Abstract
Robotic exoskeletons have considerable, but largely untapped, potential to restore mobility in individuals with neurological disorders, and other conditions that result in partial or complete immobilization. The growing demand for these devices necessitates the development of technology to characterize the human-robot system during exoskeletal-assisted locomotion (EAL) and accelerate robot design refinements. The goal of this study was to combine controlled experiments with computational modeling to build a virtual simulator of EAL. The first objective was to acquire a minimum empirical dataset comprising human-robot kinematics, ground reaction forces, and electromyography during exoskeletal-assisted and unassisted locomotion from an able-bodied participant. The second objective was to quantify the dynamics of the human-robot system using a subject-specific virtual simulator reproducing EAL compared to the dynamics of normal gait. We trained an able-bodied participant to ambulate independently in a Food and Drug Administration-approved exoskeleton, the ReWalk P6.0 (ReWalk Robotics, Yoknaem, Israel). We analyzed the motion of the participant during exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers, with simultaneous measurements of (i) three-dimensional marker trajectories, (ii) ground reaction forces, (iii) electromyography, and (iv) exoskeleton encoder data. We created a virtual simulator in OpenSim, comprising a whole-body musculoskeletal model and a full-scale exoskeleton model, to determine the joint kinematics and moments during exoskeletal-assisted and unassisted maneuvers. Mean peak knee flexion angles of the human subject during exoskeletal-assisted walking were 50.1° ± 0.6° (left) and 52.6° ± 0.7° (right), compared to 68.6° ± 0.3° (left) and 70.7° ± 1.1° (right) during unassisted walking. Mean peak knee extension moments during exoskeletal-assisted walking were 0.10 ± 0.10 Nm/kg (left) and 0.22 ± 0.11 Nm/kg (right), compared to 0.64 ± 0.07 Nm/kg (left) and 0.73 ± 0.10 Nm/kg (right) during unassisted walking. This work provides a foundation for parametric studies to characterize the effects of human and robot design variables, and predictive modeling to optimize human-robot interaction during EAL.
Introduction
Wearable robotic exoskeletons have considerable potential to transform the people’s lives by reducing the metabolic cost of walking [1–3], carrying heavy loads [4, 5], or augmenting human performance to conduct strenous tasks over extended periods of time [6, 7]. The need for such assistive devices is particularly profound in restoring mobility in individuals with neurological disorders, including those of spinal cord injury, stroke, traumatic brain injury, and multiple sclerosis. Wearable robotic exoskeletons for rehabilitation of individuals with neurological disorders are relatively new. The ReWalk (ReWalk Robotics, Yoknaem, Israel) was the first lower extremity device to receive Food and Drug Administration (FDA) approval in 2014, followed by the Ekso (Ekso Bionics, Richmond, CA) and Indego (Parker Hannifin, Cleveland, OH) in 2016, and the Keeogo (B-temia Inc., Quebec City, Canada) in 2020. Prior studies have demonstrated the ability of individuals with neurological disorders who have partial impairments to those who are completely non-ambulatory to walk independently in robotic exoskeletons [8–19]. Exoskeletal-assisted locomotion (EAL) has been shown to improve functional and motor recovery [8], mobility [9, 10, 15, 16], chronic pain [20], muscle spasticity [20–22], cardiovascular health [13, 23], bowel function [24], bladder function [21, 22], and quality of life [21]. These studies highlight the growing demand for wearable robotic exoskeletons to improve physical and psychological health, employment opportunities, and community integration in persons with neurological disorders.
The growing demand for wearable robotic exoskeletons for rehabilitation of individuals with neurological disorders has exposed two gaps in knowledge. First, the dynamics (comprising joint kinematics and joint kinetics) of the human-robot system during EAL in FDA-approved devices are not well-understood. The human-robot entity during EAL represents a complex dynamic system, and parsing out the contributions of the human from the robot, together with their interaction effects, requires a minimum empirical dataset that includes human-robot kinematics, ground reaction forces, and electromyography (EMG). Such a dataset from EAL in an FDA-approved device does not exist. Kim and colleagues quantified the dynamics of the human-robot system during EAL in a device designed to augment human performance in industrial settings [25]; however, this exoskeleton is not FDA-approved for rehabilitation of individuals with neurological disorders. Previous studies with FDA-approved exoskeletons for neurological disorders have reported joint kinematics [10, 26–28], EMG [10, 29, 30], and foot reaction forces from pressure insoles [14]. To the best of our knowledge, no prior study has reported joint kinetics of the human-robot system during EAL in an FDA-approved device. Furthermore, a minimum empirical dataset comprising joint kinematics, ground reaction forces, and EMG from a single session of controlled experiments with and without an FDA-approved exoskeleton is not available in the literature.
The second gap in knowledge exposed by the growing demand for wearable robotic exoskeletons is that existing technology to accelerate design refinements and improve human experience in these devices is extremely limited. It is not ethically feasible, and would undoubtedly be prohibitively expensive, to conduct a large number of human experiments to characterize the effects of an endless number of robot design variables, control strategies, and human parameters on the human-robot system during EAL. Computational simulation is a viable alternative to perform such parametric studies. Prior studies have used computational simulations of the human-robot system to optimize exoskeleton design [31–35], test control strategies [36–39], minimize the metabolic cost of locomotion [40–42], optimize assistance for pathological gait [43, 44], and study human-robot interaction [45–50]. However, no prior study has simulated EAL in an FDA-approved exoskeleton.
The goal of this study was to combine controlled experiments with computational modeling to build a virtual simulator of EAL in an FDA-approved lower extremity exoskeleton, the ReWalk P6.0. The first objective of this study was to acquire a minimum empirical dataset comprising human-robot kinematics, ground reaction forces, and EMG during exoskeletal-assisted and unassisted locomotion. The second objective of this study was to quantify the dynamics of the human-robot system using a subject-specific virtual simulator reproducing EAL compared to the dynamics of normal gait.
Methods
Participant recruitment
We recruited an able-bodied participant (age in their 40s, male, height 1.76 m, mass 89.4kg) for this study. The inclusion criteria were age between 18-55 years, height between 1.60-1.90 meters, and weight <90 kg, body mass index between 18.5-29.9 kg/m2, no known history of musculoskeletal disease or dysfunction, normal range of motion of the lower extremities, and willingness to commit to exoskeleton training for five sessions followed by one session of motion capture experiment performing exoskeletal-assisted and unassisted locomotion. The exclusion criteria were known history of musculoskeletal surgeries or non-surgical interventions, inability to hold crutches, pregnant, lactating, or menopausal. The participant was informed on all aspects of the study and provided informed signed consent according to the policies of our Institutional Review Board.
Exoskeletal-assisted locomotion training
The participant was trained to perform sit-to-stand, stand-to-sit, and walking maneuvers in the ReWalk P6.0. Prior to initiating training, the exoskeleton was adjusted to fit the anthropometry of the participant, including pelvic band size, thigh leg length, shank leg length, knee bracket position, foot plate size, and ankle dorsiflexion setting [51]. Walking in the ReWalk P6.0 follows a standard procedure. To initiate walking in the ReWalk P6.0, the user starts in the standing position with the hand crutches on each side. The user activates the “Walk” mode through a controller watch, which is confirmed by a single beep. To perform the device-assisted gait maneuver, the user unweights each side in an alternating manner—that is, the right leg is unloaded by leaning the torso to the left, which is followed by leaning the torso to the right, which unloads the left leg, and so forth. The action of the user swaying from side-to-side and, thereby, unloading the contralateral leg, activates the hip and knee motors to initiate the stepping motion. The next several steps are a repetition of the back-and-forth swaying of the torso to unload the trailing leg to permit the stepping action to continue.
The sit-to-stand and stand-to-sit maneuvers in the ReWalk P6.0 follow standard procedures. To initiate the sit-to-stand maneuver, a user places the hand crutches posterior to the hip while seated. Then, the trainer activates the “Stand” mode through a controller, which is confirmed by a single beep. The hip motors flex the torso (∼9°) to position the user’s upper body forward while they load the crutches. The user has to hold this position for 3 seconds, after which time the exoskeleton beeps thrice and activates the hip and knee motors to complete the sit-to-stand maneuver. Next, the stand-to-sit maneuver in the ReWalk P6.0 follows a similar procedure. A user places the hand crutches posterior to the hip while standing. Then, the trainer activates the “Sit” mode, which is confirmed by a single beep. The user gets into position by leaning backward (∼6°) and loading the crutches. The user has to hold this position for 3 seconds, after which time the exoskeleton beeps thrice and activates the hip and knee motors to complete the stand-to-sit maneuver. By the end of the fourth 1-hour session, the participant was able to perform the sit-to-stand, stand-to-sit, and walking maneuvers in the exoskeleton independent of requiring any assistance.
Motion capture experiments during exoskeletal-assisted and unassisted locomotion
We analyzed the participant during exoskeletal-assisted and unassisted locomotion from a single motion capture session, including simultaneous measurements of three-dimensional (3-D) marker trajectories, ground reaction forces, EMG, and exoskeleton encoder data. A 16-camera motion capture system (Vicon V8 and Nexus, Vicon Motion Systems, Oxford, UK) was used to track retro-reflective markers at 100 Hz. The markers were placed on the participant and the exoskeleton to capture the position and orientation of all the human body and exoskeleton segments (Figs 1A and 1B). The markers were placed on the participant according to the Conventional Gait Model 2.5 template in Vicon Nexus [52]. Markers were placed on the exoskeleton based on a custom template (Figs 1A and 1B). Next, we recorded ground reaction forces from each foot using four overground force plates (Bertec Corp., Columbus, OH) sampled at 2000 Hz. The force data were filtered using a low-pass, fourth-order Butterworth filter with a cutoff frequency of 15 Hz.
We recorded muscle EMG measurements from both legs using a 16-channel surface system (TrignoTM, Delsys Inc., Natick, MA) sampled at 2000 Hz. We measured EMG from rectus femoris, vastus lateralis, vastus medialis, semitendinosus, biceps femoris, gastrocnemius medialis, soleus, and tibialis anterior muscles using established protocols [54, 55]. The participant’s mean resting EMG was determined from unassisted standing trials with the participant instructed to remain stationary in a neutral pose with their muscles relaxed. The mean resting EMG value was subtracted from their raw EMG values from the locomotion trials to offset the data to zero. We filtered the EMG data using a fourth-order high-pass filter (30 Hz) to remove motion artifact and then full-wave rectified and filtered with a fourth-order Butterworth low-pass filter (6 Hz) to obtain the linear envelope of muscle activation [56]. The filtered EMG data were normalized to muscle-specific activations obtained from maximum voluntary contraction trials. In addition, we recorded exoskeleton encoder data from all motion capture trials, including timestamps corresponding to the activation of the motors and motor encoder angles. These data were synchronized with the 3-D marker trajectory, ground reaction force, and EMG data.
Motion capture of exoskeletal-assisted and unassisted walking
We acquired 3-D motion data while the participant walked back and forth over a 5.0 x 2.5 m even instrumented walkway with and without the exoskeleton. The participant walked 10 times back and forth with the exoskeleton at the participant’s preferred speed (0.47 ± 0.03 m/s), and 10 times unassisted at self-selected speed (1.11 ± 0.07 m/s). From this dataset, only successful trials were included for further analysis. A walking trial was determined as successful if the foot placements were entirely on single force plates and there were no missing marker, ground reaction force, and EMG data. Additional criteria for a successful trial during exoskeletal-assisting walking was that the hand crutches were placed away from the force plates so as not to interfere with the ground reaction forces from the feet and there were no missing exoskeleton encoder data. Based on these criteria, we obtained six successful trials each of exoskeletal-assisted and unassisted walking.
Motion capture of exoskeletal-assisted and unassisted sit-to-stand and stand-to-sit maneuvers
We acquired 3-D motion data while the participant performed sit-to-stand and stand-to-sit maneuvers with and without the exoskeleton. The participant used an armless piano bench with seat height of 43 cm positioned on the overground walkway that enabled each foot to be placed on a force plate during the entire duration of a maneuver. The participant performed the sit-to-stand and stand-to-sit maneuvers 10 times each with and without the exoskeleton. From this dataset, only successful trials were included for further analysis. A trial was determined as successful if the foot placements were entirely on single force plates and there were no missing marker, ground reaction force, and EMG data. Additional criteria for a successful trial during exoskeletal-assisting sit-to-stand/stand-to-sit were that the hand crutches were placed away from the force plates so as not to interfere with the ground reaction forces from the feet and there were no missing exoskeleton encoder data. Based on these criteria, we obtained four and five successful trials of exoskeletal-assisted and unassisted sit-to-stand maneuvers, respectively, and five successful trials each of exoskeletal-assisted and unassisted stand-to-sit maneuvers.
Virtual simulator reproducing exoskeletal-assisted and unassisted locomotion
We developed a virtual simulator in OpenSim [57] to reproduce exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers (Fig 1C-D). We integrated a previously published human musculoskeletal model [53] with a full-scale model of the ReWalk P6.0 exoskeleton. The musculoskeletal model comprised 24 segments and 37 degrees of freedom (DoFs): seven in each leg, six at the pelvis, three at the torso, and seven in each arm [53]. The seven DoFs in each leg included three DoFs at the ball-and-socket hip joint, a one DoF coupled knee mechanism with translations of the tibia and patella prescribed by the knee flexion angle, and one DoF revolute joints at the ankle, subtalar, and metatarsal joints. A six DoF universal joint at the pelvis was used to describe the pose of the musculoskeletal model with respect to the global origin. The three DoFs in the torso and upper body included a spherical joint connecting the torso to the pelvis. The seven DoFs in each arm included three DoFs at the ball-and-socket shoulder joint, one DoF revolute joint at the elbow, one DoF revolute joint between the radius and the ulna, and two DoFs universal joint between the radius and the hand for wrist flexion-extension and radial-ulnar deviation. The full-scale model of the exoskeleton comprised seven segments and four DoFs (Fig 1C). The four DoFs included a single DoF pin joint each at the hips and knees of the exoskeleton. The ankle joints in the exoskeleton were welded to the lower leg segments. Next, human-robot interactions were modeled using an additional six DoF universal joint at the pelvic band to anchor the exoskeleton to the participant (Fig 1C). The musculoskeletal model included 80 Hill-type muscle-tendon units [58]. The Hill-type muscle-tendon units captured the force-length-velocity properties of the lower extremity muscles, with muscle geometry and architecture based on adult cadaver data [59].
We adapted a previously published computational framework in OpenSim [53, 60] to determine the dynamics of exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers. We scaled the generic musculoskeletal model to match the mass and segment lengths of the participant. We determined joint kinematics by performing Inverse Kinematics (IK) analyses in OpenSim. IK solves for kinematics by minimizing error between the experimentally measured marker positions and the corresponding markers on the human-robot model. We performed Inverse Dynamics (ID) analyses to compute net human-robot joint torques. For ID analyses, we simplified the human-robot model by incorporating the mass and inertial properties of the exoskeleton with the musculoskeletal segments, similar to previous studies [61, 62]. Exoskeleton masses were added to the respective segments in the musculoskeletal model. The centers of mass of the segments in the musculoskeletal model were moved to the locations corresponding to the combined human-robot segments.
Data analysis and statistical methods
We compared hip, knee, and ankle flexion-extension angles from each leg during exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers. For exoskeletal-assisted maneuvers, we determined the hip and knee flexion-extension angles from the robot using IK and exoskeleton encoder data; the RMS errors between these two methods were minimal (the worst case was 3.3°), as such only joint angles from IK are presented in this study. The joint angles from multiple trials were averaged for each leg. We compared the vertical, anterior-posterior (AP), and medial-lateral (ML) ground reaction forces from each leg during exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers. The ground reaction forces from EAL were normalized to the combined weight of the participant and the exoskeleton; forces from unassisted locomotion were normalized to the participant’s body weight. Ground reaction forces from multiple trials were averaged for each leg. Next, we compared normalized EMG from the eight lower extremity muscles from each leg during exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers. Normalized EMG from multiple trials were averaged for each leg. Finally, we compared hip, knee, and ankle moments from each leg during exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers. The joint moments from EAL were normalized to the combined mass of the participant and the exoskeleton; joint moments from unassisted locomotion were normalized to the participant’s mass. The joint moments from multiple trials were averaged for each leg.
Results
Exoskeletal-assisted and unassisted walking
The virtual simulator reproduced exoskeletal-assisted and unassisted walking within acceptable tolerances, with average RMS errors between experiment and simulator markers being 1.3 cm and 1.2 cm for exoskeletal-assisted and unassisted trials, respectively. We observed differences in joint kinematics between exoskeletal-assisted and unassisted walking (Fig 1). Mean peak hip extension angles of the human during exoskeletal-assisted walking were 3.1° ± 1.9° (left) and 0.2° ± 0.9° (right), compared to 27.2° ± 1.2° (left) and 25.1° ± 1.0° (right) during unassisted walking (Figs 2A and 2B). After subtracting the offset angles due to different definitions of coordinate axes, average absolute difference between human and robot hip flexion-extension angles were 1.5° ± 1.4° (left) and 1.9° ± 1.0° (right). Mean peak knee flexion angles of the participant during exoskeletal-assisted walking were 50.1° ± 0.6° (left) and 52.6° ± 0.7° (right), compared to 68.6° ± 0.3° (left) and 70.7° ± 1.1° (right) during unassisted walking (Figs 2C and 2D). After subtracting the offset angles, average absolute difference between participant and robot mean knee flexion-extension angles were 1.7° ± 1.1° (left) and 0.8° ± 0.5° (right). Mean peak ankle plantarflexion angles of the participant during exoskeletal-assisted walking were 6.9° ± 2.8° (left) and 6.7° ± 3.9° (right), compared to 13.9° ± 2.5° (left) and 19.5° ± 1.9° (right) during unassisted walking (Figs 2E and 2F). Peak plantarflexion during exoskeletal-assisted walking occurred in early stance during braking, in contrast to peak plantarflexion at the end of propulsion (toe-off) during unassisted walking. We observed a minimal range of plantarflexion (5.3°-10.5°) in the propulsion phase during exoskeletal-assisted walking, compared to a larger range (27.1°-30.3°) during unassisted walking.
We observed differences in normalized ground reaction forces between exoskeletal-assisted and unassisted walking, especially during the braking and propulsion phases (Fig 3). Mean peak vertical ground reaction forces during braking were 0.91 ± 0.08 BW (left) and 0.89 ± 0.06 BW (right) during exoskeletal-assisted walking, compared to 1.09± 0.02 BW (left) and 1.05 ± 0.05 BW (right) during unassisted walking (Figs 3A and 3B). Mean peak vertical ground reaction forces during propulsion were 0.98 ± 0.02 BW (left) and 1.02 ± 0.02 BW (right) during exoskeletal-assisted walking, compared to 1.04 ± 0.04 BW (left) and 1.02 ± 0.02 BW (right) during unassisted walking (Figs 3A and 3B). Mean peak posterior ground reaction forces during braking were 0.08 ± 0.04 BW (left) and 0.08 ± 0.02 BW (right) during exoskeletal-assisted walking, compared to 0.17 ± 0.01 BW (left) and 0.17 ± 0.04 BW (right) during unassisted walking (Figs 3C and 3D). Mean peak anterior ground reaction forces during propulsion were 0.06 ± 0.02 BW (left) and 0.08 ± 0.06 BW (right) during exoskeletal-assisted walking, compared to 0.18 ± 0.01 BW (left) and 0.19 ± 0.01 BW (right) during unassisted walking (Figs 3C and 3D). Mean peak lateral ground reaction forces during braking were 0.07 ± 0.02 BW (left) and 0.06 ± 0.01 BW (right) during exoskeletal-assisted walking, compared to 0.09 ± 0.01 BW (left) and 0.09 ± 0.01 BW (right) during unassisted walking (Figs 3E and 3F). Mean peak lateral ground reaction forces during propulsion 0.06 ± 0.01 BW (left) and 0.07 ± 0.01 BW (right) during exoskeletal-assisted walking, compared to 0.08 ± 0.01 BW (left) and 0.10 ± 0.01 BW (right) during unassisted walking (Figs 3E and 3F). The decreased ground reaction forces during the braking and propulsion phases of exoskeletal-assisted walking were supported by EMG data (Fig 4). The tibialis anterior muscle, used for braking, was less active during the braking and terminal swing (to prepare for braking) phases during exoskeletal-assisted compared to unassisted walking (Figs 4O and 4P). The gastrocnemius medialis and soleus muscles, used for propulsion, were less active during the propulsion phase during exoskeletal-assisted compared to unassisted walking (Figs 4K-4N).
We observed differences in joint moments between exoskeletal-assisted and unassisted walking (Fig 5). Mean peak hip flexion moments of the participant during exoskeletal-assisted walking 0.23 ± 0.05 Nm/kg (left) and 0.35 ± 0.06 Nm/kg (right), compared to 0.65 ± 0.07 Nm/kg (left) and 0.64 ± 0.09 Nm/kg (right) during unassisted walking (Figs 5A and 5B). Mean peak knee extension moments of the participant during exoskeletal-assisted walking were 0.10 ± 0.10 Nm/kg (left) and 0.22 ± 0.11 Nm/kg (right), compared to 0.64 ± 0.07 Nm/kg (left) and 0.73 ± 0.10 Nm/kg (right) during unassisted walking (Figs 5C and 5D). Mean peak ankle plantarflexion moments during exoskeletal-assisted walking were 0.80 ± 0.12 Nm/kg (left) and 0.78 ± 0.19 Nm/kg (right), compared to 1.29 ± 0.04 Nm/kg (left) and 1.12 ± 0.03 Nm/kg (right) during unassisted walking (Figs 5E and 5F).
Exoskeletal-assisted and unassisted sit-to-stand
The virtual simulator reproduced exoskeletal-assisted and unassisted sit-to-stand within acceptable tolerances, with average RMS errors between experiment and simulator markers being 1.5 cm and 1.9 cm for exoskeletal-assisted and unassisted trials, respectively. Exoskeletal-assisted sit-to-stand in the ReWalk P6.0 follows a standard procedure that is different from unassisted sit-to-stand, including different temporal events and phases. Comparison of exoskeletal-assisted and unassisted sit-to-stand is included as supplemental information, including joint kinematics (S1 Fig), ground reaction forces (S2 Fig), EMG (S3 Fig), and joint moments (S4 Fig).
Exoskeletal-assisted and unassisted stand-to-sit
The virtual simulator reproduced exoskeletal-assisted and unassisted stand-to-sit within acceptable tolerances, with average RMS errors between experiment and simulator markers being 1.4 cm and 1.9 cm for exoskeletal-assisted and unassisted trials, respectively. Exoskeletal-assisted stand-to-sit in the ReWalk P6.0 follows a standard procedure that is different from unassisted stand-to-sit, including different temporal events and phases. Comparison of exoskeletal-assisted and unassisted stand-to-sit is included as supplemental information, including joint kinematics (S5 Fig), ground reaction forces (S6 Fig), EMG (S7 Fig), and joint moments (S8 Fig).
Discussion
The goal of this study was to combine controlled experiments with computational modeling to build a virtual simulator of EAL in the ReWalk P6.0, an FDA-approved lower extremity exoskeleton for rehabilitation of patients with neurological disorders. The first objective of this study was to acquire a minimum empirical dataset comprising human-robot kinematics, ground reaction forces, and EMG during exoskeletal-assisted and unassisted locomotion. Here, we present a minimum empirical dataset from a single session of controlled experiments with an able-bodied participant performing exoskeletal-assisted and unassisted walking, sit-to-stand, and stand-to-sit maneuvers. Our results provide a direct comparison of joint kinematics (Fig 2, S1 Fig, and S5 Fig), ground reaction forces (Fig 3, S2 Fig, and S6 Fig), and EMG (Fig 4, S3 Fig, and S7 Fig) during exoskeletal-assisted and unassisted locomotion in an FDA-approved exoskeleton. The second objective of this study was to quantify the dynamics of the human-robot system using a subject-specific virtual simulator reproducing EAL, and to compare with the dynamics of unassisted locomotion. The subject-specific virtual simulator reproduced EAL within acceptable tolerances (RMS errors of 1.2-1.3 cm for walking, 1.5-1.9 cm for sit-to-stand, and 1.4-1.9 cm for stand-to-sit maneuvers). Our results provide a direct comparison of joint moments during exoskeletal-assisted and unassisted locomotion in an FDA-approved exoskeleton (Fig 5, S4 Fig, and S8 Fig).
We are not aware of any prior literature that has provided a direct comparison of joint kinematics, ground reaction forces, EMG, and joint moments from a single session of controlled experiments with and without an FDA-approved exoskeleton. As such, we have compared our findings with data that is available in the literature. Our joint kinematics results from exoskeletal-assisted walking in the ReWalk P6.0 are comparable to those reported by Chang et al. using the Ekso device [28]. Our mean peak hip extension, knee flexion, and ankle plantarflexion angles were 0.2°-3.1°, 50.1°-52.6°, and 6.7°-6.9°, respectively (Fig 2); in comparison, mean peak hip extension, knee flexion, and ankle plantarflexion angles reported in Chang et al. are ∼3.0°-4.0°, ∼48.0°-50.0°, and ∼6.0°-9.0°, respectively [28]. Ekelem et al. reported median peak hip and knee flexion angles of ∼2.0°-16.0° and ∼23.0°-25.0°, respectively, during exoskeletal-assisted walking in the Indego device [27]. Since they do not report peak hip extension angles during stance, a direct comparison with our hip extension angle data is not possible. Their peak knee flexion values (∼23.0°-25.0°) are lower than our values (50.1°-52.6°). This is likely because the two participants in their study had high levels of spasticity; with functional electrical stimulation, they were able to increase the participants’ peak knee flexion angles to ∼35.0°-47.0° [27]. Our hip, knee, and ankle kinematics are in agreement with the results from Talaty et al., who also analyzed exoskeletal-assisted walking in the ReWalk [26]. Next, Fineberg et al. reported vertical ground reaction forces obtained from in-shoe pressure insoles (F-Scan, TekScan, Boston, MA) during exoskeletal-assisted walking in the ReWalk device [14]; our vertical ground reaction forces are in agreement with Fineberg et al. There is no prior study to compare our measured anterior-posterior and medial-lateral ground reaction forces during EAL. Regarding muscle activity during EAL, Bellitto et al. compared EMG data from three incomplete spinal cord injured participants during exoskeletal-assisted (with the Ekso device) and unassisted walking [29]. The timings of our EMG activity are in agreement, but our normalized magnitudes are lower than those reported in Bellitto et al. This is likely because of two reasons: 1) our results are from an able-bodied participant, compared to incomplete spinal cord injured participants in Bellitto et al.; and 2) we normalized our EMG to maximum voluntary contraction data, while Bellitto et al. used a dynamic normalization technique [63], where EMG data from each muscle were normalized to the maximum value over all trials with and without the exoskeleton. There is no prior study to compare our results from sit-to-stand and stand-to-sit maneuvers.
This study provides a much-needed empirical dataset to develop virtual simulators reproducing EAL in FDA-approved devices. To the best of our knowledge, there is not a single virtual simulator of EAL in an FDA-approved exoskeleton. Prior studies have developed virtual simulators of EAL in custom-built exoskeletons [44-47, 49, 64], but these devices are not FDA-approved and are likely many years away from widespread application for the rehabilitation of patients with neurological disorders. Prior studies have also developed virtual simulators of EAL based on idealized or imaginary exoskeletons [37, 40–42, 48, 50]. For example, Bianco et al. developed a virtual simulator to examine how multi-joint assistance affects the metabolic cost of exoskeletal-assisted and unassisted walking [37]. This simulator, however, was not based on a physical exoskeleton and empirical data from unassisted walking was used as model inputs, with the assumption that joint kinematics, ground reaction forces, and EMG do not change between exoskeletal-assisted and unassisted walking. They stated in their limitations that incorporating the kinematic adaptations due to an exoskeleton would alter their metabolic cost predictions. It is likely that Bianco et al. and other studies that developed virtual simulators based on idealized or imaginary exoskeletal-assistance did so because they did not have empirical data from EAL.
A potential limitation of this study is that the ReWalk P6.0 was designed for patients with spinal cord injury, but the results from this study were obtained from an able-bodied participant. These results provide new understanding of the dynamics of the human-robot system during EAL in an FDA-approved device. We anticipate building upon this work by analyzing participants with spinal cord injury. A second limitation of this study is that our EAL experiments did not measure the external crutch reaction forces or the interaction forces due to the knee brackets and straps that were used to secure the participant in the exoskeleton. These forces were also not included in the virtual simulator and, as such, their effects on our findings remain undefined. A third potential limitation is that we incorporated the mass and inertial properties of the exoskeleton with the musculoskeletal segments, similar to that reported in previous studies, and thus simplified our virtual simulator for ID analyses [61, 62]. A more thorough approach would have been to segregate the human and robot components, include human-robot interaction forces as input and to solve the human and robot kinematic chains separately.
This study provides a framework for determining the dynamics of the human-robot system during EAL in an FDA-approved exoskeleton using controlled experiments and a subject-specific virtual simulator. We created a first-of-a-kind empirical dataset comprising joint kinematics, ground reaction forces, EMG, and joint moments from exoskeletal-assisted and unassisted locomotion. The virtual simulator will provide a low-risk and cost-effective platform for parametric studies of human-robot interaction during EAL, a necessary step for researchers and clinicians to characterize the effects of human factors on EAL and establish standards for safe and sustained use of robotic exoskeletons. Furthermore, this virtual simulator will provide a platform for exoskeleton companies to conduct rapid design-phase evaluations to accelerate device refinements. We will make all empirical data and virtual simulator code from this study freely available to the research community. We invite investigators to build on our work to develop accurate virtual simulators reproducing EAL to address a broad range of questions, permitting the design and manufacture of exoskeletal devices that more closely replicate a normal gait pattern.
Data Availability
All empirical data and virtual simulator code from this study will be freely available at simtk.org
Conflict of interest statement
The authors have no conflict of interest to disclose related to this work.
Supporting information
Acknowledgments
We thank Assaf Tzioni from ReWalk Robotics for his support with access to the exoskeleton encoder data. We thank Lina Alsauskaite from ReWalk Robotics for training the research team on safe use of the exoskeleton. ReWalk Robotics did not play any role in study design, participant recruitment, methodology, and data analysis.