Impedance control and test of an automatic rotational orthosis for walking with arm swing

Juan Fang; Bilibin Tan; Wei Zhang; Le Xie; Guo-Yuan Yang

doi:10.4103/2773-2398.386228

RESEARCH ARTICLE

Year : 2023 | Volume : 2 | Issue : 3 | Page : 63-72

Impedance control and test of an automatic rotational orthosis for walking with arm swing

Juan Fang¹, Bilibin Tan², Wei Zhang², Le Xie², Guo-Yuan Yang²
¹ School of Mechanical Engineering, Jiangnan University, Wuxi, Jiangsu Province; The Joint Lab of the Institute of Rehabilitation Center and Chejing Robotics Technology (Shanghai) Co., Ltd., Med-X Research Institute, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China
² The Joint Lab of the Institute of Rehabilitation Center and Chejing Robotics Technology (Shanghai) Co., Ltd., Med-X Research Institute, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China

Date of Submission	11-Apr-2023
Date of Decision	05-Jun-2023
Date of Acceptance	13-Sep-2023
Date of Web Publication	26-Sep-2023

Correspondence Address:
Guo-Yuan Yang
The Joint Lab of the Institute of Rehabilitation Center and Chejing Robotics Technology (Shanghai) Co., Ltd., Med-X Research Institute, School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai
China

Source of Support: None, Conflict of Interest: None

DOI: 10.4103/2773-2398.386228

Abstract

Neurological damage after stroke and spinal cord injury often results in walking impairments. The theory of interlimb neural coupling implies that synchronized arm swing should be included during gait training to improve rehabilitation outcomes. We previously developed an automatic rotational orthosis for walking with arm swing (aROWAS), which produced coordinated interlimb movement when running in passive mode. The current case-series study had three aims: to develop impedance control algorithms for generating flexible movement in the aROWAS system, to validate its technical feasibility, and to investigate interlimb muscle activity when using it. A force-free controller was developed to compensate for gravity and friction, and an impedance controller was developed to produce a flexible movement pattern. Experiments were performed on three able-bodied volunteers to evaluate the feasibility of the flexible aROWAS system and muscle activity in their upper and lower limbs was recorded. In force-free mode, the leg rig was static but easily moved by small external forces, and the subjects reported very little resistance when attempting to walk synchronously in the aROWAS system. In impedance mode, the leg rig performed the pre-defined gait pattern, but the joint trajectories were adaptable to external forces. All participants produced earlier hip extension and greater knee flexion during active walking than during passive walking. Furthermore, the arm and lower limb muscles simultaneously produced higher electromyography activity. The control algorithms enabled the aROWAS system to produce walking-like coordinated joint performance in the upper and lower limbs, and also allowed for some degree of adjustment in response to voluntary input from the users. Stronger interlimb muscle activity was produced when participants walked actively in the system. This aROWAS system has the technical potential to serve as an effective tool for investigating interlimb neural coupling and as a novel testbed for walking rehabilitation with synchronized arm swing.

Keywords: flexible movement; impedance control; interlimb neural coupling; rehabilitation robotics

How to cite this article:
Fang J, Tan B, Zhang W, Xie L, Yang GY. Impedance control and test of an automatic rotational orthosis for walking with arm swing. Brain Netw Modulation 2023;2:63-72

How to cite this URL:
Fang J, Tan B, Zhang W, Xie L, Yang GY. Impedance control and test of an automatic rotational orthosis for walking with arm swing. Brain Netw Modulation [serial online] 2023 [cited 2023 Dec 2];2:63-72. Available from: http://www.bnmjournal.com/text.asp?2023/2/3/63/386228

Funding: This study was supported by a grant from the National Natural Science Foundation of China, No. 81401856, an Innovation and Entrepreneurship Project Funding fro-m the Government of Jiangsu Province, No. 1076010241170110 and the Youth Research Funding Scheme of Jiangnan University, No. JUSRP115A12 (all to JF).

Introduction

Neurological damage after stroke and spinal cord injury often results in lower limb dysfunction and walking impairments. The plasticity of the central nervous system provides the basis for neurological and functional recovery of walking when one participates in regular rehabilitative activities (Barbeau and Fung, 2001; Onifer et al., 2011). The theory of interlimb neural coupling defines the synchronized motion between the upper and lower limbs as the most primitive and basic movement pattern in human gait (Fang et al., 2014). A recent study showed that compared with general gait-training with only the lower limbs, the addition of rhythmic arm swinging resulted in better rehabilitation in patients after subacute stroke (Kang et al., 2018). Because the theory and clinical results indicate that synchronized arm swing should be included during gait training to improve rehabilitation outcomes, we previously developed a rotational orthosis for walking with arm swing (ROWAS) (Fang et al., 2017b). Subsequently, we improved its stability and functionality with an automatic prototype (aROWAS) that uses high-precision lift columns and automatic size adjustment (Fang et al., 2017a). With predefined trajectories for the upper and lower limbs as the references, both prototypes produced coordinated interlimb movement when they ran in the passive mode (Fang et al., 2017a, b).

Even though the trajectory-tracking performance produced by the passive position controller is very good, the aROWAS system should also provide flexible walking patterns to encourage active user participation. A trajectory-tracking approach, although often implemented as the basic movement controller in rehabilitation robotics, has limited rehabilitation outcomes (Hornby et al., 2008). The passive position-tracking method guides the patient's lower limbs to move along fixed paths that are similar to the pattern of normal walking (Colombo et al., 2000). However, such a stiff position controller creates the possibility that patients will not use enough physical effort during gait training (Marchal-Crespo and Reinkensmeyer, 2009). The metabolic costs and the muscle responses in the hip flexors have been reported to be significantly lower during passive robotic gait training than those during therapist-assisted walking (Israel et al., 2006). Passive training has also been reported to result in reduced volitional physical effort, and also interfered with motor relearning (Marchal-Crespo and Reinkensmeyer, 2009). To encourage patients to participate actively in gait training, the aROWAS system should therefore provide flexible movement, which might enhance interlimb neural coupling and promote gait restoration (Kaupp et al., 2018).

Several control approaches have been extensively investigated for generating flexible movement (Cao et al., 2014), including patient-cooperative strategies (Jezernik and Morari, 2002; Riener et al., 2005), disturbance and/or force observers (Murakami et al., 1993; Oh et al., 2014; Ugurlu et al., 2015), and path-control algorithms (Duschau-Wicke et al., 2010; Krishnan et al., 2012; Schück et al., 2012). To produce patient-cooperative robotic gait training, preliminary studies focused on adapting the gait pattern according to how the robot and the user interacted (Jezernik and Morari, 2002; Riener et al., 2005). By modifying the elastic and viscous coefficients, the robotic systems generated different impedances in reaction to the users' voluntary movements (Riener et al., 2005). This approach only allowed joint-angle deviations from the active force produced by the user (Jezernik and Morari, 2002; Riener et al., 2005). Therefore, subsequent research next focused on accurate interaction forces between the users and robots, and force/torque sensors were often adopted. However, due to the influence of the horizontal ground-reaction force, the measured interaction force between the users and the robots was only a rough estimate, especially during the stance phase (Riener et al., 2005). In power-assisted robotics, the active force from the user is the key piece of information because it guides how much torque the drive should produce to achieve the required assistance (Oh et al., 2015). To avoid the use of force sensors, which are usually fragile and costly (Capurso et al., 2017), disturbance and/or force observers were developed to estimate external forces (Murakami et al., 1993; Oh et al., 2014), and were also validated in assistive robotic exoskeletons (Ugurlu et al., 2015).

Although the interaction force is useful for the control strategies described above, it is not an indispensable parameter if other approaches are adopted for generating flexible movement. For example, the path-control strategy creates a compliant virtual “tunnel” around the desired spatial path (Krishnan et al., 2012; Schück et al., 2012). Using this strategy, if a patient walks actively along a path within the virtual tunnel, they can do the training on their own without any assistance. However, if they walk out of the virtual tunnel, a certain force is applied to pull their legs back. This approach did not require force information, but rather used the trajectory error as the input to the controller for movement control (Vallery et al., 2009; Krishnan et al., 2012).

Ideally, the aROWAS rehabilitation robotics system should produce flexible movement that users can modify according to their individual volition (Duschau-Wicke et al., 2010). Although many studies of adaptable controllers used external forces to guide the drive's output, we opted to adjust the drive based on the error between the target joint trajectories and the actual trajectories. The three aims of the current case series study were (1) to develop impedance control algorithms for generating flexible movement in the aROWAS system, (2) to validate its technical feasibility, and (3) to investigate the interlimb muscle activity produced when using it.

Materials and methods

Design

A force-free controller was developed for aROWAS to compensate for gravity and friction. An impedance controller was then implemented to produce flexible movement patterns. Experiments were performed on the aROWAS system to evaluate technical feasibility and to investigate interlimb muscle activity from able-bodied volunteers as they used the system.

Description of the aROWAS system

As described in our previous study (Fang et al., 2017a), the aROWAS system [Figure 1] comprises an elevating rotational bed and size-adjustable rigs for the upper and lower limbs. With support from two lifting columns, the aROWAS system can be automatically raised to a suitable height. Its bed frame can be rotated by two linear actuators to the various tilted or upright positions that a user prefers. The rigs for the upper and lower limbs are assembled on the bed frame (Fang et al., 2017a). The rigs have actuators for bilateral shoulder, hip, knee, and ankle joints. The drive assembly for each hip joint is an assembly of a 200 W motor (RE50, Maxon Motor, Sachseln, Switzerland), a gearbox with a gear ratio of 160, and a synchronous transmission belt with a ratio of two. Each of the shoulder, knee, and ankle joints has the same drive structure, which is a combination of a flat motor (EC90, Maxon Motor) and a gearbox with a ratio of 120. Using a PID (proportional, integral, derivative) position controller, the aROWAS system produced good trajectory-tracking performance in passive mode (Fang et al., 2017a).

Figure 1: The aROWAS system and the experimental setup.
Note: The aROWAS system comprises an elevating rotational bed and size-adjustable rigs for the upper and lower limbs. aROWAS: Automatic rotational orthosis for walking with arm swing.

Click here to view

Force-free controller

The function of the force-free controller (the dashed rectangular frame in [Figure 2]) was to compensate for gravity and friction so that the joint drive responded only to the external force exerted by the user. Based on inverse dynamics, the required torque to move the rig of the aROWAS system was

Figure 2: The overall control strategy.
Note: The portion of the diagram surrounded by the dashed rectangle is the force-free controller. aROWAS: Automatic rotational orthosis for walking with arm swing; T_ext: torque produced by external forces; T_f: torque to compensate friction; T_g : torque induced by gravity; T_imp : torque calculated based on the impedance controller; T_R. torque provided by the drive; θ: joint angle; θ_error: difference between the target and actual joint angle; θ: angular velocity.

Click here to view

where M_R, C_R, T_p and T_g respectively represent the inertia, the centrifugal/Coriolis force, and the torques to compensate for friction and gravity. The variable T_R is the torque provided by the drive, and T_ext is the torque produced by external forces. Due to the low speed of the rig, the Coriolis/centrifugal force C_R is negligible. Therefore,

By analyzing the mechanical configuration, as well as the masses of the rig and the attached drive, the theoretical torque needed to compensate for the rig's gravity T_g can be calculated. To obtain the friction compensator T_f for each joint assembly, many repetitive experiments were performed in the physical aROWAS system using the method described in Morante et al. (2016). Friction was modelled as a combination of Coulomb and viscous components. The Coulomb friction torques were determined as the torques when the joint ran at speeds of 0.5°/s and −0.5°/s.

Repetitive testing for friction compensation was performed on the hip and knee joints. As the load in the shoulder joint was very small, the friction in the shoulder joint was considered negligible. When the friction of the hip joint was tested, the knee drive was locked. The currents from the joint motors were recorded and used to calculate the torques from the joint drives. The torques and angular speeds were filtered using a 3^rd-order Butterworth digital filter, and finally the friction compensators for the hip and knee joints were obtained.

With the gravity and friction compensators, the hip and knee joints ran in force-free mode, meaning that the leg rig theoretically moved without any influence from gravity or joint friction.

Impedance controller

Based on the above force-free model, a feedback impedance controller [Figure 2] was developed,

where K and B are the linear coefficients for elasticity and viscosity, respectively. The actual joint trajectory was recorded by a position sensor integrated in the motor. The difference between the actual trajectories and the target trajectories, θ_error, was fed into the impedance controller, resulting in an extra torque T_imp. As gravity and friction were well compensated, θ_error under these circumstances primarily came from the user's active movement, in addition to minor factors such as mechanical inertia of the rig, the user's segment inertia, and the passive elastic/viscous joint moments (Riener and Edrich, 1999). The impedance controller worked as a virtual spring-damper system that constrained the joint to follow the target trajectory. If the user exerted volitional force on the rig, the force was applied on an analog spring-damper system. The force-free controller and the impedance controller combined to produce flexible movement in the system.

Experimental evaluation of the aROWAS system

Two tests were performed to evaluate performance: a preliminary test on the aROWAS system to assess the performance of the force-free and impedance controllers, and a formal test with able-bodied volunteers using the aROWAS system. During the latter, we collected user feedback and recorded interlimb muscle activity. Ethical approval was obtained from the Ethics Committee at the Med-X Research Institute, Shanghai Jiao Tong University (Shanghai, China; December 17, 2017). Written informed consent was obtained from the three volunteers prior to participation.

During testing, the aROWAS system was in an upright position. The pre-defined walking parameters (reference) for the rigs were the normal walking trajectories recorded in our previous study (Fang et al., 2016). The gait period of the reference was 5 seconds. Because stable support was required during the tests with volunteers, the actuators for the bilateral ankle joints were locked in an upright standing position during the whole test. The shoulder, hip, and knee joints on the right side always ran in passive mode, i.e., tracking the predefined trajectories as described above. In contrast, the left side of the system ran in passive, force-free, and impedance modes during the test.

The preliminary test on the physical aROWAS system without any users aimed to evaluate the feasibility of running the system in the three different modes. In passive mode, the left shoulder, hip, and knee joints tracked the reference trajectories. In impedance mode, these three joints ran the control algorithms [Figure 2] using the same reference. In force-free mode, the left shoulder drive still ran the impedance controller, while the left hip and knee joints had gravity and friction compensated for, as shown in the rectangular frame in [Figure 2]. A disturbance force was applied to the rigs by manually pushing a force sensor. In passive and impedance modes, an external force was exerted on the middle of the shank during the stance phase when the rig was performing the reference walking pattern. In force-free mode, the leg rig was manually adjusted to an upright position. To test only the hip joint, the knee joint was locked before an external force was applied at the middle of the thigh. Similarly, when testing the knee joint, a force was applied to the middle of the shank rig while the hip joint was locked. The joint angles and motor currents were measured during all tests, and the disturbance force was recorded using a force sensor (JLBM100, 1 kN, Zhongwan Sensor Co., Ltd., Bengbu, China).

Three able-bodied volunteers were recruited for the second test [Table 1]. Sixteen electromyography (EMG) electrodes were attached bilaterally on eight muscles of the upper and lower limbs: biceps brachii (BB), triceps brachii (TB), deltoid (DT), vastus medialis (VM), rectus femoris (RF), biceps femoris (BF), tibialis anterior (TA), and gastrocnemius medialis (GM). The specific positions were determined as described in Winter and Yack (1987). The targeted skin areas were cleaned with alcohol to reduce contact impedance. EMG signals were collected at 1927 Hz with a bandwidth of 20–450 Hz using a commercial EMG system (Trigno™ Wireless system, Delsys Inc., Natick, MA, USA).

Table 1: Participant information

Click here to view

Before each test, the aROWAS bed was adjusted to a lying position. The participant lay on the bed, and their upper and lower limbs were fixed to the rigs. After they were secured with the body weight support system, the aROWAS bed frame was moved to an upright position [Figure 1]. Minor adjustments were made to make the user comfortable. In passive mode, the user followed the movement produced by the system without voluntarily exerting any force. In force-free mode, the participant was encouraged to walk independently in the system, trying to move the left leg in step with the passively moved right side. In impedance mode, the participant was asked to walk actively, especially with the left leg. Each walking mode lasted for 30 seconds and was repeated three times (Additional Video 1). After each test, participants were asked the following feedback questions:

[Additional file 1]

How did you feel when you used the system in force-free mode?

How well could you perform active walking when the system ran in impedance mode?

Which mode did you like best, passive, force-free, or impedance?

Do you have any advice for improving the performance?

Data analysis

Raw EMG signals were full-wave rectified and low-pass filtered at 5 Hz with a zero-lag 2^nd-order Butterworth digital filter, and were further smoothed using a 240-point window root-mean-square (RMS) algorithm (Hesse et al., 2010). After synchronizing the joint angle with the muscle response, the EMG signals for each gait cycle (GC) were summarized. The duration of one GC was normalized to 100%, with heel strikes at 0% and 100%. A stride-to-stride ensemble average for each muscle from each participant was selected as the final EMG profile within a GC for each walking mode (Gui and Zhang, 2016). To compare the EMG responses from different modes, the mean RMS values from all participants during one GC were calculated for each of the three modes. The RMS EMG during the force-free and impedance modes is presented relative to that during the passive mode.

Although the joint angles and EMG from both sides were recorded, only the results from the left side are presented in this study.

Results

Preliminary test results

The aROWAS system tracked the trajectories well in passive mode. When an external force of about 50 N was applied on the middle of the shank, the torques from the hip and knee drives varied accordingly, resulting in robust tracking of the reference trajectory [Figure 3]A. Because the force had a longer moment arm to the hip joint than to the knee joint, the hip drive produced a larger torque increase than the knee drive. In force-free mode, as no reference trajectory was provided, the rig remained static if no disturbance force was applied. Applying the manual push moved the rig accordingly. When a small force of about 30 N was applied on the thigh, and the hip moved about 11°, and the hip drive responded with a maximal torque output of about 25 N·m [Figure 4]A. After the force was removed, the speed of the hip joint lessened and gradually stopped, primarily as a result of mechanical inertia. When a force of about 40 N was applied to the shank, the knee joint moved about 43° before it finally stopped [Figure 4]B.

Figure 3: The responses during manual push (solid lines) compared with those when lacking an external disturbance (dashed lines) in passive and impedance modes.
Note: The second and fourth rows show the results for the hip, while the third and fifth rows show the results for the knee joint.

Click here to view

Figure 4: The responses for the hip and knee during manual push in force-free mode.

Click here to view

In impedance mode, the parameters K and B were tentatively set to be 1.0 N·m/° and 1.0 N·ms/°, respectively. When a force of about 50 N was applied to the middle of the shank rig, trying to push the leg backwards, the speed of the hip joint increased. The maximal extension occurred at ~6% of GC earlier than it did for the reference, and the extension amplitude increased by ~6o [Figure 3]B. Simultaneously, the speed of the knee joint also increased, resulting in a slightly larger stance flexion. With the external force pushed the leg backwards, reduced torque was required from the hip drive to extend the hip joint. In contrast, the knee drive produced a larger torque to resist the disturbance, so as to extend the knee joint, as defined by the reference. The hip and knee joints returned to track the target trajectory soon after the external force was removed.

Main test results

In agreement with the preliminary results, when used by healthy volunteers, the aROWAS system produced good trajectory-tracking performance in passive mode [Figure 5]. The actual trajectories of the shoulder, hip, and knee joints almost perfectly matched the references. In contrast, during force-free mode, the hip and knee trajectories varied widely. None of the participants extended their hip joints as much as they did in normal hip movements. Participant 1 (P1) [Figure 5]A walked fairly well, with a kinematic pattern similar to normal walking. However, the hip extended about 10% of GC earlier than it did during normal gait. Furthermore, the minor flexion of the knee joint in the initial stance phase that occurs in normal gait was hardly observed. P2 did not walk as well as P1, in that the ranges of hip and knee joint motion were much reduced [Figure 5]B. Among the three participants, P3 produced movements that were the most different from the normal pattern; he (1.73 m, 72 kg) failed to produce phasic extension or flexion in the hip or knee joint. The hip joint did not flex much at heel strike, and the knee joint constantly flexed. Range of motion for P3 was 18° for the hip and 10° for the knee joint during the whole GC [Figure 5]C.

Figure 5: The kinematics of the aROWAS system used by a healthy volunteer in passive (dot lines), force-free (dashed lines), and impedance (dash-dot lines) modes.
Note: The solid lines are the reference trajectories. aROWAS: Automatic rotational orthosis for walking with arm swing; P1–3: three participants.

Click here to view

In impedance mode, the actual joint trajectories were similar to the references with minor differences in the phases and amplitudes. The hip extended about 10% of GC earlier than the reference. Furthermore, for P2 and P3, the maximal hip flexion occurring at about 90% of GC was 2° larger than the reference value. Regarding the knee joint, P1 and P2 produced 3° more flexion at 75% of GC, while P3 had a maximal knee flexion that was reduced by 4°. The knee joint of P3 was constantly flexed in the stance phase, with about 11° more flexion than the reference. During the whole test, all three participants produced a range of motion in the shoulder joint that was about 3° lower than the reference value.

Although different people produce different EMG responses during walking (Wootten et al., 1990), some general EMG patterns were observed in all three participants when they walked passively in the aROWAS system [Figure 6]. During the whole GC, the arm muscles produced very small EMG responses, while the leg muscles produced relatively large responses. TA and GM had phasic peaks during the swing and stance phases, respectively. There were a few slight differences among the three participants: P1 produced peak muscle activity in the VM and RF at heel strike, P2 produced EMG bursts in the DT and BF, and P3 had higher EMG in the TB and DT.

Figure 6: EMG responses for P1, P2, and P3 when they walked passively in the aROWAS system.
Note: A–H: Biceps brachii, triceps brachii, tibialis anterior, vastus medialis, rectus femoris, biceps femoris, tibialis anterior, gastrocnemius medialis. aROWAS: Automatic rotational orthosis for walking with arm swing; EMG: electromyography; P1–3: three participants.

Click here to view

When the participants walked in force-free mode, their EMG activity generally increased [Figure 7]A. P1 produced much higher EMG in the TA during the whole GC, with the peak occurring at the mid-stance phase [Figure 8]. In contrast to walking in passive mode, in force-free mode, P1 did not produce phasic peaks in the VM or RF at heel strike. Instead, we observed a slight burst in the BF. The other two participants demonstrated varied muscle patterns, but both P2 and P3 exhibited greatly increased responses in the BB, BF, TA, and GM [Figure 7]A.

Figure 7: The ratio of the overall RMS EMG values during the whole GC when participants walked in force-free (upper) or impedance (lower) modes relative to those during walking in passive mode.
Note: The bars in white, gray, and black, respectively show the muscle activity from P1, P2 and P3. The dashed lines show the RMS EMG during passive mode. BB: Biceps brachii; BF: biceps femoris; DT: deltoid; EMG: EMG: electromyography; GC: gait cycle; GM: gastrocnemius medialis; RF: rectus femoris; RMS: root-mean-square; P1–3: three participants; TA: tibialis anterior; TB: triceps brachii; VM: vastus medialis.

Click here to view

Figure 8: EMG responses from participant P1 walking in passive, force-free, and impedance modes.
Note: A–H: Biceps brachii, triceps brachii, tibialis anterior, vastus medialis, rectus femoris, biceps femoris, tibialis anterior, gastrocnemius medialis. EMG: Electromyography; P1: participant 1.

Click here to view

When participants walked in impedance mode, the general muscle patterns became similar to those observed during passive walking, but with much higher amplitudes. For P1, the BF produced a peak at heel strike, which was 2.5 times the amplitude of that during passive mode [Figure 8]. The overall RMS in the BF during the whole GC was ~2.5 times higher than that in passive mode [Figure 7]. P2 displayed higher EMG values from the BF and GM [Figure 7]. Among the three participants, P3 had the lowest EMG increase compared with passive mode. Nevertheless, his EMG values in the BF were twice as high as those in passive mode [Figure 7]. Furthermore, all three participants presented phasic EMG increases in the arm muscles, especially the BB, in impedance mode [Figure 7] and [Figure 8].

The participants reported that following the movements when the aROWAS system ran in passive mode was very comfortable, and that it felt similar to walking. In force-free mode, all participants mentioned that they were uncomfortable because they did not feel enough resistance. P1 thought it was interesting to walk without friction. She described it as “easy walking.” P2 dared not push his left leg back when he did not feel much resistance. He described it as “unbalanced walking” because while his right leg was guided to walk in a normal gait pattern, he could not move his left leg synchronously, which made him feel uncomfortable. P3 found it similar to skating, and had difficulty lifting his shank up because he was concerned he would lose his balance. All participants reported a much greater sense of security when walking in impedance mode than in force-free mode. All participants tried to walk actively by stepping further, but felt resistance when they tried to walk too fast. P1 and P2 preferred walking in impedance mode, while P3 liked passive mode the most. To improve performance, P1 thought it desirable to walk at a higher speed, while P2 thought that having some kind of performance feedback on a screen would be useful, especially in impedance mode. P3 did not like stepping in the air, and thought that walking on the floor or on a firm support/plate would help create a higher sense of security.

Discussion

For this case-series study, we developed impedance control algorithms for generating flexible movement in the aROWAS system, validated the technical feasibility, and investigated interlimb muscle activity of participants as they used the machine. Gravity and friction were well compensated for, and the participants reported that they detected little resistance when walking in force-free mode. In impedance mode, the joint trajectories were adaptable to the external force. All participants walked actively in the aROWAS system, and adjusted their trajectories voluntarily. Increased EMG responses were observed simultaneously in upper and lower limb muscles.

The force-free controller performed satisfactorily when tested, although the friction models have room for improvement. Friction in the hip and knee joints was modelled as a combination of Coulomb and viscous components. If gravity and friction were accurately compensated for, in force-free mode, the leg rig is theoretically responsive to any small external force. Manual push with a force of 30 N and 40 N moved the hip and knee joint approximately 11° and 43°, respectively [Figure 4]. This friction compensator might be improved if advanced friction models such as LuGre (Madi et al., 2004) or elastoplastic (Temeltas and Aktas, 2006) friction models are incorporated. Nevertheless, the tests in force-free mode demonstrated that the friction models gave satisfactory performance. All participants reported that resistance was so small in force-free mode that they were reluctant to push their left leg backwards, resulting in reduced hip extension [Figure 5]. Although secured by the body weight support system, two of the three participants were worried about losing their balance. Even though the passively moved right upper/lower limbs and the left shoulder joint were guided, only one participant (P1) walked with an acceptable gait pattern. All participants had greater flexed knee movement during the stance phase, and reduced knee flexion in the swing phase [Figure 5]. This sense of insecurity was further revealed by the EMG activity. All three participants had increased EMG signals in the TA and GM [Figure 8], which are the primary muscles used to maintain balance (Winter, 1995). These kinematic and muscle responses demonstrated that friction and gravity were well compensated for, which provides a good basis for further development of the impedance controller. The asynchronous leg movement exhibited in force-free mode highlights the necessity of a reference movement to guide the walking, which was implemented in impedance mode.

With target trajectories as the references, the impedance controller assisted the participants in walking with a normal gait pattern, but also created the potential for them to actively adjust the joint performance. Apart from the torque needed to compensate for gravity and friction, an additional torque T_imp was provided, which depended on the deviation of the actual angle from the reference θ_error [Figure 2]. If a user walked actively, as the reference trajectory described, the impedance controller did not produce any output T_imp. Otherwise, the impedance controller served as a virtual spring-damper system that assisted the joint in following the target trajectory. In contrast to the impedance controller described in Riener et al. (2005), which allowed angle deviations only from the user's voluntary force, the controller in our study did not produce assisted torque T_imp until angle deviations were observed. Apart from the small influence of the rig's mechanical inertia, the segment inertia of the participants, and the passive elastic and viscous joint moments, θ_error was mainly affected by the force exerted by the participants (their volitional input). The general goal of flexible control strategies is to generate gait trajectories that can be adapted in response to a user's volitional effort, thus encouraging active participation (Riener et al., 2005). Therefore, the control algorithms developed in this study achieved this goal.

Enhanced interlimb muscle activity was observed during active walking in the aROWAS. In impedance mode, participants were asked to move their left leg actively, and the resulting muscle activity was measured by recording EMG amplitudes. Compared with passive mode, all participants produced earlier hip extension in the other two modes, accompanied by an obvious EMG increase in the BF [Figure 8]B. P2 walked with a more flexed knee joint during the stance phase, which was confirmed by an increased EMG in the GM [Figure 8]B. Although no special attention was paid to arm movements, participants simultaneously produced slightly adjusted shoulder movement during active walking in impedance mode [Figure 5], which was accompanied by higher EMG in the arm muscles [Figure 7] and [Figure 8]). Compared with passive mode, active walking in the aROWAS with impedance algorithms produced stronger interlimb neural coupling, which is believed to be helpful for promoting gait rehabilitation in patients following stroke (Kang et al., 2018; Kaupp et al., 2018). However, due to the limited number of volunteers recruited in this study, detailed investigation of interlimb neural coupling when using the aROWAS requires further investigation. Nevertheless, the current study demonstrates that the aROWAS system has the technical potential to serve as an effective tool for investigating interlimb neural coupling.

Using the combined algorithms for friction compensation, gravity compensation, and impedance control, our study found that flexible movement in the aROWAS system is feasible in able-bodied volunteers. Future work will focus on using the aROWAS system as a tool for investigating interlimb neural coupling and as a testbed for rehabilitating walking impairments by employing synchronized arm swing.

A limitation of this study is that the impedance algorithms did not measure the user/robot interaction, which might be another useful parameter when observing patients during active training. In our experiments with able-bodied volunteers, only tentative stiffness and viscous coefficients were used to test the technical feasibility of the impedance algorithms. However, these parameters could be adjusted online to obtain different impedances during clinical tests. Another limitation was the small sample size of three able-bodied participants. However, as a case-series study, we believe that a sample of three participants is enough to validate technical feasibility. Nevertheless, more tests and a larger sample size is desirable when further investigating the system's performance. Further work will incorporate visual feedback on performance and include patients who use the system with synchronized arm swing to rehabilitate walking.

In conclusion, we developed, tested, and validated control algorithms for the aROWAS. The novelty of this work is that the control algorithms enable the aROWAS system to produce walking-like coordinated joint performance in the upper and lower limbs, while at the same time, allowing for some degree of adjustment in response to voluntary input. The results show that the flexible aROWAS system is feasible and has the technical potential to serve as an effective tool for both investigation of interlimb neural coupling and as a testbed for rehabilitation of walking with synchronized arm swing.

Acknowledgments

The authors acknowledge the staff of the Chejing Robotics Technology (Shanghai) Co., Ltd for their support in testing the aROWAS system.

Author contributions

JF developed the control algorithms, designed and conducted the experiment, performed data collection, analysis and interpretation, and drafted and revised the manuscript. BT performed substantial contributions to developing the controller, conducting the experiment and data collection, and revised the manuscript. WZ participated in controller development, experiment design and manuscript revision. LX and GYY participated in the study design and data interpretation, and revised the work critically for important intellectual content. All the authors read and approved the final manuscript.

Conflicts of interest

The authors declare that they have no competing interests.

Editor note: GYY is an Editorial Board member of Brain Network and Modulation. He was blinded from reviewing or making decisions on the manuscript. The article was subject to the journal's standard procedures, with peer review handled independently of the Editorial Board member and his research groups.

Availability of data and materials

All data generated or analyzed during this study are included in this published article and its Additoonal file.

Open access statement

This is an open access journal, and articles are distributed under the terms of the Creative Commons AttributionNonCommercial-ShareAlike 4.0 License, which allows others to remix, tweak, and build upon the work non-commercially, as long as appropriate credit is given and the new creations are licensed under the identical terms.

Additional file

Additional Video 1: A representative subject using the aROWAS system in the three modes.[33]

References

1.	Barbeau H, Fung J (2001) The role of rehabilitation in the recovery of walking in the neurological population. Curr Opin Neurol 14:735-740.
2.	Cao J, Xie SQ, Das R, Zhu GL (2014) Control strategies for effective robot assisted gait rehabilitation: the state of art and future prospects. Med Eng Phys 36:1555-1566.
3.	Capurso M, Ardakani MMG, Johansson R, Robertsson A, Rocco P (2017) Sensorless kinesthetic teaching of robotic manipulators assisted by observer-based force control. Proc IEEE Int Conf Robot Autom 2017: 945-950.
4.	Colombo G, Joerg M, Schreier R, Dietz V (2000) Treadmill training of paraplegic patients using a robotic orthosis. J Rehabil Res Dev 37:693-700.
5.	Duschau-Wicke A, von Zitzewitz J, Caprez A, Lunenburger L, Riener R (2010) Path control: a method for patient-cooperative robot-aided gait rehabilitation. IEEE Trans Neural Syst Rehabil Eng 18:38-48.
6.	Fang J, Xie L, Yang G-y (2014) Review on the interlimb neural coupling and its potential usage in walking rehabilitation. J Shanghai Jiaotong Univ (Sci) 19:561-564.
7.	Fang J, Yang GY, Xie L (2017a) Development of an automatic rotational orthosis for walking with arm swing. IEEE Int Conf Rehabil Robot 2017:264-269.
8.	Fang J, Hunt KJ, Xie L, Yang GY (2016) Modelling of the toe trajectory during normal gait using circle-fit approximation. Med Biol Eng Comput 54:1481-1489.
9.	Fang J, Xie Q, Yang GY, Xie L (2017b) Development and feasibility assessment of a rotational orthosis for walking with arm swing. Front Neurosci 11:32.
10.	Gui K, Zhang D (2016) Influence of locomotion speed on biomechanical subtask and muscle synergy. J Electromyogr Kinesiol 30:209-215.
11.	Hesse S, Waldner A, Tomelleri C (2010) Innovative gait robot for the repetitive practice of floor walking and stair climbing up and down in stroke patients. J Neuroeng Rehabil 7:30.
12.	Hornby TG, Campbell DD, Kahn JH, Demott T, Moore JL, Roth HR (2008) Enhanced gait-related improvements after therapist-versus robotic-assisted locomotor training in subjects with chronic stroke: a randomized controlled study. Stroke 39:1786-1792.
13.	Israel JF, Campbell DD, Kahn JH, Hornby TG (2006) Metabolic costs and muscle activity patterns during robotic- and therapist-assisted treadmill walking in individuals with incomplete spinal cord injury. Phys Ther 86:1466-1478.
14.	Jezernik S, Morari M (2002) Controlling the human-robot interaction for robotic rehabilitation of locomotion. In: 7th International Workshop on Advanced Motion Control. Proceedings (Cat. No.02TH8623), pp 133-135.
15.	Kang TW, Oh DW, Lee JH, Cynn HS (2018) Effects of integrating rhythmic arm swing into robot-assisted walking in patients with subacute stroke: a randomized controlled pilot study. Int J Rehabil Res 41:57-62.
16.	Kaupp C, Pearcey GEP, Klarner T, Sun Y, Cullen H, Barss TS, Zehr EP (2018) Rhythmic arm cycling training improves walking and neurophysiological integrity in chronic stroke: the arms can give legs a helping hand in rehabilitation. J Neurophysiol 119:1095-1112.
17.	Krishnan C, Ranganathan R, Kantak SS, Dhaher YY, Rymer WZ (2012) Active robotic training improves locomotor function in a stroke survivor. J Neuroeng Rehabil 9:57.
18.	Madi MS, Khayati K, Bigras P (2004) Parameter estimation for the LuGre friction model using interval analysis and set inversion. In: 2004 IEEE International Conference on Systems, Man and Cybernetics (IEEE Cat. No.04CH37583), pp 428-433 vol.421.
19.	Marchal-Crespo L, Reinkensmeyer DJ (2009) Review of control strategies for robotic movement training after neurologic injury. J Neuroeng Rehabil 6:20.
20.	Morante S, Victores JG, Martínez S, Balaguer C (2016) Force-Sensorless Friction and Gravity Compensation for Robots. In: Robot 2015: Second Iberian Robotics Conference (Reis LP, Moreira AP, Lima PU, Montano L, Muñoz-Martinez V, eds), pp 57-68. Cham: Springer International Publishing.
21.	Murakami T, Yu F, Ohnishi K (1993) Torque sensorless control in multidegree-of-freedom manipulator. IEEE Trans Ind Electron 40:259-265.
22.	Oh S, Kong K, Hori Y (2014) Design and analysis of force-sensor-less power-assist control. IEEE Trans Ind Electron 61:985-993.
23.	Oh S, Baek E, Song S-k, Mohammed S, Jeon D, Kong K (2015) A generalized control framework of assistive controllers and its application to lower limb exoskeletons. Rob Auton Syst 73:68-77.
24.	Onifer SM, Smith GM, Fouad K (2011) Plasticity after spinal cord injury: relevance to recovery and approaches to facilitate it. Neurotherapeutics 8:283-293.
25.	Riener R, Edrich T (1999) Identification of passive elastic joint moments in the lower extremities. J Biomech 32:539-544.
26.	Riener R, Lünenburger L, Jezernik S, Anderschitz M, Colombo G, Dietz V (2005) Patient-cooperative strategies for robot-aided treadmill training: first experimental results. IEEE Trans Neural Syst Rehabil Eng 13:380-394.
27.	Schück A, Labruyère R, Vallery H, Riener R, Duschau-Wicke A (2012) Feasibility and effects of patient-cooperative robot-aided gait training applied in a 4-week pilot trial. J Neuroeng Rehabil 9:31.
28.	Temeltas H, Aktas G (2006) State observation for elastoplastic friction models in positioning systems by utilizing leunberger observers. Proc Inst Mech Eng Part I J Syst Control Eng 220:417-426.
29.	Ugurlu B, Nishimura M, Hyodo K, Kawanishi M, Narikiyo T (2015) Proof of concept for robot-aided upper limb rehabilitation using disturbance observers. IEEE Trans Hum Mach Syst 45:110-118.
30.	Vallery H, Duschau-Wicke A, Riener R (2009) Generalized elasticities improve patient-cooperative control of rehabilitation robots. In: 2009 IEEE International Conference on Rehabilitation Robotics, pp 535-541.
31.	Winter DA (1995) Human balance and posture control during standing and walking. Gait Posture 3:193-214.
32.	Winter DA, Yack HJ (1987) EMG profiles during normal human walking: stride-to-stride and inter-subject variability. Electroencephalogr Clin Neurophysiol 67:402-411.
33.	Wootten ME, Kadaba MP, Cochran GV (1990) Dynamic electromyography. II. Normal patterns during gait. J Orthop Res 8:259-265.