute pair. There are three rotation DOFs at the shoulder joint and one rotation DOF at the elbow joint. Each joint is designed as a functional module and link together to implement the functions of the whole upper-limb. The symbol ‘-’ presents the exoskeleton.) with three active joints at the shoulder joint and one movable joint at the elbow joint. This type exoskeleton configuration mainly includes ANYexo [10], Hommory [11], MGA [12], MEDARM [13], ARAMIS [14], ABLE [15] and IDEAS [16]. The above researches improve the DOF and movement performance of the exoskeleton by adding rotating joints or sliding joints on the basic configuration of RRR-R. The RRR-R configuration has a compact structure. It can move in any direction with 3 DOFs in the shoulder joint, so it is competitive to improve the human-robot interface. However, this structure generally requires at least four motors to control the exoskeleton, which increases the cost of the exoskeleton and makes the control strategy more complicated. The other type is a configuration (RR-R-R, Here, R represents the passive joint. There are two rotation DOFs at the shoulder joint, one rotation DOF at the forearm and one rotation DOF at the elbow joint.) with two active joints at the shoulder joint of the exoskeleton, one active joint at the forearm, and one active joint at the elbow joint of the exoskeleton. This type exoskeleton configuration mainly includes ARMin III [17], Co-Exos [18], iReHave [19], IntelliArm [20], LIMPACT [21], L-EXOS [22], SUEFUL-7 [23], MARSE-5 [24] and ARMOR [25]. Compared with the RRR-R configuration, this configuration significantly reduces the difficulty of the control system. In addition, the human-robot interface in active rehabilitation training is limited in these designs. According to the mechanical principle, the flexibility of this configuration is inferior to the RRR-R configuration. As can be seen from the above existing systems, more motors are used on many series of mechanisms according to the above requirements for upper-limb joints. The demand for more motors not only increases the cost, but also increases the difficulty of control.

In addition, rehabilitation therapy characteristics are important bases for the design of existing upper-limb rehabilitation robots [26, 27] due to the different characteristics of active movement training and passive movement training. The requirements of low cost, high kinematical performances and active and passive rehabilitation training should be considered in the optimal design of effective rehabilitation exoskeleton. Therefore, this paper proposes a novel of 7-DOF exoskeleton configuration based on the characteristics of the upper-limb movements under active and passive movement training for the stroke patients in order to reduce the costs of driven motors, improve human-robot interface.

2. Mechanical design

2.1 The characteristics analysis of active and passive movement training

The functional anatomy of the upper-limb is the base of designing an exoskeleton with the better human-robot interface performance. Recently, the kinematical of the upper-limb has been modeled by using several methods. In general, the shoulder, the elbow, the wrist joints and the forearm structure are included in the design of the upper-limb exoskeletons. Shoulder joint is often modeled as a spherical joint with 3 DOFs to implement the movements of flexion/extension, abduction/adduction, and the internal/external rotation [28]. The elbow joint is commonly modeled as a revolute joint with 1 DOF to perform the movement of flexion/extension. Wrist joint is also considered as a universal joint with 2 DOFs to perform the flexion/extension and abduction/adduction movements. The forearm structure is simplified as a revolute joint with 1 DOF to perform the pronation/supination movements.

For 95% of the population, the range of motion (ROM) required for activities of daily living is derived from the maximum values of the related work [29, 30, 31] and translated to the ISB system [32] as shown in Table 1. The wrist joint has little effect on the workspace of the human upper-limb, so it needs to maintain specific physiological position.

Table 1
The ROM of the upper-limb joints in the ISB system

DOF	ROM ( ${}^{\circ}$ )
Plane of elevation	0–115
Angle of elevation	0–130
Internal/external rotation	$-$ 90–60
Elbow flexion/extension	0–120

According Zhu et al. [33], therapeutic training can promote the recovery of athletic ability and learning ability for stroke patients. Therapeutic training includes two kinds of training approaches. They are active movement training and passive movement training. Active movement training is the training that the patient actively completes in the form of muscle contraction, which requires neither assistance nor overcoming external resistance during training. Passive movement training refers to the help of external forces, such as the strength of other people, patients’ healthy limbs or equipment, to help muscles and joints to restore their functions. However, active and passive movement training in the procedure of rehabilitation has shown different movement characteristics with the robots’ assistance. In the passive movement training, the robots will drive the patients’ arm to implement the specific movements combined the shoulder joint and the elbow joint. The DOFs on flexion/extension and abduction/adduction of the shoulder joint, and the DOF on the flexion/extension of the elbow joint should be driven strictly by the robot because of the weak muscles. For instance, as shown in Fig. 1a, the positions and the direction of the blue arrows indicate that the external forces produced by the robot to drive the human joints. In the active movement training, the robots will follow the patients’ arm movements. The DOFs on the shoulder joint and the elbow joint should be free in case of restricted movements. Therefore, there are three DOFs on the shoulder joints, one DOF on the elbow joint, one DOF on the forearm and three DOFs on the wrist joint as shown in Fig. 1a. Unfortunately, there are a lot of patients who need to do the active movement training but do not have enough muscle strength. The robot exoskeleton should combine the passive movement training characteristics and the active movement training characteristics simultaneously. The characteristics on driving and the free DOFs should be optimized strictly in the same robot exoskeleton. Such this, the number of the active DOF and the number of the passive DOF for an exoskeleton should be discussed before its configuration design. The more numbers of the active DOFs the more movement limitations. The fewer active DOFs the more movement freedom and the more dangerous in rehabilitation training. On the other side, the more numbers of the active DOFs the more cost of the exoskeleton. This paper discusses a 7-DOF exoskeleton with 2 active DOFs based on the characteristics of the active movement training and the passive movement training. Therefore, as shown in Fig. 1b, the two active DOFs (the green rotation arrows) are placed on the flexion/extension of the shoulder joint and the elbow joint in order to implement the passive movement training. The reason for the choice is that the movements of flexion/extension of the shoulder joint and the elbow joint are main loads of the human arm than the others [34, 35]. The other DOFs (the blue rotation arrows) of the human arm are set to be the passive DOF in order to improve the freedom in active movement training procedure.

Figure 1.

The human upper-limb model. (a) The 7 DOFs of the model. (b) The active and passive DOFs of the model.

The International Society for Biomechanics (ISB) recommends that the three DOFs at the shoulder joint are presented as three serial rotations around the vertical axis, they are including rotation of $z$ axis (plane of elevation), $\theta_{1}$ , $y$ axis (angle of elevation), $\theta_{2}$ , and $x$ axis (internal/external rotation), $\theta_{3}$ . The DOF at the elbow joint is presented as around $z$ , $\theta_{4}$ is utilized to describe the rotation angle. In addition, the wrist joint can be directly modeled without 3 DOFs in this paper. Figure 2 shows the equivalent model of the human upper-limb, where {s}, {e}, {f} {w} present the coordinate systems of shoulder, elbow, forearm, and wrist joints, respectively.

Figure 2.

The configuration (a) Common configuration with RRR-R at shoulder and elbow joints. (b) New configuration with RR-RR at shoulder and elbow joints.

2.2 Structure design of the 7-DOF upper-limb exoskeleton

The structure configuration is important to design the exoskeleton mechanism because the kinematical performance should inflect the human-robot interface. Traditionally, the 7-DOF upper-limb exoskeleton is designed in detail on the structure of shoulder and elbow joint. As what mentioned above, there are 3 DOFs on the shoulder joint that is often designed with RRR configuration, and there is 1 DOF on the elbow joint that is often designed with R. Therefore, the RRR-R configuration is commonly established to analyze how to control the shoulder and elbow joints of the human upper-limb, as shown in Fig. 2a. $\underline{R}_{C1}$ , $\underline{R}_{C2}$ , and $\underline{R}_{C3}$ are three active joints that assist the flexion/extension, abduction/adduction, and internal/external rotation movement of the human shoulder joint. $\underline{R}_{C4}$ is an active joint that assist the flexion/extension movement of the human elbow joint. The active movement of the three DOFs at the exoskeleton shoulder joint can drive the human shoulder joint to move in any direction. The configuration can naturally meet the characteristics of rehabilitation training. However, it requires a more complex control strategy more motors and sensors. Rehabilitation training principles are methods to maintain and regain function or prevent secondary loss of function by using mechanical factors to carry out targeted treatment and training for patients with motor dysfunction. As mentioned above, according to the principles of rehabilitation training, it is unnecessary for the exoskeleton to completely control the function of human upper-limb for training [33].

Based on the previous analysis, a new RR-RR-PRR (Here, P presents prismatic pair. There are two rotation DOFs at the shoulder joint, two rotation DOFs at the elbow joint, and one prismatic DOF and two rotation DOFs at the wrist joint.) configuration of the exoskeleton combining active and passive joints is proposed, as shown in Fig. 2b. $\underline{R}_{N3}$ is the active joint that drives the flexion and extension of the shoulder joint. By rotating the exoskeleton around $R_{N2}$ to 90 ${}^{\circ}$ , $\underline{R}_{N3}$ can be switched to drive adduction and abduction of the shoulder joint. $\underline{R}_{N1}$ is the active joint to drive the elbow joint to extend. In active rehabilitation training, the exoskeleton should satisfy the movement requirements of the human upper-limb as much as possible. Because the internal and external rotation of the shoulder joint is used for the flexion and extension of the elbow in a different direction, the elbow joint can be equivalent to a rotating pair with a rotatable axis. Therefore, to better match the movement of the human elbow joint, a passive joint $R_{N1}$ is added perpendicularly to $\underline{R}_{N4}$ at the elbow joint of the exoskeleton. The forearm and wrist joints of patients can move in any direction, passive joints meet the requirements of the movement are set at the corresponding positions of the exoskeleton. $P_{C5}$ is a sliding pair that compensating the axis deviation between the human elbow joint and the mechanical elbow joint. $R_{C6}$ and $R_{C7}$ are combined to satisfy the movement requirements of the forearm and wrist joints. Compared with the existing exoskeleton that requires at least four motors, the proposed 7-DOF configuration (RR-RR-PRR) with only 2 active DOFs controlled by 2 motors can meet the requirements of rehabilitation training. Therefore, this mechanism is an underactuation. The exoskeleton cannot be completely controlled. In the passive movement training procedure, the DOFs of adduction/abduction, the internal/external rotation on the shoulder joint, and the DOFs on the forearm and the wrist joint will be fixed to ensure the passive movement. But in the active movement training procedure, all of the passive DOFs are released because the human arm can provide the external active DOFs. In addition, the shoulder joint and the elbow joint of the exoskeleton are composed of an active joint and a passive joint, which will make the mechanism compact.

3. Kinematical performance analysis of the 7-DOF upper-limb exoskeletons

In this work, three kinematical performance indexes, which are the workspace, manipulability and manipulability ellipsoid, are chosen to assess the human-robot interface for the proposed exoskeleton. The rationality of the design structure is verified by comparing the performance of the RR-RR-PRR structure and the RRR-RR-PRR structure. Since the configuration of the wrist joint has little effects on the kinematical performance of the exoskeleton, the reference configuration is designated as RRR-R-PRR, as shown in Fig. 2a. For the convenience of description in the following paper, the RRR-R-PRR and the RR-RR-PRR are represented by Co-C (common configuration) and Co-N (new proposed configuration) (Here, Co represents configuration, C represents common while N represents new proposed), respectively.

3.1 Workspace analysis

The workspace refers to the space formed by the points of all positions that the robot’s end-effector can reach in the basic coordinate system. The workspace of the human upper-limb and two configurations are established to make the movement space of the exoskeleton satisfy the movement space of the human upper-limb as much as possible. Firstly, in the ISB system, the joints of the exoskeleton are restricted by the DOFs of the upper-limb joints to ensure safety when using the exoskeleton, as shown in Table 2. Secondly, the joints and the two configuration angles of the upper-limb are established using the Monte Carlo method [36]. Finally, the angles are substituted into the kinematic equation to obtain the workspace of the upper-limb and the two configurations, as shown in Table 3. The red space in the table represents the workspace of the upper-limb. The blue space and the yellow space are the workspace of Co-C and Co-N. It can be seen that, compared with the workspace of Co-C, the workspace of Co-N has a higher degree of coincidence with the upper-limb workspace.

Table 2
The angle of the upper-limb and exoskeleton joints

ROM	$\theta_{1}$ ( ${}^{\circ}$ )		$\theta_{2}$ ( ${}^{\circ}$ )		$\theta_{3}$ ( ${}^{\circ}$ )		$\theta_{4}$ ( ${}^{\circ}$ )
The upper-limb	$-$ 115	–0	0	–130	$-$ 90	–60	$-$ 120	–0
The Co-C	0	–130	$-$ 90	–25	0	–130	0	–120
The Co-N	$-$ 90	–25	0	–130	0	–120	0	–120

Table 3

The workspace of the upper-limb and the two configurations

To quantify the differences, the workspace of the two different configurations are filtered through the boundaries of the human upper-limb workspace. The result is that the effective position points (the workspaces of the two configurations for the human upper-limb can reach the total workspace) of Co-C are 290119 and the effective position points of Co-N are 273536. Summation points of per point cloud are 400000. Therefore, the percentage of the effective workspace of Co-C is 75.52% and the percentage of the effective workspace of Co-N is 68.38%. It shows that the proposed configuration increases the effective workspace by 10.44%.

3.2 Manipulability analysis

The workspace of the exoskeleton reflects the kinematical performance of the rehabilitation robot to a certain extent. However, due to the high flexibility of the upper-limb, the exoskeletons need to be flexible enough to improve the effect of the rehabilitation. Manipulability means the ability of the exoskeleton to change its position and orientation, that is considered to be a quantitative and performance measure of configuration selection of rehabilitation robots [37]. Manipulability analysis of Co-C and Co-N is performed as follows.

3.2.1 Jacobian matrix

Generally, the input of the Jacobian matrix can be checked for operability. For a revolute joint, we calculate that by

$\displaystyle J_{i}=[{(p\times n)_{z}(p\times o)_{z}(p\times a)_{z}n_{z}o_{z}a% _{z}}]^{T}$ (1)

The Jacobian matrix of a prismatic joint can be computed by

$\displaystyle J_{i}=[{n_{z}o_{z}a_{z}000}]^{T}$ (2)

Where, $n_{z}$ , $o_{z}$ , $a_{z}$ and $p$ are the four-column vectors of the matrix ${}_{7}^{i}A$ . From the homogeneous transformation matrix between adjacent links, ${}_{7}^{6}A,{}_{7}^{5}A,{}_{7}^{4}A,{}_{7}^{3}A,{}_{7}^{2}A,{}_{7}^{1}A,{}_{7}% ^{0}A$ can be obtained. The, $n_{z}$ , $o_{z}$ , $a_{z}$ and $p$ of ${}_{7}^{i}A$ are substituted into Eqs (1) and (2), the Jacobian matrix $J_{1}=[J_{1,1}J_{1,2}J_{1,3}J_{1,4}J_{1,5}J_{1,6}J_{1,7}]$ and $J_{2}=[J_{2,1}J_{2,2}J_{2,3}J_{2,4}J_{2,5}J_{2,6}J_{2,7}]$ are obtained.

3.2.2 Manipulability measure

The manipulability measure [37, 38], defined as

$\displaystyle\omega=(\det[J(q)J(q)^{T}])^{1/2}=(\lambda_{1}\lambda_{2}\cdots% \lambda_{m})^{1/2}=\sigma_{1}\sigma_{2}\cdots\sigma_{m}$ (3)

The $\sigma_{i}$ is the singular value of Jacobian matrix, the $\lambda_{i}$ is the eigenvalue of the $J(q)J(q)$ , and m is the dimension of operation space, $i=1,2,\cdots,m$ .

Due to the difference between the manipulability peaks of different robots, for the convenience of comparison, the relative manipulability is defined as

$\displaystyle\mu_{i}=\omega_{i}/\omega_{\max}(i=1,2,\cdots,n)$ (4)

Where $\omega_{i}$ is the manipulability of the coordinate point, $\omega_{\max}$ is the maximum value of the configuration, the value range of $\mu_{i}$ is 0–1. Generally speaking, the higher $\mu_{i}$ , the better the dexterous, while the lower $\mu_{i}$ the operation is limited.

To compare the manipulability of Co-C and Co-N, the values of $\mu$ are divided into five categories. As the terminal posture points beyond the workspace of the human upper-limb are meaningless for analysis, the filtered points of the workspace of the human upper-limb are used to Eqs (3) and (4). Table 4 is the specific statistical results.

Table 4

The details of the subject

Age	Sex	Height	Upper arm length	Forearm length	Hand length
26 year old	Male	176 mm	35 mm	26.7 mm	19.1 mm

The area with $\mu$ above 0.6 is the area where the robot is more flexible. The operability percentage with Co-C above 0.6 is 94.03% and that with Co-N above 0.6 is 95.63%. It shows that the proposed configuration improves flexibility by 1.7%.

During upper-limb rehabilitation training, the probability that the end of the exoskeleton reaches the edge of the workspace is less than that inside of the workspace. Therefore, the figures of relative manipulability combined with workspace are plotted separately to compare the level of flexibility in the whole domain, as shown in Fig. 3. The color of the workspace is closer to red, the value of $\mu_{i}$ is closer to 1, the flexibility of this point is better.

Figure 3.

The figures of relative manipulability combined with the workspace. (a) Co-C. (b) Co-N.

As shown in Fig. 3, the operability in both configurations are situated at the edge of the workspace, with little impact on global flexibility. However, global $\mu_{i}$ value of Co-N is greater than that of the Co-C, indicating that the flexibility of Co-N is better than that of Co-C.

3.3 Manipulability ellipsoid analysis

Circular motion is common during rehabilitation exercises, and multiple joints can be trained at the same time. Based on the kinematical model of human upper-limb, an arc trajectory is established and divided into 10 sample points. According to the Newton-Raphson method [39], the inverse kinematical solutions of two joint configurations are obtained, and the kinematical performance of two upper-limb rehabilitation robots on the arc is analyzed.

3.3.1 The joint angle values

The joint trajectory of the upper-limb is set between two joint configurations. The initial joint position is set to $\theta_{1}=-90^{\circ}$ , $\theta_{2}=90^{\circ}$ , $\theta_{3}=90^{\circ}$ , $\theta_{4}=-90^{\circ}$ , while the final joint position is set to $\theta_{1}=0^{\circ}$ , $\theta_{2}=0^{\circ}$ , $\theta_{3}=90^{\circ}$ , $\theta_{4}=0^{\circ}$ . The joint trajectory is generated by MATLAB and divided into 10 sample points. The angular position of the exoskeleton joint is calculated by obtaining the Cartesian coordinates of the sample points. According to the obtained coordinates, the inverse kinematical solutions of the two configurations are obtained by the Newton-Raphson method.

Figure 4.

Manipulability ellipsoids for Co-C and Co-N.

3.3.2 Manipulability ellipsoid

The manipulability ellipsoid is a geometric visualization tool that determines in which directions and to what extent the robot’s end-of-motion ability will be weakened. It can be used to determine the easiest and most difficult specific orientation of the robot’s movement. The volume of the maneuverability ellipsoid is proportional to the maneuverability index, which can indicate the degree of maneuverability in the visual operation space, that is $V\propto(\det[J(q)J^{T}(q)])^{1/2}=(\lambda_{1}\lambda_{2}\cdots\lambda_{m})^{% 1/2}$ [40, 41]. The manipulability ellipsoids are constructed based on the Jacobian matrix and its singular values. In addition to its geometric characteristics, two additional parameters are defined to measure the difficulty of the exoskeleton moving in a giving position.

The ratio of the shortest semi-axis to the longest semi-axis of the manipulability ellipsoid P describes the kinematical performance of the exoskeleton in all directions. When the ratio P is closer to 1, the exoskeleton is easy to move in any direction, the robot is isotropic. Therefore,

$\displaystyle P=(\lambda_{\min}(J(q)J^{T}(q)))^{1/2}/(\lambda_{\max}(J(q)J^{T}% (q)))^{1/2}\leqslant 1$ (5)

The manipulability ellipsoid volume can directly represent the kinematical performance of the exoskeleton in a certain posture. The volume is larger that represents the performance is better. Therefore,

$\displaystyle V=(\det[J(q)J^{T}(q)])^{1/2}=(\lambda_{1}\lambda_{2}\cdots% \lambda_{m})^{1/2}$ (6)

The manipulability ellipsoids corresponding to each step of the joint trajectory are obtained by MATLAB, as shown in Fig. 4. The manipulability ellipsoids of these two configurations are drawn on the corresponding sample joints. The size of the ellipsoids is reduced by 25 times for the convenience of expression. It is found that the values of P and V of Co-C and Co-N are close as a whole. The values of Co-N are larger than that of Co-C in the early stage of the movement, and the results are opposite in the later stage of the movement. The P values of Co-N are much larger than 0, which means the exoskeleton cannot reach the singularity. In summary, the kinematical performance of Co-N is roughly the same as that of Co-C, which proves the rationality of Co-N.

4. Performance analysis on human-robot interface in active and passive movement training

4.1 Implementation of the prototype

The structure of the upper-limb exoskeleton robot verified above is designed in detail in this section. The robot prototype consists of a base, a cantilever beam, and an exoskeleton body as shown in Fig. 8. The entire exoskeleton connection cantilever is installed in front of the driving lifting column at the base to accommodate different height. A certain range of sliding can be carried out from the base to adapt to different shoulder width and the switch between left-arm and right-arm. The robot is used with the left arm as shown in Fig. 5.

Figure 5.

The system of the upper-limb rehabilitation robot.

The $R_{1}$ , $R_{2}$ , $R_{3}$ , $R_{4}$ , $T_{5}$ , $R_{6}$ , and $R_{7}$ are motor joint arranged in series. The EC-motor power in joint $R_{2}$ is transmitted to the output shaft through a harmonic reducer. The torque sensor is mounted between the reducer and the shaft to monitor the output torque. An absolute angular encoder is fixed on the output shaft to measure the position of $R_{2}R_{4}$ and $R_{2}$ are similar in structure. $T_{5}$ is a sliding joint that can slide on the slide rail built into the exoskeleton forearm to meet the joint asymmetry in rehabilitation training. $R_{1}$ , $R_{3}$ , $R_{6}$ and $R_{7}$ are passive joints with absolute encoders installed on the shaft. The position information collected by all encoders is used to control the virtual reality game.

4.2 Experiments

A series of experiments are conducted on a 26-year-old healthy male subject to evaluate the ability of the active and passive rehabilitation training on single-joint movement and multi-joint movement [42], the details of the subject as shown in Table 4. The study is approved by the ethical review board of Shanghai University of Medicine and Health Sciences (Study No. 2019-ZYXM1-04–420300197109053525). The system can measure the ROM of the shoulder joint, the elbow joint and the wrist joint of a person who sits on a common chair playing a virtual reality game. We set up seven sensors on the human upper-limb to accurately measure the joint angles: two inertial sensors located at the upside and downside of backbone, and five inertial sensors located at the upper-limb (shoulder, upper arm, forearm, palm, and hand), respectively. All of the sensors in this system can measure the angles in x-, y- and z-axis. The flexion/extension and adduction/abduction of the shoulder joint, the flexion/extension of the elbow joint and the flexion/extension and internal rotation/external rotation of the wrist joint are performed respectively and joint angles are collected in real time. We define that the cycle of motion is between the attainable maximum ROM of the corresponding DOFs, for instance, the flexion and the extension. Each couple motion has been done for 50 cycles. The accelerate and decelerate cycles are not included in the 50 cycles due to instability problems.

4.2.1 Experiments of single-joint movement

The single-joint exercise test is designed to verify the ability of the passive rehabilitation training. Setting three groups of experiments to test the single-joint motion characteristics between the passive motion of the human arm and the active motion of the exoskeleton arm, as shown in Figs 6–8. These test motions include flexion/extension of the shoulder, adduction/abduction of the shoulder, and flexion/extension of the elbow. The motion analysis system tests the motion angles of the human arm in real-time. Two groups of motion angles of each group experiment are compared together. The meaning of the root mean square error and Pearson correlation coefficient are utilized to describe the difference of the rotation angles between the human and the exoskeleton at the same temporal points. It is proved that the root mean square error and Pearson correlation coefficient are also being utilized to verify the human-robot interface [43, 44, 45]. The smaller of the root mean square the better human-robot interface. The value of Pearson correlation coefficient is closer to 1, the human-robot interface is better. Therefore, this paper utilizes the root mean square error and Pearson correlation coefficient to verify the human-robot interface.

Figure 6.

Single-joint movement experiment for shoulder flexion/extension movement.

Figure 7.

Single-joint movement experiment for shoulder abduction/adduction movement.

Figure 8.

Single-joint movement experiment for elbow flexion/extension movement.

Root mean square error and Pearson correlation coefficient are used to analyze the motion error between the human arm and exoskeleton, as shown in Eqs (7) and (8).

$\displaystyle\delta=\left(1/N\sum_{t=1}^{N}{(\gamma_{t}-\beta_{t})^{2}}\right)% ^{1/2}$ (7)

Where, $\delta$ – root mean square error, $\gamma$ – motor movement angle, $\beta$ – human angle, N-data bulk.

$\displaystyle P=\left(\sum_{t=1}^{N}{(\gamma_{t}-\overline{\gamma})(\beta_{t}-% \overline{\beta})}\right)\left/\left(\sum_{t=1}^{N}{(\gamma_{t}-\overline{% \gamma})}^{2}\times\sum_{t=1}^{N}{(\beta_{t}-\overline{\beta})}^{2}\right)\right.$ (8)

Where, $P$ – Pearson correlation coefficient, $\bar{\gamma}$ – average value of the motor movement angle, $\bar{\beta}$ – average value of the angle. The single-joint experiment about shoulder flexion and extension movement is shown in Fig. 6. The red dotted line denotes the real-time angle values obtained by the pose sensor at the shoulder of the exoskeleton and the blue solid line denotes the real-time angle values obtained by the motion analysis system. The $\delta$ is 6.986 and the P is 0.989. The single-joint experiment of shoulder adduction and abduction movement is shown in Fig. 7. The $\delta$ is 7.568 and the P is 0.984. The single-joint experiment about elbow flexion and extension movement is shown in Fig. 8. The $\delta$ is 5.846 and the P is 0.988. The movement of the human joints is completely driven by the exoskeleton, so the movement curve of the exoskeleton precedes that of the human body. However, the movement curves of the human upper-limb are close to the ones of the exoskeleton. The time-delay differences between both curves are not beyond 3.1%. Therefore, the results show the good human-robot interface during the movement that the exoskeleton can effectively drive the human joints to perform single-joint rehabilitation exercises in the passive movement training.

Figure 9.

The experiments of multi-joint movement (a) The game interface of the multi-joint rehabilitation training. (b) Analysis of the cooperative capability of the exoskeleton for flexion/extension of the shoulder. (c) Analysis of the cooperative capability of the exoskeleton for flexion/extension of the elbow.

4.2.2 Experiments of multi-joint movement

The multi-joint exercise test is designed to verify the ability of active movement training of the proposed upper-limb exoskeleton. The healthy male subject as mentioned above wears the exoskeleton to play a virtual reality game, as shown in Fig. 9a. In this model, the user performs an active joint movement with the help of the exoskeleton. The human upper-limb are assisted by the exoskeleton to resist gravity during the movement, and the bird on the screen is driven by the angle change signal of the exoskeleton joints to eat randomly generated gold coins. The white curve is the partial trajectory of the motion. This process records the changes in the angles of the exoskeleton and upper-limb joints. The flexion and extension movement of the shoulder joints of the exoskeleton and the human upper-limb are shown in Fig. 9b. The elbow joint movement curve is shown in Fig. 9c. The blue line is the real-time angle values obtained by the shoulder pose sensor and the red line is the real-time angle values obtained by the motion analysis system. The $\delta$ and P at the shoulder joint are 9.312 and 0.906 and those at the elbow joint are 7.677 and 0.968. According to the results of calculation, the change of the exoskeleton joint angle values is synchronous with the change of the upper-limb joint angle values in the multi-joint training. The time-delay differences between both the curves are not beyond 3.28%. Since the trajectory of the human upper-limb is random, the synchronization of the angles of the exoskeleton and the human upper-limb shows that the exoskeleton can follow the human upper-limb well in the active movement training phase. It proves the excellent manipulability of the exoskeleton in the multi-joint training.

5. Discussion

The 7-DOF configuration with 2 active DOFs proposed in this paper is kind of configuration is an underactuation, and it cannot be completely controlled. In order to improve the human-robot interface, this paper proposes a novel configuration (RR-RR-PRR) based on the specified DOFs by synthesizing the characteristics of active movement training and passive movement training. As for the novel configuration, the DOFs of adduction/abduction, the internal/external rotation on the shoulder joint, and the DOFs on the forearm and the wrist joint are fixed to ensure the passive movement during the passive movement training procedure However, all of the passive DOFs are released because the human arm can provide the external actives DOFs in the active movement training procedure Under the restriction of this condition, fewer motors should promote the economy of the proposed exoskeleton. However, the underactuation characteristics of the exoskeleton will bring some disadvantages in motion control, such as the workspace, the manipulability. The workspace of the exoskeleton should keep the same one as a human being. Unfortunately, most exoskeleton (e.g. [4, 5, 6]) could not reach the human beings because their configuration is limited. This paper analyzes this workspace and manipulability ellipsoid by comparing with the common configuration RRR-R-PRR. It is noted that the workspace of the proposed configuration is increased by 10.44% even under the same constriction conditions [15, 16]. On the other hand, the same workspace is achieved by increasing the number of active degrees of freedom. Thus, the economic characteristics of the exoskeleton will be affected (e.g. [10, 12]). The results also indicate that the configuration could improve the man-robot interface on the aspect of workspace. In addition, the human-robot interface can be expressed by the movement characteristics, i.e. the manipulability or manipulability ellipsoid. According to the computing results shown in Section 4, it is found that the values of P and V are close as a whole. It indicates that the exoskeleton can follow the movement of human arm in active movement training and drive the human arm fluently than the common configuration that owns the same active DOFs. In operating experiments of the prototype, the single-joint movement experiment proves the end effector trail of the prototype has been following the human arm when they are moving together. The time-delay differences between both curves are not beyond 3.1%. The man-robot interface of 7-DOF exoskeleton can be improved by this design. However, it is obvious that the increasing number of DOF can have a better interface (e.g. Hommory [11]). Another solution would be to improve the man-robot interface. Unfortunately, the number of active DOF could be increased in order to implement it. Therefore, it is one solution of changing the position of DOF under the same active DOF. The proposed exoskeleton can be applied to help patients complete specific training tasks in active or passive movement training, so as to achieve clinical rehabilitation of patients with upper-limb dysfunction or functional limitations. It is valuable to verify which DOF can be changed in case to improve better man-robot interface in the design of rehabilitation exoskeletons.

6. Conclusions

In this paper, a novel type of exoskeleton with RR-RR-PRR configuration for stroke patients with upper-limb dysfunction is proposed in passive and active movement rehabilitation training towards improving the human-robot interface. The number of DOF of the exoskeleton is 7, while the number of the active DOF is 2. Therefore, this mechanism is an underactuation. In order to verify the rationality of the proposed configuration, the kinematical performance parameters of the configuration including workspace, relative manipulability and manipulability ellipsoid are analyzed and compared with those of the RRR-R configuration to evaluate the kinematical performance of the proposed configuration. The effective workspace and the flexibility of the exoskeleton with the RR-RR-PRR configuration are increased by 10.44% and 1.7%. Finally, a 7-DOF exoskeleton rehabilitation robot for upper-limb is designed according to the RR-RR-PRR configuration. Two sets of the experiments of single-joint passive movement training and multi-joint active movement training have been proceeded in order to show the human-robot interface of the proposed exoskeleton. In the single-joint passive movement training experiment, the time-delay differences are not beyond 3.1%. In the multi-joint active movement training experiment, the time-delay differences are not beyond 3.28%. In addition, safety plays a key role in the design of rehabilitation robots. For patients, it is necessary to avoid any risks. Kinematical performance analysis and the experiments have also verified the safety of the exoskeleton [46, 47]. The experiments also show the proposed upper-limb exoskeleton has good human-robot interface during active and passive movement training.

Footnotes

Conflict of interest

None to report.

References

Bejot

Daubail

Giroud

. Epidemiology of stroke and transient ischemic attacks: Current knowledge and perspectives. Rev Neurol-France. 2016; 172(1): 59-68. doi: 10.1016/j.neurol.2015.07.013.

Bertani

Melegari

De Cola

Bramanti

Calabro

. Effects of robot-assisted upper limb rehabilitation in stroke patients: A systematic review with meta-analysis. Neurol Sci. 2017; 38(9): 1561-9. doi: 10.1007/s10072-017-2995-5.

Manna

Dubey

. A mechanism for elbow exoskeleton for customised training. Int C Rehab Robot. 2017; 1597-602. doi: 10.1109/ICORR.2017.8009476.

Hunt

Lee

Artemiadis

. A novel shoulder exoskeleton robot using parallel actuation and a passive slip interface. J Mech Robot. 2017; 9(1). doi: 10.1115/1.4035087.

Klein

Spencer

Allington

Bobrow

Reinkensmeyer

. Optimization of a parallel shoulder mechanism to achieve a high-force, low-mass, robotic-arm exoskeleton. Ieee T Robot. 2010; 26(4): 710-5. doi: 10.1109/Tro.2010.2052170.

Herbin

Pajor

. Human-robot cooperative control system based on serial elastic actuator bowden cable drive in ExoArm 7-DOF upper extremity exoskeleton. Mechanism and Machine Theory. 2021; 163. doi: 10.1016/j.mechmachtheory.2021.104372.

Ikemoto

Kimoto

Hosoda

. Shoulder complex linkage mechanism for humanlike musculoskeletal robot arms. Bioinspir Biomim. 2015; 10(6). doi: 10.1088/1748-3190/10/6/066009.

Hsieh

Chen

Chien

Lan

. Design of a parallel actuated exoskeleton for adaptive and safe robotic shoulder rehabilitation. Ieee-Asme T Mech. 2017; 22(5): 2034-45. doi: 10.1109/Tmech.2017.2717874.

JFrench

JARC

Malley

MKO

. System characterization of MAHI Exo-II: A robotic exoskeleton for upper extremity rehabilitation. Proc ASME Dyn Syst Control Conf. 2016; 2014: 1-5. doi: 10.1115/DSCC2014-6267.

10.

Zimmermann

Forino

Riener

Hutter

. ANYexo: A versatile and dynamic upper-limb rehabilitation robot. Ieee Robot Autom Let. 2019; 4(4): 3649-56. doi: 10.1109/Lra.2019.2926958.

11.

Kim

Deshpande

. An upper-body rehabilitation exoskeleton harmony with an anatomical shoulder mechanism: Design, modeling, control, and performance evaluation. Int J Robot Res. 2017; 36(4): 414-35. doi: 10.1177/0278364917706743.

12.

Carignan

Tang

Roderick

. Development of an exoskeleton haptic interface for virtual task training. 2009 Ieee-Rsj International Conference on Intelligent Robots and Systems. 2009: pp. 3697-702. doi: 10.1109/Iros.2009.5354834.

13.

Ball

. MEDARM: A rehabilitation robot with 5-DOF at the shoulder complex. 2007 IEEE/ASME International Conference on Advanced Intelligent Mechatronics. 2007: pp. 251.

14.

Colizzi

Lidonnici

Pignolo

. The aramis project: A concept robot and technical design. J Rehabil Med. 2009; 41(12): 1011-5. doi: 10.2340/16501977-0407.

15.

Garrec

. ABLE, an innovative transparent exoskeleton for the upper-limb. 2008 Ieee/Rsj International Conference on Robots and Intelligent Systems, Vols 1–3, Conference Proceedings. 2008: pp. 1483-8. doi: 10.1109/Iros.2008.4651012.

16.

Carignan

Liszka

. Design of an arm exoskeleton with scapula motion for shoulder rehabilitation. 2005 12th International Conference on Advanced Robotics. 2005: pp. 524-31.

17.

Nef

TGM

Riener

. ARMin III – arm therapy exoskeleton with an ergonomic shoulder actuation. Appl Bionics Biomech. 2009; 6: 127-42. doi: 10.1080/11762320902840179.

18.

Zhang

Song

Dong

Cao

. Improvement of human-machine compatibility of upper-limb rehabilitation exoskeleton using passive joints. Robot Auton Syst. 2019; 112: 22-31. doi: 10.1016/j.robot.2018.10.012.

19.

Wang

Tian

Tang

\alpha

-Variable adaptive model free control of iReHave upper-limb exoskeleton. 2020; 148: 102872. doi: 10.1016/j.advengsoft.2020.102872.

20.

Park

Ren

Zhang

. IntelliArm: An exoskeleton for diagnosis and treatment of patients with neurological impairments. P Ieee Ras-Embs Int. 2008; 109. doi: 10.1109/Biorob.2008.4762876.

21.

Otten

Voort

Stienen

Aarts

van Asseldonk

van der Kooij

. Limpact: A hydraulically powered self-aligning upper limb exoskeleton. Ieee-Asme T Mech. 2015; 20(5): 2285-98. doi: 10.1109/Tmech.2014.2375272.

22.

Frisoli

Rocchi

Marcheschi

Dettori

Salsedo

, M B. A new force-feedback arm exoskeleton for haptic interaction in Virtual Environments. World Haptics Conference: First Joint Eurohaptics Conference and Symposium on Haptic Interfaces for Virutual Environment and Teleoperator Systems, Proceedings. 2005: pp. 195-201. doi: 10.1109/WHC.2005.15.

23.

Gopura

RA KK

. SUEFUL-7: A 7-DOF upper-limb exoskeleton robot with muscle-model-oriented EMG-based control. 2009 IEEE/RSJ Intelligent Robots Systems. 2009: pp. 1126-31. doi: 10.1109/IROS.2009.5353935.

24.

Rahman

MH TK-O

Saad

Kenné

Archambault

. Development and control of a robotic exoskeleton for shoulder, elbow and forearm movement assistance. Appl Bionics Biomech. 2012; 9: 275-92. doi: 10.3233/ABB-2012-0061.

25.

Mayr

A KM

Saltuari

. ARMOR: Elektromechanischer roboter für das bewegungstraining der oberen extremität nach schlaganfall. Prospektive Randomisierte Kontrollierte Pilotstudie. 2008; 40: 66-73. doi: 10.1055/s-2007-989425.

26.

Reinkensmeyer

. Neurorehabilitation technology. 2016.

27.

L PJ. Wearable robots: biomechatronic exoskeletons. 2008.

28.

,Decomposition of transformation matrices for robot vision, 1984; 130-9. doi: 10.1109/ROBOT.1984.1087163.

29.

Gates

Walters

Cowley

Wilken

Resnik

. Range of motion requirements for upper-limb activities of daily living. Am J Occup Ther. 2016; 70(1). doi: 10.5014/ajot.2016.015487.

30.

Magermans

Chadwick

EKJ

Veeger

HEJ

, FCT. H. Requirements for upper extremity motions during activities of daily living. Clin Biomech. 2005; 20(6): 591-9. doi: 10.1016/j.clinbiomech.2005.02.006.

31.

Namdari

Yagnik

Ebaugh

Nagda

Ramsey

Williams

, et al. Defining functional shoulder range of motion for activities of daily living. J Shoulder Elb Surg. 2012; 21(9): 1177-83. doi: 10.1016/j.jse.2011.07.032.

32.

Helm

FCT

Veeger

HEJ

Makhsous

, al el. ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion – Part II: Shoulder, elbow, wrist and hand. J Biomech. 2005; 38(5): 981-92. doi: 10.1016/j.jbiomech.2004.05.042.

33.

Zhu

Zhang

. Neurorehabilitation. 2010; 315-25.

34.

Hee D, Lee, Byeong K, Lee, Wan S, Kim, Jung S, Han. Human-robot cooperation control based on a dynamic model of an upper limb exoskeleton for human power amplification. Mechatronics. 2014; 24: 168-76. doi: 10.1016/j.mechatronics.2014.01.007.

35.

Xin

Qiu

Peng

. Kinematics and dynamics analysis of a 3-DOF upper-limb exoskeleton with an Internally rotated elbow Joint. Applied Science. 2018; 8: 464. doi: 10.3390/app8030464.

36.

Rubinstein

Kroese

. Simulation and the monte carlo method. 2016.

37.

Kim

JO KP

. Dexterity measures for design and control of manipulators. 1991; 2: 758-63. doi: 10.1109/iros.1991.174572.

38.

Cardou

Bouchard

Gosselin

. Kinematic-sensitivity indices for dimensionally nonhomogeneous jacobian matrices. Ieee T Robot. 2010; 26(1): 166-73. doi: 10.1109/Tro.2009.2037252.

39.

Ahmad

Kumari

. A hybrid genetic algorithm approach to solve inverse kinematics of a mechanical manipulator. International Journal of Scientific & Technology Research. 2019; 8: 1772-82.

40.

Gunasekara

Gopura

Jayawardena

. 6-REXOS: Upper limb exoskeleton robot with improved pHRI. Int J Adv Robot Syst. 2015; 12. doi: 10.5772/60440.

41.

Rozo

Jaquier

Calinon

Caldwell

. Learning manipulability ellipsoids for task compatibility in robot manipulation. 2017IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2017: pp. 3183-9. doi: 10.1109/IROS.2017.8206150.

42.

Meng

Xie

Shao

Cao

Wang

, et al. Pilot study of a powered exoskeleton for upper limb rehabilitation based on the wheelchair. Biomed Res Int. 2019; 2019. doi: 10.1155/2019/9627438.

43.

Zhong

, S ML, L L. Research on rehabilitation training bed with action prediction based on NARX neural network. International Journal of Imaging Systems & Technology. 2019; 29: 539-46. doi: 10.1002/ima.22334.

44.

Fei

Xin

. Research on rehabilitation effect prediction for patients with SCI based on machine learning. World Neurosurgery. 2021. doi: 10.1016/j.wneu.2021.11.040.

45.

Xie

Meng

Zeng

Fan

Dai

. Human-exoskeleton coupling dynamics of a multi-mode therapeutic exoskeleton for upper limb rehabilitation training. IEEE Access. 2021; 9: 1. doi: 10.1109/ACCESS.2021.3072781.

46.

Carbone

Gherman

Ulinici

Vaida

Pisla

. Design Issues for an Inherently safe robotic rehabilitation device. Mech Mach Sci. 2018; 49: 1025-32. doi: 10.1007/978-3-319-61276-8_110.

47.

Cordero

Carbone

Ceccarelli

Echávarri

Muñoz

. Experimental tests in human-robot collision evaluation and characterization of a new safety index for robot operation. Mechanism and Machine Theory. 2014; 80: 184-99. doi: 10.1016/j.mechmachtheory.2014.06.004.

Design and kinematical performance analysis of the 7-DOF upper-limb exoskeleton toward improving human-robot interface in active and passive movement training

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Mechanical design

2.1 The characteristics analysis of active and passive movement training

Table 1 The ROM of the upper-limb joints in the ISB system

3. Kinematical performance analysis of the 7-DOF upper-limb exoskeletons

3.1 Workspace analysis

Table 2 The angle of the upper-limb and exoskeleton joints

3.2.1 Jacobian matrix

3.3.1 The joint angle values

4.1 Implementation of the prototype

4.2.1 Experiments of single-joint movement

5. Discussion

6. Conclusions

Footnotes

Conflict of interest

References

Table 1
The ROM of the upper-limb joints in the ISB system

Table 2
The angle of the upper-limb and exoskeleton joints