Study design
This study was designed to assess the safety and efficacy of using a bilateral hip-based exoskeleton to improve gait function in community dwelling older adults. We utilized both the assistive and resistive modes of the device to personalize the intensity of exercise and to provide an active recovery period. Twelve participants over 65 years old completed a total of twelve 30-min gait training sessions over a period of 4–6 weeks using the GEMS-H. All gait training sessions were completed in the community spaces at The Merion, a senior living community located in Evanston, Illinois.
Functional outcome measures
Outcomes were collected at two separate time points in order to compare pre- and post-intervention status. Functional outcome measures included five times sit-to-stand (5xSTS), ten meter walk test (10MWT), six minute walk test (6MWT), Berg-Balance Scale (BBS), and functional gait assessment (FGA). Patient-Reported Outcomes included the Quebec User Evaluation of Satisfaction with assistive Technology (QUEST). These outcomes were selected based on their common use in prior literature assessing gait function, balance, endurance, and device satisfaction.
Activity monitoring
Each participant’s activity levels were monitored using an ankle-worn sensor (ActiGraph wGT3X-BT; ActiGraph LLC., Pensacola, FL). For sedentary analysis, participant’s activity was monitored for six days prior to the baseline session, and for 6 days following the final assessment. ActiGraph recorded 3D-acceleration data at a sampling frequency of 30 Hz. Raw acceleration data is analyzed using ActiGraph’s proprietary software (ActiLife 6.13).
Sedentary time/inactivity analysis
Each participant was issued an ActiGraph system and instructed to wear the device at all times for six consecutive days before and after the training intervention. This data was analyzed using a 10-s actigraph epoch data format and data were subjected to the Choi algorithm for wear time and non-wear time separation [19]. Further, ActiGraph data was only analyzed for sedentary behavior between 5 AM and 11 PM, assuming most physical activity happens during waking hours, and this was verified with individual reports. Active to sedentary transition was characterized by a boundary condition of three or more minutes of inactivity. This analysis resulted in both the number and duration of sedentary bouts each day, where a bout duration is defined as the number of consecutive minutes spent in a sedentary state. After six consecutive days of data collection, the average daily bout number and duration were calculated. These daily averages from the PRE intervention state to the POST intervention state were compared using a one-tailed paired t-test and \(\alpha\)=0.05. The sedentary analysis procedures were based on literature related to aging research [20, 21].
Participants and inclusion/exclusion criteria
Twelve individuals above the age of 65 years old were recruited to participate in twelve sessions that occurred 2–3 times per week over a 4–6 week period using the GEMS-H. In addition to selecting individuals over the age of 65, qualifying participants also had to be able to walk with or without an assistive device for greater than three meters. Medical clearance was obtained from each participant’s primary physician prior to training with the device. Participants were excluded from recruitment if they were unable to comprehend or provide consent, were unable to physically fit within the device, or had any significant neurological diagnoses that would impact safe use of the device.
Hip assist exoskeleton and personalized tuning
The GEMS-H is a hip-based robotic exoskeleton worn around the waist and fastened to the thighs to assist with hip flexion and extension (Fig. 1). The GEMS-H device has a pair of actuators that generate assistive or resistive forces at each hip joint. The device weighs 2.1 kg and comes in three sizes. The width of each version can be adjusted to fit individual body size. There are magnetic joint angle sensors in each hip of the exoskeleton to continuously track the user’s kinematics and provide feedback to the controller which stabilizes assistance/resistance based on instantaneous user needs.
DOFC for gait assistance and resistance
The GEMS-H implements a simple self-excited DOFC method to generate assistive and resistive torques. As shown in Fig. 2, the DOFC method does not include a gait phase estimator or a reference lookup for generating torque profiles, yet it can be generalized to operate under various walking speeds and environments (e.g., stairs and ramps) without the need for task-specific parameter adjustments [12, 13]. The assistance/resistance torque is immediately applied following the movement of the user by reflecting the change of leg motion from reading the wearers hip kinematics at every control period (= 0.01 s i.e. 100 Hz) [12, 13]. This time delayed, self-feedback controller approach is known for stabilizing oscillatory systems like human locomotion, and can be generalized to operate under various walking speeds and environment (e.g., stairs and ramps) without the need for task-specific parameter adjustments [12, 13]. The magnitude of assistance/resistance can be varied over a range of values to personalize the external assistance/resistance to every user’s self-chosen comfort level. An in-depth description of this controller and earlier hardware design are described in previously published manuscripts [12, 13, 22].
As the current study is the first time older adults trained with this exoskeleton, we required the engagement of clinicians to train our participants to safely use the device. Thus, based on clinician feedback and practicality of using in real-world settings, we chose two control parameters that the clinicians could tune to personalize the device setting for each user. These are the feedback gain \(\kappa\)(positive gain for assistive torque and negative gain for resistive torque respectively) the feedback time delay \(\Delta t\). The magnitude of gain determines the strength of the assistive/resistive torque generated. The combination of these two settings will determine the type of torque generated, i.e. assistive or resistive, and the timing of torque input with respect to the gait cycle.
Smoothing hip motion state with low pass filter
Figure 2 show the hip assistance and resistance strategies with DOFC framework. We define an output state \({y}_{\mathrm{raw}}\) as the ground projected leg motion:
$${y}_{\mathrm{raw}}\left(t\right)=\mathrm{sin}{q}_{\mathrm{r}}\left(t\right)-\mathrm{sin}{q}_{\mathrm{l}}(t),$$
(1)
where \({q}_{\mathrm{r}}\) and \({q}_{\mathrm{l}}\) are the right and left joint angles, respectively. For both terms, hip extension is considered a positive angle.
The original noisy output state is smoothed by passing it through a simple first-order low-pass filter (also known as an exponential moving average filter):
$${y}^{\mathrm{cur}}={(1-\alpha )y}^{\mathrm{prv}}+\alpha {{y}_{\mathrm{raw}}}^{\mathrm{cur}},$$
(2)
where \({y}^{\mathrm{cur}}\) denotes the currently smoothed output state, \({y}^{\mathrm{prv}}\) is the previously smoothed state, \({{y}_{\mathrm{raw}}}^{\mathrm{cur}}\) is the currently sensed original state, \(\alpha\) is the smoothing factor. The current smoothed state \({y}^{\mathrm{cur}}\) is expressed as a weighted sum of the previous sample time state \({y}^{\mathrm{prv}}\) and the original state value of the current sample time \({{y}_{\mathrm{raw}}}^{\mathrm{cur}}\), and the smoothing rate can be modified by changing the smoothing factor. The − 3 dB cutoff frequency (the frequency over which the signal power is halved, denoted \({f}_{c}\)), is given by this equation for discrete time systems:
$${f}_{c}=\frac{{f}_{s}}{2\pi }{\mathrm{cos}}^{-1}\left(1- \frac{{\alpha }^{2}}{2(1-\alpha )}\right),$$
(3)
where \({f}_{s}\) is the sampling frequency. The smoothing factor \(\alpha\) = 0.05 in (2) was selected to generate the smoothed interaction torque. Combined with our sampling rate \({f}_{s}=100 Hz\), the − 3 dB cutoff frequency is 0.8165 Hz.
Assistive or resistive torque generation from delayed feedback state
The assistive or resistive torque \(\tau\) is generated through a combination of appropriate time delays \(\Delta t\) and positive (assistive) or negative (resistive) gains \(\kappa\):
$$\tau \left(t\right)=\kappa y\left(t-\Delta t\right)=\left\{\begin{array}{c}\kappa >0, {\text{assist mode}}\\ \kappa <0, {\text{resist mode.}}\end{array}\right.$$
(4)
The base control strategy in Fig. 2 can be extended for both right/left hip torque generation \({\tau }_{\mathrm{r},\mathrm{des}}\), \({\tau }_{\mathrm{l},\mathrm{des}}\) by modifying the original torque equation in (4).
The terms for right hip flexion, left hip extension are:
$${\tau }_{\mathrm{r},\mathrm{des}}\left(t\right)=-\tau \left(t\right)$$
(5)
$${\tau }_{\mathrm{l},\mathrm{des}}\left(t\right)=\tau \left(t\right)\cdot \delta ,$$
whereas for left hip flexion, right hip extension:
$${\tau }_{\mathrm{l},\mathrm{des}}\left(t\right)=\tau \left(t\right)$$
(6)
$${\tau }_{\mathrm{r},\mathrm{des}}\left(t\right)=-\tau \left(t\right)\cdot \delta ,$$
where \(\delta\) denotes the hip extension-flexion torque ratio. Torque in the direction of hip extension is considered positive. The extension-flexion ratio \(\delta =\) 1 was used in this study to generate equal hip extension and flexion torque strength [13].
Device interface
The device is controlled through a custom built application on a hand-held tablet. Through the application, the trained physical therapist assisting each participant is able to turn on/off torque, switch between assist and resistance modes, and modify the gain (\(\kappa )\) and delay (\(\Delta t\)) parameters. Gain increases or decreases the amplitude of assistance or resistance. The maximum value for gain in assistance mode is 15 (about 12 Nm peak torque), while the maximum (absolute) value in resistance mode is − 5 (about − 4 Nm peak torque). Delay allows the assistance or resistance to be applied earlier or later in the gait cycle. The range of delay is between 0.15 and 0.25 s. The tablet also displays real time information such as joint angle and torque values.
Training progression
The GEMS-H training program was based on prior systems designed to improve the walking performance of the user [24, 25]. For this study, training sessions were conducted in regions of the participant’s community living facility, including indoor hallways, ramps, curbs, stairs, and multi-terrain surfaces. Older adults traditionally struggle with motivation to walk due to physical impairment and fear of falling. To encourage walking in the community environment, the physical therapists alternated the exoskeleton settings between resistance torque to help strengthen muscles and assistance torque to provide an active rest between resistance training. Seated breaks were allowed as necessary, but assistance mode was preferred when possible. Every subject’s first training session had at least ten minutes of resisted walking during their 30-min training sessions (in accordance to their ability), but the physical therapists sought to increase the time in resistance mode gradually throughout the 12 sessions.
Physical therapists were also able to adjust the resistance gain, assistance gain, and time delay to help tune the device to an appropriate resistance/assistance level and match the user’s gait. These parameters would change over the course of the twelve training sessions to match the subject’s change in gait, strength, or endurance. Priority was placed on increasing resistance and decreasing assistance during each new training session, in order to encourage progressive training.
After selecting the resistance time and the assist/resist gains, activities during each session were selected and modified based on the subject's ability to perform them successfully while providing a challenge. Activities were selected based on the following progression of increasing difficulty: level ground walking, speed changes, multi-directional/backward stepping, inclines/ramps, stair climbing, and obstacle negotiation. Obstacle negotiation included stepping over, weaving between, or stepping onto selected obstacles. Following this protocol, subjects 2 and 10 were additionally challenged by not always using their assistive devices (a cane and rolling walker, respectively) during the walking training, see Additional file 1: Table S1 for more details. All training was done under the supervision of a trained physical therapist, who would guard participants to prevent falls, particularly in the case that the subject’s regular assistive device was not being used.
Statistical analysis
All outcomes/values are presented as mean ± standard deviation (SD), and the alpha value was set at P < 0.05 for indicating significance, and unless otherwise noted normality assumptions were checked and appear reasonable. Two-tailed paired t tests were used to compare the outcomes from pre and post testing. SigmaPlot 14.0 (Systat Software Inc., San Jose, CA, USA) was used to perform all statistical analyses.