Several different worn exoskeletons have been developed to support the elderly's daily activities [1,2,3,4,5,6]. A number of exoskeletons based on quasi-passive elements and soft materials have also been recently developed [7, 8] for the elderly, but their functionality is limited in the sense that it is very challenging for these exoskeletons to provide a varying level of desired assistance to the user over the complete gait cycle. Therefore, for worn exoskeletons current-controlled DC drives (CCDC drives) are still the most recommended joint actuators to use for their superior load torque compensation performance, quick response and their ability to provide the desired level of varying assistance to the user [9]. Lower weight requirement for these exoskeletons limits the number and size of the joint actuators to be used. This weight reduction invariably implies a lower count and size of the joint actuators. Hence, the low number of joint actuators entails a low active degree of freedom (LADOF) for the exoskeleton, while the smaller actuator size implies limited actuator bandwidth and gear ratio [10]. This limitation, combined with rigid contact supports (primarily used to attach the human user physically), deteriorates the physical human–robotic interaction (pHRI) performance, primarily due to four main reasons. Firstly, due to limited gear ratio, the disturbing net load torque on the joint actuators is not sufficiently reduced to be ignored and severely deteriorates the joint actuator's tracking performance [9]. This joint net load torque, which arises from a complex interaction between the exoskeleton and the human limbs at multiple contact points, is highly nonlinear, uncertain and coupled. All the nonlinearity of the human exoskeleton joint space dynamics is contained in this net joint load torque [9]. Therefore, it is of paramount importance that this disturbing load torque on the exoskeleton's joint actuators is minimized to linearize the human exoskeleton system and allow the design of simple linear joint controllers for the system, with improved tracking performance. Secondly, the limited power actuators suffer from limited force bandwidth and actuation signal saturation. In turn, this inadequacy results in poor transient response and high impedance due to the reflected moment of inertia [11, 12]. Therefore, the performance of any impedance control strategy [13,14,15,16,17] trying to reduce the endpoint contact impedance of the exoskeleton gets severely affected. The desired contact point impedance is very difficult to realize in practice due to limited force bandwidth and actuation signal limitation of the actuators [10]. This issue, combined with rigid contact supports for a LADOF exoskeleton, strongly affects the assistive device's pHRI performance. The effect of rigid contact supports on pHRI for a LADOF upper body exoskeleton has been analyzed in detail in [10]. Thirdly, the exoskeleton's active degree of freedom (ADOF) is reduced due to fewer actuators. This, coupled with the unavoidable slight axial misalignment between the human and the exoskeleton joints, using rigid contact support results in increased and fast undesired interactive forces between the human and assistive devices [10]. This effect, combined with the actuator's limited force bandwidth, strongly reduces the exoskeleton's pHRI performance. Lastly, due to high load torques and large controller bandwidths, the joint actuators can potentially go into saturation which can significantly reduce their tracking performance and, in turn, can strongly reduce the pHRI performance of the exoskeleton [18].
ARCSEC DRIVE supply professional and honest service.
Quite recently, several techniques have been proposed to improve the pHRI performance of the LADOF exoskeletons, suffering due to the factors mentioned above [9, 10, 18]. For example, it has been shown in [10] (for a 4-ADOF fixed upper body exoskeleton test rig) that instead of using the classical rigid supports, the end support impedance control performance in task space is effectively improved (with reduced undesired interactive forces) by using 3-D (mechanically decoupled) passive compliant supports. These passive compliant supports contain three passive compliant elements that can independently sense the 3-D interactive forces between the human and the exoskeleton. Furthermore, it has been mathematically shown in [10] that the stiffness of the respective passive compliant element of the support limits the maximum contact point impedance of the exoskeleton in a particular direction. It is also shown theoretically and practically in [10] using the upper body test rig that this approach results in an improved and safer pHRI performance for an assistive device.
To efficiently compensate the coupled nonlinear human–machine joint torques and to effectively linearize the human–machine joint dynamics, a joint space disturbance observer-based dynamic load torque compensator (DOB-based-DLTC) is initially proposed in [9], with its vectorial form recently proposed in [10]. Furthermore, a stability and performance analysis of the proposed DOB-based-DLTC w.r.t to different bandwidths has been recently presented in [18] along with a hybrid switching strategy to prevent the joint actuators from going into saturation. It has been shown in [18] that this approach simultaneously improves the stability and performance of the proposed DOB-based-DLTC in [9] and, in turn, the pHRI performance of the exoskeleton.
The work proposed in this paper is in continuation of the work done in [1, 2, 9, 10, 18, 19]. The motivation for this work is to design and develop a lower body exoskeleton so that the latest developed techniques in [9] [18] and [10] could be effectively applied to improve the pHRI performance of lower body exoskeletons. The exoskeleton initially developed in [1, 19] is designed to follow particular defined desired trajectories in joint space, resulting in strong interactive forces between the human and the exoskeleton. The exoskeleton also suffers from considerable weight and inertia. These drawbacks, coupled with very rigid contact supports, result in poor pHRI performance of this exoskeleton. The lower body exoskeleton designed in [2] has a lower weight and uses standard force control techniques to follow the human movement, but uses rigid contact supports, which result in interactive forces, not in the desired direction, which causes the force sensors at the supports to get corrupted and hence results in poor force control and poor pHRI performance for the exoskeleton.
Therefore, to practically verify the pHRI improvement for an actual wearable lower body exoskeleton, afforded by the improved task space impedance control techniques (with mechanically decoupled passive compliant contact supports) as proposed in [10], in conjunction with the new joint space load torque compensation techniques as proposed in [9, 10, 18], the design of a novel 4-ADOF lower body exoskeleton is presented with four passive degrees of freedom (PDOF) per leg. The mechanical design is presented in detail in Sect. 2. These load torque compensation techniques require accurate sensing of the joint load torques; therefore, innovative compact joint assemblies have been designed with integrated torque sensors to correctly sense the joint load torques (as shown in Sect. 2) and with incremental and absolute encoders to sense the joint positions and velocities accurately as well. Novel contact supports for thigh and shank have also been designed, with mechanically decoupled 3-D passive compliance to allow the implementation of the improved impedance control techniques in [10] as shown in Sect. 2. Special design considerations have been considered to reduce the undesired interactive forces between the human and assistive device. A methodology of selecting and verifying the joint actuators and estimating the desired assistive forces at the contact supports based on human user joint torque requirements and the degree of assistance is also thoroughly presented in Sect. 3. A new CAN bus-based distributive master–slave control architecture has been proposed in Sect. 5 to properly implement the new techniques based on [9, 10, 18]. Operational state flowcharts for both the master and slave controllers are also proposed in Sect. 5. The design details of the developed embedded control card and control box are also presented in the same section. A new control strategy capable of imparting simultaneous impedance-based force tracking control at the passive, compliant supports of the exoskeleton while using DOB-based-DLTC at the joint space is presented in Sect. 6. Experimental gait data-based simulation results, justifying the use of passive compliant arm supports in combination with DOB-based-DLTC for assistive exoskeletons, employing the proposed control strategy, are finally presented in Sect. 7.
To provide physical assistance to the human user for his lower body daily activities (walking, sit to stand), a 4-ADOF LADOF exoskeleton is designed as shown in Fig. 1. Apart from reducing the ADOF for the exoskeleton, special design considerations are considered to reduce the exoskeleton's overall estimated weight (without the control box) to be less than 23 kg. Since joint actuators are the biggest contributed to the exoskeleton’s overall weight, their proper selection is pivotal in this regard. Proper selection of joint actuators in turn requires accurate estimation of each joint’s load torque and power requirements. (These requirements have been found and are presented in detail in Sect. 3.2 and Sect. 3.3) Furthermore, to ensure that the designed exoskeleton complies with low-risk physical assistance robot standards, namely EN-ISO , the exoskeleton is designed to provide a maximum of 50% assistance to human users weighing 70 –110 kg with a height from 1.5 to 1.8 m. To reduce the size and weight of the exoskeleton but still be able to assist the human effectively, the actuators for hip and knee joints were selected to meet torque and power requirements for a medium gait speed of \(2.4\text{ km}/\text{h}\). State-of-the-art harmonic drives (as described in detail in Sect. 2.2) were used as gearheads for each actuator to reduce the weight and size of joint assemblies and, in turn, the overall weight of the exoskeleton.
Aluminum alloy with high tensile strength and low weight density properties was used to design a lightweight but robust structure of the exoskeleton (as described in Sect. 2.1). Since the ADOFs for both the exoskeleton's hip and knee joints are along the sagittal plane, the respective 3-D compliant contact supports are designed to sense the contact forces only along the same plane. To properly sense these contact forces, the compliant contact supports for the thigh and shank are correspondingly attached to their respective support structures near the knee and ankle joint assemblies, as shown in Fig. 3. The exoskeleton's thigh, shank and waist structures are adjustable using respective unique sliding adjustment mechanisms to allow proper fitting for users with different body frames and physiques.
The sliding mechanism for the thigh support structure is shown explicitly in Fig. 3. The dimensions of these structures are shown in Fig. 2, while the corresponding values are shown in Table 1. The designed range of motion (ROM) values for the different DOF are listed in Table 2.
The proposed structural design for the designed exoskeleton is shown in Fig. 2. The structural design of the proposed exoskeleton is based on some initial work done in [2]. Special aluminum grade material namely AL--T73 was used to design the exoskeleton. This material promises high tensile and yield strengths but with low weight density which in turn has enabled a robust structure design but with reduced overall weight. FEM-based stress analysis simulations were done in Autodesk Inventor™ to optimize the structure design for strength and robustness. To effectively assist the human in his daily activities while ensuring a reduced weight of the exoskeleton, assistance to the human is provided only in the sagittal plane of hip and knee joints (for respective flexion/extension DOFs). On the other hand, to allow the human's unassisted natural movement, the exoskeleton is also designed to support 8-PDOF with 4-PDOF per leg. Furthermore, for each hip joints, internal–external rotation and abduction–adduction DOFs are passive, while for each ankle joint DOFs, i.e., dorsi–plantar flexion and inversion–eversion DOFs are passive, as listed in Table 2. The supported active and passive DOFs for the designed exoskeleton are also, respectively, shown in Fig. 2.
New joint space load torque compensation techniques like DOB-based-DLTC and hybrid DOB-based-DLTC have been proposed in [9, 18]. These techniques are shown to effectively compensate for the highly uncertain and nonlinear joint torques at the exoskeleton's active joints due to nonlinear human–machine dynamics and interaction. For proper implementation, these compensation techniques require accurate sensing of the joint load torques at all the exoskeleton's active joints. The proper implementation of the impedance control technique proposed in [10], on the other hand, requires accurate sensing of the joint position and velocity along with a proper gear head for all the active joints of the exoskeleton. Furthermore, the joint assemblies need to be designed with the lowest possible weight and size, making the design and concept of such an assembly quite challenging. Therefore, novel joint assemblies are designed to meet the above requirements compactly to ensure proper implementation of the proposed techniques for all four active joints of the lower body exoskeleton in Fig. 1. Each joint assembly primarily consists of a motor torque–sensor assembly, a harmonic drive as a gearhead and an absolute encoder with housing for absolute position sensing, as shown in Fig. 4c. The cross-sectional view of the developed joint assembly is shown in detail in Fig. 4d. The motor torque–sensor assembly, in turn, consists of a brushless DC motor as a joint actuator, an incremental encoder as a joint speed sensor and an integrated torque sensor with proper housing. The designed motor torque–sensor assembly is shown in Fig. 4a. The load torque sensor is proposed to be designed with three sensing beams to effectively sense the axial load torques about the axis of rotation of the respective joint. Four strain gauges on each side of the sensing beams are to be mounted in a full-bridge configuration for enhanced sensitivity, effective noise cancelation and efficient temperature compensation [20, 21]. The housing of the torque sensor is so designed that only the axial rotational torques are to be transmitted to the sensor while all the other undesired torques primarily due to axial or radial load forces are to be taken up by the housing, using two independent support bearings, as shown in the cross-sectional view in Fig. 4b. This, hence, in turn, ensures accurate sensing of the shaft load torque.
To improve the endpoint impedance control performance of the designed lower body exoskeleton so as to reduce the undesired interactive forces between the human and the machine, innovative 3-D passive compliant limb supports are proposed for the respective thigh and shank limbs, as shown in Fig. 1. The design of passive compliant supports for the lower body exoskeleton is primarily motivated by [10], where improvement in pHRI by using such supports has been demonstrated for an impedance-controlled upper body exoskeleton. The designed thigh compliant support for the lower body exoskeleton is shown explicitly in Fig. 5, while its attachment to the respective support structure is shown in Fig. 3. A load sensor is proposed along the Y-axis of the support to sense the sagittal plane's interactive forces. Since the exoskeleton is designed to support the human only along the sagittal plane, only one such load sensor is required per support. It is imperative from a functionality perspective that mechanically decoupled passive compliance is provided by the support while maintaining low friction and undesired play [10]. This requirement is ensured by suggesting a pair of passive springs about an independent miniature roller-based slide table along each axis, as shown in Fig. 5. A roller-based slide table for each axis is suggested to ensures mechanical decoupling and low friction, which is imperative for imparting the desired level of passive compliance along each support axis. Limited PDOF is also suggested about the Y-axis to allow proper alignment of the human limb with that of the support axis, as illustrated in Fig. 5. Although from a control perspective, the passive compliance only along the support direction, i.e., Y-axis, is of relevance, the proposed passive compliance along the X- and Z-axes and the limited PDOF about Y-axis provides the necessary flexibility (for human limb agronomics) required to protect the load cell sensor against the forces not aligned with the desired sensing direction (i.e., the Y-axis). In turn, this design approach ensures accurate sensing of the interactive forces along the direction of control and suggests better fitting and comfort to the human by reducing the amount of stress and undesired interactive forces in line with the non-supportive directions. A combination of Velcro belt with mechanical back limb support is designed to attach the respective human limb support, as shown in Fig. 5.
To properly assist the human during each phase of the walking cycle (stance phase, single-support phase and double-support phase), each phase's start and end instant must be appropriately measured [22,23,24]. Therefore, a 2-PDOF foot assembly is designed for each exoskeleton leg with two load cells, one at the toe to detect toe-off and the other at the heel to detect heel strike. The load cells are suggested to be mounted through a separate pressure plate with sliding steel pins to allow proper force application and sensing, as shown in Fig. 6. The two PDOFs, on the other hand, provide the necessary freedom to the human user for proper walking. In addition, the rubber studs at each foot assembly transfer the exoskeleton's weight to the ground during stance and single-support phases, relieving the human of the exoskeleton's weight during these phases. Furthermore, the load cells allow proper measurement of the ground reaction forces at each foot of the exoskeleton to estimate the center of pressure. Estimating the center of pressure is pivotal in applying advanced stability-controlled techniques to the exoskeleton [25, 26] (Fig. 7).
The exoskeleton's power and load torque requirements for active hip and knee joints must be correctly estimated to ensure proper selection of the joint actuators. This, in turn, requires proper estimation of the assistive load torque and estimation of dynamic load torque for the exoskeleton itself. The activities considered to be assisted by the designed lower body exoskeleton are walking, sit to stand and standing. Since walking is an essential activity entailing the most considerable power and torque requirements [27], it is selected to estimate the actuator requirements for the three different gait speeds of the elderly.
The exoskeleton needs to assist the human user; therefore, the torque required (for human walking) must be first estimated accurately to correctly find the exoskeleton's assistance torque requirement. For this purpose, mean walking gait curves for human knee and hip joints are first found in the sagittal plane using walking gait data in [28] for eighteen elderly persons (age 62–79 years, mean height 161 cm, mean weight 66.33 kg and mean stride length of 1.0 m). The gait curves have then been found (using this data) over a complete gait cycle at three different walking speeds, i.e., for a slow speed of 1.68 \(km/hr\), a medium speed of 2.4 \( \text{km}/\text{h}\) and a fast speed of 3.1 \(km/hr\). The mean gait curves for the angular position, velocity and acceleration of human knee and hip joints are, respectively, shown in Fig. 8a–c, while the associated required human joint torques \({\tau }_{{h}_{\text{hip}}}\) and \({\tau }_{{h}_{\text{knee}}}\) are shown in Fig. 8d. It is to be noted that the gait curves in Fig. 8 have been found by considering joint flexion and joint extension as positive and negative quantities, respectively. Figure 8d shows that a considerably large human torque is required for a small increase in walking speed, hence indicating that the human torque requirement for the exoskeleton is strongly dependent on the walking speed of the elderly.
The designed exoskeleton's ankle joint has two PDOF, as shown in Fig. 6; hence, no load torques are expected to be produced on the exoskeleton's active hip and knee joint due to ground reaction forces. The human user himself is supposed to provide the corresponding torque for the ground reaction forces. Therefore, there is no considerable coupling between the dynamics of the exoskeleton's two legs, and hence, the dynamic load torques for the active hip and knee joints could be found by considering each leg's dynamics individually. Since both the legs are symmetrical, a kinematic and dynamic model for a single leg with 2-ADOF is found using the Denavit–Hartenberg (D–H) parametric approach. The frame definitions for the exoskeleton's single leg are shown in Fig. 7, while the corresponding D–H parameters are listed in Table 3. If \({{q}_{m}}_{1}\),\({{q}_{m}}_{2}\) are the measured joint angles of hip and knee joints, respectively, then the joint angle vector \({\varvec{q}}\in {\mathbb{R}}^{2}\) is defined as
$${\varvec{q}} = \left[ {q_{{\left( {hip} \right)}}\ q_{{\left( {knee} \right)}} } \right]^{{\varvec{T}}} ,$$ (1)where \({q}_{(hip)}={{q}_{m}}_{1}+{{q}_{o}}_{1}\) and \({q}_{(knee)}={{q}_{m}}_{2}+{{q}_{0}}_{2}\). Here \({{q}_{0}}_{1}\),\({{q}_{0}}_{2}\) are the associated joint angle offsets, as shown in Table 3. The dynamic torque \({{\varvec{\tau}}}_{exo}\) \(\in\) \({\mathbb{R}}^{2}\) required for the 2-ADOF exoskeleton leg is therefore simply given by the forward dynamic equation as
$${{\varvec{\tau}}}_{\text{exo}}=\left[\begin{array}{c}{{\tau }_{\text{hip}}}_{exo}\\ {{\tau }_{\text{knee}}}_{exo}\end{array}\right]={{\varvec{M}}}_{\text{exo}}\left({\varvec{q}}\right)\ddot{{\varvec{q}}}+{{\varvec{C}}}_{\text{exo}}\left({\varvec{q}},\dot{{\varvec{q}}}\right)\dot{{\varvec{q}}}+{{\varvec{g}}}_{\text{exo}}\left({\varvec{q}}\right)\boldsymbol{ }.$$ (2)where \({\varvec{q}}\) is the joint angle vector defined in (1), \({{\varvec{M}}}_{\mathbf{e}\mathbf{x}\mathbf{o}\boldsymbol{ }}\left({\varvec{q}}\right),\) \({{\varvec{C}}}_{\mathbf{e}\mathbf{x}\mathbf{o}}\left({\varvec{q}},\dot{{\varvec{q}}}\right)\) \(\in\) \({\mathbb{R}}^{2\times 2}\) are the respective inertial Mass and Coriolis matrices for the exoskeleton leg while \({{\varvec{g}}}_{\mathbf{e}\mathbf{x}\mathbf{o}}\left({\varvec{q}}\right)\) \(\in\) \({\mathbb{R}}^{2}\) is the gravity vector found using the respective transformations and the inertial parameters [29].
To effectively assist the human in its daily activities, it is imperative that the exoskeleton closely follows the angular joint positions, velocities and accelerations of the corresponding human joints. Therefore, the dynamic load torque \({{\varvec{\tau}}}_{exo}\) is found for three different gait speeds using (2) with the angular position \({\varvec{q}}\), angular velocity \(\dot{{\varvec{q}}}\) and angular acceleration \(\ddot{{\varvec{q}}}\) given by the corresponding mean gait curves in Fig. 8a–c. The found dynamic load torques (\({{\tau }_{hip}}_{exo}\), \({{\tau }_{knee}}_{exo})\) for the hip and knee joint are then, respectively, shown in Fig. 9a and it is seen that both these torque are also increasing functions of human gait speed.
The lower body exoskeleton is required to provide maximum assistance of 50% to the human user. The required assistance torque \({{\varvec{\tau}}}_{assist}\boldsymbol{ }\in {\mathbb{R}}^{2}\) is, therefore, given in terms of the required human torque as
$${{\varvec{\tau}}}_{\text{assist}}=({a}_{\text{assi}st})\left[\begin{array}{c}{{\tau }_{\text{hip}}}_{assist}\\ {{\tau }_{\text{knee}}}_{\text{assist}}\end{array}\right]=\boldsymbol{ }{{(a}_{\text{assist}}){\varvec{\tau}}}_{h}.$$ (3)Here, \({a}_{\text{assist}}=\) 0.50 for maximum assistance and \({{\varvec{\tau}}}_{h}=\left[\begin{array}{c}{\tau }_{{\text{h}}_{\text{hip}} }\\ {\tau }_{{\text{h}}_{\text{knee}}}\end{array}\right]\) is the estimated human torque, as shown in Fig. 8d.
The assistance torque \({{\varvec{\tau}}}_{assist}\) found using (3) is shown in Fig. 9b for three different gait speeds. The net torque requirement for active joints of the exoskeleton can therefore be given from (1), (2) and (3) as
$${{\varvec{\tau}}}_{\text{net}}=\left[\begin{array}{c}{{\tau }_{\text{hip}}}_{\text{net}}\\ {{\tau }_{\text{knee}}}_{\text{net}}\end{array}\right]=\boldsymbol{ }{{\varvec{\tau}}}_{\text{exo}}+\boldsymbol{ }{{\varvec{\tau}}}_{\text{assist}}.$$ (4)The estimated net torques for hip and knee joint actuators using (4) are shown in Fig. 9c. The output power requirement \({{\varvec{p}}}_{\text{req}}\boldsymbol{ }\in\) \({\mathbb{R}}^{2}\) for the actuators can therefore be computed from (4) as.
$${{\varvec{p}}}_{\text{req}}=\left[\begin{array}{c}{p}_{\text{hip}}\\ {p}_{\text{knee}}\end{array}\right]={{\varvec{Q}}}_{{\varvec{\tau}}}\dot{{\varvec{q}}}.$$ (5)Here, \({{\varvec{Q}}}_{{\varvec{\tau}}}\) =\(\text{diag}\left\{{{\varvec{\tau}}}_{\text{net}} \right\} \in\) \({\mathbb{R}}^{2x2}\) is the net torque matrix and \(\dot{{\varvec{q}}}=\left[\begin{array}{c}{\dot{q}}_{\text{hip}}\\ {\dot{q}}_{\text{knee}}\end{array}\right]\) is the desired joint velocity vector for respective hip and knee joints and given by the human gait velocity curves in Fig. 8b.
The estimated output power required for hip and knee joint actuators for three gait speeds is shown in Fig. 9d. To keep the size and weight of the exoskeleton small but still be able to assist the human effectively (\({{\varvec{\tau}}}_{\text{assist}}\) =50% of\({{\varvec{\tau}}}_{h}\)), the actuators for hip and knee joints are suggested for the peak motoring power requirements corresponding to the medium gait speed (\( 2.4\text{ km}/\text{h})\) in Fig. 9d. Figure 9d also shows that the selected actuators need to perform in both the motoring and the generation modes to meet the said requirements. In motoring mode, the power is to be consumed from the battery and considered positive, while for generation mode, the power is to be delivered back to the battery and considered negative. The peak motoring power requirement for medium gait speed is found to be about 100 W, as shown in Fig. 9d. Therefore, 48 V, 100 W brushless motors (EC-60 from Maxon™) with respective harmonic drive gearheads of 80:1 and 50:1 were selected for the exoskeleton's active hip and knee joints to meet the said requirements. The specifications of the selected actuators are shown in Table 4.
To verify the performance of the suggested actuators, the output torque speed requirements (in Figs. 8b and 9c) for both the joints, at three different speeds are mapped on the characteristic curves of the respective actuators and shown in Fig. 10a and b. The allowed operational area for the actuators' motoring mode is limited by the torque speed characteristic curve, while the actuators' continuous torque operation region is limited by its power rating curves, as shown in Fig. 10a and b. The input current and voltage requirements (\({{\varvec{i}}}_{act}\),\({{\varvec{v}}}_{act}\in\) \({\mathbb{R}}^{2}\)) for both the actuators can be found in terms of the actuator’s parameter (in Table 4) as
$$\begin{array}{*{20}{l}}{{\varvec{i}_{{\text{act}}}} = }&{\left[ {\begin{array}{*{20}{l}}{{i_{{\text{act}}_{hip}}}}\\{{i_{{\text{act}}_{knee}}}}\end{array}} \right] = {\varvec{\eta} ^{ - 1}}\varvec{K}_{{\tau}} ^{ - 1}{\varvec{\tau} _{{\text{net}}}} + {{\varvec{i}}_{nl}},}\\{{{\varvec{v}}_{{\text{act}}}} = }&{\left[ {\begin{array}{*{20}{c}}{{{{v}}_{{\text{act}}}}_{{{hip}}}}\\{{{v}_{{\text{act}}}}_{{{knee}}}}\end{array}} \right] = {\varvec{K}}_{\text{s}}^{ - 1}\mathop {\dot{{\varvec{q}}}}_{nl}\, +\, {{\varvec{R}}_{\text{w}}}{{\varvec{i}}_{{{nl}}}}.}\end{array}$$ (6)where
\({{\varvec{i}}}_{{nl}}={{{\varvec{K}}}_{{\varvec{\tau}}}}^{-1}{{\varvec{\tau}}}_{f},\) \({\dot{{\varvec{q}}}}_{nl}=\dot{{\varvec{q}}}+{{\varvec{K}}}_{s},\)
$${\varvec{\eta}}=\text{diag}\{{\eta }_{\text{hip}},{\eta }_{\text{knee}}\},$$$${{\varvec{K}}}_{\tau }=diag\{{k}_{{\tau }_{hip}},{{k}_{\tau }}_{knee}\},$$$${{\varvec{K}}}_{s}=\text{diag}\left\{{k}_{{\text{s}}_{\text{hip}}},{{k}_{\text{s}}}_{\text{knee}}\right\}, {{\varvec{R}}}_{w}=\text{diag}\left\{{r}_{{\text{w}}_{\text{hip}}},{{r}_{\text{w}}}_{\text{knee}}\right\}.$$Vectors \({{\varvec{\tau}}}_{net}\) and \(\dot{{\varvec{q}}}\) in (6) are the torque and speed requirements shown in Figs. 8b and 9c, respectively, while matrices \({\varvec{\eta}},\)\({{\varvec{K}}}_{\tau }\), \({{\varvec{K}}}_{s}\), \({{\varvec{R}}}_{w}\) \(\in\) \({\mathbb{R}}^{2x2}\), respectively, represent the gear ratio, torque constant, speed constant and phase-to-phase winding resistance matrices of the actuators. Vectors\({{\varvec{i}}}_{nl}\), \({\dot{{\varvec{q}}}}_{nl}\) in (6), respectively, represent the no-load current and no-load speed of the actuators. The found values of \({{\varvec{i}}}_{\text{act}}\) and \({{\varvec{v}}}_{\text{act}}\) and plotted for three different gait speeds and, respectively, shown in Fig. 10c and d. The continuous and discontinuous current operation areas, along with operational limits for both actuators, are also clearly shown in Fig. 10c and d. Figure 10 shows that the actuators for the hip and knee joints of the exoskeleton need to operate in all four quadrants (2-motoring and 2-generation regions) to satisfy their respective torque-speed requirements. Since the maximum peak input power required for both the actuators is found to be 356W with a maximum peak actuator current of 10A (Table 5), a four-quadrant compatible 400W motor driver (EPOS4-Compact-50/8-CAN from Maxon™) was selected for each actuator. Since the power is to be drawn and returned to the battery during the motoring and generation operation of the actuators, the battery selected must also properly source and sink the maximum peak actuator current of 10.0 A to avoid large voltage spikes at the input, especially during generation operation.
Figure 10a and b shows that the speed and torque requirements for both the knee and hip joints at low and medium gait speeds remain within the continuous torque area of the respective actuators. Furthermore, the corresponding input current and voltage requirements for these speeds also remain in the continuous-current operation area of the respective actuator, as shown in Fig. 10c and d. Therefore, at low and medium-gait speeds, the selected actuators can safely deliver the respective required torques and speeds for the exoskeleton's active joints without exceeding their respective power ratings. For fast gait speed, though the torque speed requirements for hip and knee joint map onto the discontinuous area of operation of the respective actuators for a short duration per gait cycle, but these requirements always remain within the allowed operational area of the respective actuator as seen in Fig. 10a and b. Hence, the respective joint actuators also achieve fast-gait speed requirements at the cost of higher power dissipation, as seen by the corresponding current and voltage requirements in Fig. 10c and d.
The selected actuators' performance is evaluated based on the actuators' respective safety factors found in terms of the actuators' parameters listed in Table 5. These factors are obtained for peak input power \({p}_{{\text{joint}}_{\text{pk}}}\), peak net torque \({\tau }_{{\text{joint}}_{{\text{net}}_{\text{pk}}}}\) and peak speed \({\dot{q}}_{{\text{joint}}_{\text{pk}}}\) requirements for three different gait speeds and are defined as follows.
Peak torque safety factor \(=\frac{{\tau }_{{\text{act}}_{\text{max}}}}{{\tau }_{{\text{joint}}_{{\text{net}}_{\text{pk}}}}} ,\)
Peak power safety factor \(=\frac{{p}_{\text{act}}}{{p}_{{\text{joint}}_{\text{pk}}}}\),
Achievable speed safety factor \(=\frac{{\dot{q}}_{{\text{act}}_{\text{max}}}}{{\dot{q}}_{{\text{joint}}_{\text{pk}}}}\).
Here, \({\tau }_{{\text{act}}_{\text{max}}}\), \({p}_{\text{act}}\) and \({\dot{q}}_{{\text{act}}_{\text{max}}}\), respectively, represent the maximum actuator torque, power and speed. From these factors, it is seen that the selected actuators for hip and knee joints are expected to give high performance for slow-gait speed with high power, torque and speed safety factors together with respective average motoring efficiencies of 71 and 68%. The RMS input power requirements per gait cycle for hip and knee actuators at slow-gait speed are also reasonably low compared to actuators' power ratings and are found to be only 11 and 10 W, respectively. The safety factors obtained for medium gait speed are also noticeably good, with respective average motoring efficiencies of 67 and 64%. Though the power factor for the hip actuator is seen to be 0.89 for the medium-gait speed, the hip actuator's performance is not affected as the actuator is expected to be slightly overloaded only for a short time per gait cycle with a satisfactory speed safety factor of 1.8, under peak load conditions. The RMS input power requirements per gait cycle for both actuators at medium gait speed are also considerably less than the actuators' power ratings and are found to be 35 and 23 W, respectively. For fast-gait speed, it is seen that though the peak power safety factor is relatively low with respective values of 0.28 and 0.73 for hip and knee actuators, the torque and speed safety factors are reasonably satisfactory with respective average motoring efficiencies of 61 and 62%. The RMS input power requirements per gait cycle for both the joints at fast-gait speed are still less than the actuators' power ratings and are found to be 95 and 52 W, respectively. Therefore, the selected actuators can still work at a fast-gait speed but with larger overloading and more considerable power dissipation per gait cycle.
To properly implement the advanced control techniques [9, 10, 18] as suggested in Sect. 1 for the proposed design of lower body exoskeleton, a distributed master–slave control structure (motivated by some initial work done in [19]) is suggested, as shown in Fig. 12. The top-level control is to be implemented by the master controller, while the local joint-level control is intended to be implemented by the respective joint-level slave controller for each leg's active joint. The operational flow of impedance-based assistive force control employing DOB-based-DLTC is, respectively, shown in Figs. 13 and 15 for master and slave controllers of the exoskeleton during walking assistance mode.
To ensure robust and fast communication between the master and slave controllers, each leg's master and respective slave controllers are linked through two dedicated system CAN buses, i.e., system CAN bus 1 and system CAN bus 2, as shown in Fig. 12. A CAN message-based communication protocol has been developed based on the work done in [19] to ensure proper communication between the slaves and the master controller. The sender and receiver IDs are embedded into the message type for proper reception and decoding. The CAN message is broadcasted on the CAN bus by the sender, only to be received by the intended recipient. Three different types of CAN messages have been defined for the system CAN busses. All slave controllers sent the current joint positions' velocities and accelerations at fixed intervals to the master as CAN data messages. Master and the controllers send CAN error messages to transfer error information, while a CAN time stamp message is sent by the master controller to all the slaves every second to ensure proper synchronization of all the controllers in time, which is indicated visually by toggling an LED on the salve and master controllers.
The operational flow of CAN messages for master and slave controllers is shown in Figs. 13 and 15. The CAN bus is to be continuously polled by the controller (master or slave) to check for CAN message reception. Upon receiving a CAN message, the message is decoded for type (data or error message) and sender ID. The data are then to be extracted and stored in the respective data buffer for a CAN data message reception depending on the received ID. For master CAN error message reception, the message is evaluated for criticality, and if found critical, a critical error message is sent to all the slaves, and the user is informed accordingly. If the error message is not critical, a corrective error message is to be sent to the respective slave controller with corrective action embedded in the error message. If the received error message is critical for slave controllers, an emergency stop is to be activated for the local joint; otherwise, the error message is decoded to perform the suggested corrective action in the error message.
The master controller is suggested to be interfaced with the user HMI module via a Bluetooth connection, as shown in Fig. 12.
Serial Bluetooth radio modules (SPBTC2A) are recommended for this purpose. The HMI module is suggested to provide the user with the exoskeleton's critical control functionality. It has to configure the exoskeleton's functionality by sending configuration data to the master controller, i.e., mode and level of assistance, sample time, desired impedance/controller parameters and kinematic/dynamic parameters. The HMI module is also responsible for informing the user of the current state and errors of the exoskeleton system.
The master controller is suggested to implement the top-level system control for the exoskeleton in real time. The master controller performs proper initialization of all the system CAN busses, ADCs and timers after receiving configuration data from the user. It also has to scale the already stored mean-normalized desired assistive load torque curve with respect to the user's height and weight.
Master controller is to then register and acknowledge all the slave joint controllers on the two system CAN busses as registered CAN nodes. At each interrupt, the task space position velocity and acceleration of end-effectors compliant supports are first to be calculated from the respective supports kinematics for the currently received joint data (position velocity and acceleration of active joints). The direction and magnitude of desired assistive forces in the respective support reference frames are then to be estimated from the scaled desired assistive load torque curve using the information of the current gait phase, which, in turn, is to be continuously estimated by using the toe and heel force sensors and IMU data for each foot assembly, as discussed in Sect. 4.
The current interactive forces between the human limb and the exoskeleton are suggested to be measured at each compliant support to compute the force error (w.r.t the estimated desired assistive force) in the respective support’s frame of reference. This force error must then be transformed to the inertial frame using the respective rotation matrices of the supports to properly implement the force-based impedance control in task space for each compliant support. Furthermore, the current task space position, velocity and acceleration of the human limb at the contact supports must be fed to the impedance control law as desired values. These desired values could be estimated using human limb forward kinematics for the mean joint position, velocity and acceleration curves as shown in Fig. 8a–c, which in turn require continuous estimation of the current gait phase [30,31,32]. Alternatively, these desired values could also be estimated by estimating the deformation of compliant elements using the measured interactive forces and then subtracting it from the respective estimates of current task space positions for contact supports, as discussed in [10]. The desired task space accelerations for each compliant support are then computed using the desired task space impedance control law (having its desired parameters set by the user) [10]. The desired velocities are then in turn computed from the desired accelerations and sent to respective slave joint controllers to be tracked in real time over the system CAN busses as CAN data messages.
The detail of the suggested joint-level control structure is shown in Fig. 14. It comprises of a slave controller, a brushless motor controller, a joint assembly (as explained in Sect. 2.2), a sensorized 3-D compliant support (as discussed in Sect. 2.3), digital and absolute encoders, an inline transducer amplifier for compliant support and a torque sensor signal conditioning card. The slave controller is responsible for providing the local joint-level control to the joint assembly. The slave controller is interfaced through two CAN busses. The system CAN bus interfaces the respective slave controllers to the master controller while a local CAN bus interfaces a CANopen™-based motor driver ((EPOS4-Compact-50/8-CAN from Maxon™) to its respective slave controller, as shown in Fig. 14. This suggested interface allows the slave controller to configure, control and monitor the motor driver over the local CAN bus and allows it to receive and send joint position and velocity information from the motor driver. The high-resolution digital encoder, coupled to the motor's back-end, is proposed to ensure an accurate measurement of joint position and velocity even for short sampling times. The absolute encoder, coupled directly to the joint shaft, is proposed to ensure a correct joint position estimation, especially at startup. In addition, a CAN protocol compatible with CANopen™ protocol is developed for the slave controller to ensure robust communication with the motor driver over the local CAN bus. The slave controller's operational flow is shown in Fig. 15 for the walking assistance mode using DOB-based-DLTC impedance control. After initializing its ADCs, timers and CAN busses, the slave controller has to send its slave ID to the master controller to be registered as CAN node. The slave, after successful acknowledgment, then waits for the reception of configuration data (i.e., control mode, sample time, local controller parameters desired and starting joint position) from the master controller. Based on the data received, the local motor driver is to be initialized and checked for successful initialization in the desired mode (current control mode in our case) several times over the local CAN bus before sending a critical error message to the master controller. Upon success, the sample timer with the desired sample time is to be initialized to allow local joint-level control to be implemented in real time.
At each timer interrupt, the local CAN bus and the CAN data buffers are to be read (by the slave controller) to get the respective current joint velocity (sent by the local motor driver) and the latest desired velocity (sent by the master controller) for implementing the local servo velocity control. The current load torque on the local joint actuator shaft is then measured from the interfaced load torque sensor to find the desired reference values for the motor driver, using the DOB-based-DLTC technique as proposed in [9, 18]. The computed desired reference values are then to be sent to the local motor driver over the local CAN bus for implementing motor control. The motor driver is continuously monitored for error messages and communicated to the master controller for necessary corrective measures.
To provide the necessary gait phase and ground reaction information to the master controller, load cells at the toe and heel of each foot assembly (as shown in Fig. 6) are interfaced to the master controller through dedicated inline (UV-Series-24 V), Honeywell™ transducer amplifiers. These amplifiers feature a highly regulated excitation supply (for the sensor bridge), a high S/N ratio, programmable gain and a wide-range zero-error calibration. The ability to sense the ground reaction forces allows the master controller to not only correctly estimate the continuous phase of the gait cycle [30,31,32] but also allows it to make critical decisions (for the respective slave controllers) to correctly implement the high-level strategies to ensure exoskeleton's stability and control [25, 26]. In addition, force load sensors at each compliant support are suggested to be interfaced with the respective slave controller through dedicated UV-Series inline transducer amplifiers.
To practically implement the proposed CAN bus-based master–slave control architecture for the designed lower body exoskeleton, a dedicated embedded control card has also been developed, as shown in Fig. 16. The developed control card can serve as both the slave and the master controllers. The card supports two CAN buses, which could then, in turn, be used as system or local CAN buses. The cards use a 32-bit RISC floating-point microcontroller (AT32UC3C) at 66-MHz with 512-KB as internal flash memory and 8-MB as additional external flash memory (AT45DB081). The card also incorporates a nine-axis (gyro + accelerometer + magnetometer) inertial measurement unit (MPU-) to facilitate inertial measurement and a Bluetooth radio module (SPBTC2A) to ensure radio communication between the master controller and the user HMI module. The card that also supports 6-ADC channels (for sensor interfacing), 2-DAC outputs, 2-PWM ports and 2 digital encoder input interfaces (to facilitate servo motor control applications) is also shown in Fig. 16.
The design of a control box is suggested to properly house and mount all the exoskeleton's associated electronics and to provide proper interfaces to all the sensors and joint actuators. The proposed implementation of the master–slave control architecture for the lower body exoskeleton requires an embedded control card to serve as the master controller, while another four embedded control cards are required to serve as slave controllers. Master and slave controllers are linked locally within the control box through the two system CAN busses on the respective control cards, while each slave is linked to its respective motor driver over the local CAN bus (as discussed in detail in Sect. 5.3 and Sect. 5.4) using four 2-pin shielded connectors. All the control cards are facilitated to be appropriately mounted using stackable mounting assemblies, as shown in Fig. 17. The four inline transducer amplifiers for the compliant supports and four transducer amplifiers for the foot sensors (as discussed in Sect. 5.5) are suggested to be mounted using double-stack mounting assemblies and interfaced to the respective master and slave control cards, as shown in Figs. 12 and 14, respectively. These amplifiers are suggested to be interfaced to their respective load sensors through eight 4-pin shielded connectors. To provide JTAG programming interfaces to master and slave control cards, five DB9 programming ports are also suggested in the control box.
The control box is also designed to house a 48 V 10Ah, 13S4P battery, with 25R lithium cells. The maximum average current requirement per gait cycle for each joint is estimated to be 1.4A. Hence, the four joint actuators' net current requirement is about 5.6 A; therefore, the selected battery is expected to power the exoskeleton for at least 1.5 h. The 48 V, 10A power supply to the four joint actuators is to be provided through respective 2-pin power connectors. A charging interface is also suggested for the battery through a 15A, 2-pin power connector. The power turn-on/off control for the exoskeleton is to be provided through a 10A indicator power switch, as shown in Fig. 17. A regulated 24 V supply to all the embedded control cards and inline amplifiers in the control box is ensured through a dedicated 48 V to 24 V, 5A, DC–DC convertor.
Housing the control cards, inline transducer amplifiers, DC–DC converter and battery in one place not only localizes the system CAN busses between master and slave controllers but also localizes the respective wires for power supply and interfacing, resulting in shorter wires and therefore reduced electromagnetic interference.
One possible way to safely provide the desired level of assistance by the exoskeleton is to control the shank contact support using the impedance control law while controlling the thigh contact support only in the null space of the shank contact support, as suggested in [10]. This approach is not possible in the suggested LADOF-based design of the exoskeleton, as there is no null space available in the Jacobian of the shank support in which to control the thigh support to provide the desired level of assistance (since the active degree of freedom at the shank support is only two which is less than the desired 3-ADOF). Therefore, a new control strategy is proposed for the designed lower body exoskeleton, capable of imparting simultaneous impedance-based force tracking control for both the thigh and shank contact supports, while using both the proposed passive compliant contact supports and DOB-based-DLTC for improved pHRI performance of the exoskeleton. The proposed control strategy while considering the operational flow of master and slave controllers (shown in Figs. 13 and 15) is shown in detail in Fig. 18. The simultaneous assistive force tracking control for both the thigh and shank supports is made possible by proposing separate force-based impedance control laws in task space for each contact support, whereas the joint reference acceleration for each joint, i.e., \({\ddot{q}}_{{\text{r}}_{(hip)}}, {\ddot{q}}_{{\text{r}}_{(knee)}}\), is individually generated by the respective impedance control law, as shown in Fig. 18. The complete control law for the proposed strategy is given as
$$\begin{aligned}{\ddot{\varvec{x}}}_{ra} = & {\ddot{\varvec{x}}}_{{h_{\left( {{\rm{thigh}}} \right)}}}^{\ast} + {\left( {{{\varvec{M}}_{{d_{\left( {{\rm{thigh}}} \right)}}}}} \right)^{ - 1}}\left( { - {{\varvec{K}}_{{f_{\left( {{\rm{thigh}}} \right)}}}}{{\varvec{e}}_{{f_{\left( {{\rm{thigh}}} \right)}}}} - {{\varvec{B}}_{{d_{\left( {{\rm{thigh}}} \right)}}}}{{\dot {\varvec{e}}}_{\left( {{\rm{thigh}}} \right)}} - {{\varvec{K}}_{{d_{\left( {{\rm{thigh}}} \right)}}}}{{\varvec{e}}_{\left( {{\rm{thigh}}} \right)}}} \right),\\ {\ddot{\varvec{x}}}_{rb} = & {\ddot{\varvec{x}}}_{{h_{\left( {{\rm{shank}}} \right)}}}^{\ast} + {\left( {{{\varvec{M}}_{{d_{\left( {{\rm{shank}}} \right)}}}}} \right)^{ - 1}}\left( { - {{\varvec{K}}_{{f_{\left( {{\rm{shank}}} \right)}}}}{{\varvec{e}}_{{f_{\left( {{\rm{shank}}} \right)}}}} - {{\varvec{B}}_{{d_{\left( {{\rm{shank}}} \right)}}}}{{\dot {\varvec{e}}}_{\left( {{\rm{shank}}} \right)}} - {{\varvec{K}}_{{d_{\left( {{\rm{shank}}} \right)}}}}{{\varvec{e}}_{\left( {{\rm{shank}}} \right)}}} \right),\\ {\ddot{\varvec{q}}}_{ra} = & \left[ {\begin{array}{*{20}{c}}{\ddot{\varvec{q}}}_{r{a_1}}\\ {\ddot{\varvec{q}}}_{r{a_2}}\end{array}} \right] = {\left( {~{{\varvec{J}}_v}{{\left( {\varvec{q}} \right)}_{\left( {{\rm{thigh}}} \right)}}} \right)^{\dagger}} \left( {{{\ddot{\varvec{x}}}_{ra}} - {{\dot {\varvec{J}}}_v}{{\left( {\varvec{q}} \right)}_{\left( {{\rm{thigh}}} \right)}}\dot {\varvec{q}}} \right),\\ {\ddot{\varvec{q}}}_{rb} = & \left[ {\begin{array}{*{20}{c}} {\ddot{\varvec{q}}}_{r{b_1}}\\ {{{\ddot{\varvec{q}}}_{r{b_2}}}}\end{array}} \right] = {\left( {~{{\varvec{J}}_v}{{\left( {\varvec{q}} \right)}_{\left( {{\rm{shank}}} \right)}}} \right)^{\dagger}} \left( {{{\ddot{\varvec{x}}}_{rb}} - {{\dot {\varvec{J}}}_v}{{\left( {\varvec{q}} \right)}_{\left( {{\rm{shank}}} \right)}}\dot {\varvec{q}}} \right),\\ {\ddot{\varvec{q}}}_r = & \left[ {\begin{array}{*{20}{c}} {{{\ddot{\varvec{q}}}_{{r_{\left( {{\rm{hip}}} \right)}}}}}\\ {{{\ddot{\varvec{q}}}_{{r_{\left( {{\rm{knee}}} \right)}}}}}\end{array}} \right] = \left[ {\begin{array}{*{20}{c}} {{{\ddot{\varvec{q}}}_{r{a_1}}}}\\{{{\ddot{\varvec{q}}}_{r{b_2}}}}\end{array}} \right].\end{aligned}$$ (9)Here,
$${{\varvec{e}}}_{\left(thigh\right)}={{\varvec{x}}}_{{h}_{\left(thigh\right)}}^{*}-{{\varvec{x}}}_{{exo}_{\left(thigh\right)}},$$$${{\varvec{e}}}_{\left(shank\right)}={{\varvec{x}}}_{{h}_{\left(shank\right)}}^{*}-{{\varvec{x}}}_{{exo}_{\left(shank\right)}},$$$${{\varvec{e}}}_{{{\varvec{f}}}_{\left(thigh\right)}}={{\varvec{f}}}_{{assist}_{\left(thigh\right)}}^{*}-{{\varvec{f}}}_{{assist}_{\left(thigh\right)}},$$$${{\varvec{e}}}_{{{\varvec{f}}}_{\left(shank\right)}}={{\varvec{f}}}_{{assist}_{\left(shank\right)}}^{*}-{{\varvec{f}}}_{{assist}_{\left(shank\right)}},$$$${{\varvec{f}}}_{{assist}_{\left(thigh\right)}}^{*}={{\varvec{R}}\left({\varvec{q}}\right)}_{(thigh)}{{\varvec{f}}}_{{e}_{{assist}_{\left(thigh\right)}}}^{*},$$$${{\varvec{f}}}_{{assist}_{\left(shank\right)}}^{*}={{\varvec{R}}\left({\varvec{q}}\right)}_{(shank)}{{\varvec{f}}}_{{e}_{{assist}_{\left(shank\right)}}}^{*},$$$${\varvec{q}}={\left[{q}_{(hip)}{q}_{(knee)}\right]}^{{\varvec{T}}},$$$${{\varvec{q}}}_{h}^{\boldsymbol{*}}={\left[{q}_{{h}_{(hip)}}^{*} {q}_{{h}_{(knee)}}^{*}\right]}^{{\varvec{T}}}$$\({{\varvec{x}}}_{{h}_{\left(thigh\right)}}^{*}={col vec}_{3}\left\{{{\varvec{T}}\left({q}_{{h}_{\left(hip\right)}}^{*}\right)}_{\left(thigh\right)}\right\},\) \({{\varvec{K}}}_{{f}_{(thigh)}}=diag\left\{{k}_{{f}_{(thigh)j}}\right\}\),
\({{\varvec{x}}}_{{h}_{(shank)}}^{*}={col vec}_{3}\left\{{{\varvec{T}}\left({q}_{(hip)},\boldsymbol{ }{q}_{{h}_{(knee)}}^{*}\right)}_{(shank)}\right\},\) \({{\varvec{K}}}_{{f}_{(shank)}}=diag\left\{{k}_{{f}_{(shank)j}}\right\}\),
\({{\varvec{x}}}_{{exo}_{(thigh)}}={col vec}_{3}\left\{{{\varvec{T}}\left({\varvec{q}}\right)}_{(thigh)}\right\},\) \({{\varvec{x}}}_{{exo}_{(shank)}}={col vec}_{3}\left\{{{\varvec{T}}\left({\varvec{q}}\right)}_{(shank)}\right\}\),
\({{\varvec{M}}}_{{{\varvec{d}}}_{(thigh)}}={{\varvec{M}}}_{{{\varvec{d}}}_{(shank)}}=diag\left\{{m}_{{d}_{j}}\right\},\) \({{\varvec{B}}}_{{{\varvec{d}}}_{(thigh)}}={{\varvec{B}}}_{{{\varvec{d}}}_{(shank)}}=diag\left\{{b}_{{d}_{j}}\right\}\),
\({{\varvec{K}}}_{{{\varvec{d}}}_{(thigh)}}={{\varvec{K}}}_{{{\varvec{d}}}_{(shank)}}=diag\left\{{k}_{{d}_{j}}\right\},\) for j = 1 to 3.
Here \({q}_{{h}_{(hip)}}^{*}\), \({q}_{{h}_{(knee)}}^{*}\) represent the desired human joint positions values and are given by the respective hip and knee joint trajectory curves in Fig. 8a while \({\varvec{q}}={\left[{q}_{(\text{hip})}{q}_{(\text{knee})}\right]}^{{\varvec{T}}}\) represents the current joint position vector of the exoskeleton. The vectors \({{\varvec{x}}}_{{h}_{(\text{thigh})}}^{*}\), \({{\varvec{x}}}_{{h}_{(\text{shank})}}^{*}\) represent the desired human task space positions of the respective contact supports while \({{\varvec{x}}}_{{\text{exo}}_{(\text{thigh})}}\), \({{\varvec{x}}}_{{\text{exo}}_{(\text{shank})}}\) represent the actual task space position of the respective contact supports. The \({{\varvec{f}}}_{{e}_{{\text{assist}}_{(\text{thigh})}}}^{*}\), \({{\varvec{f}}}_{{e}_{{\text{assist}}_{(\text{shank})}}}^{*}\) represent the desired assistive forces on the human in the respective thigh and shank support frames. These desired forces have been found using the net torque requirement of the exoskeleton in Sect. 3.3 and are shown in Fig. 11. It is seen in Fig. 11 that \(x\) and \(z\) component of both the desired assistive forces, i.e., \({f}_{{e}_{{x}_{(\text{thigh})}}}^{*},{f}_{{e}_{{z}_{(\text{thigh})}}}^{*},{f}_{{e}_{{x}_{(\text{shank})}}}^{*}\) and \({f}_{{e}_{{z}_{(\text{shank})}}}^{*}\) are zero, and the assistance is only provided along the y-axis by the y-force components, i.e., \({f}_{{e}_{{y}_{(\text{thigh})}}}^{*}\) and \({f}_{{e}_{{y}_{(\text{shank})}}}^{*}\), respectively. The required gait phase needs to be linearly estimated from the measured cycle time of the previous gait cycle. The necessary gait cycle time is to be measured from the heel-strike event to the toe-off event for each leg. Both heel-strike and toe-off events are to be reliably detected using the respective foot sensor data of each leg. The actual assistive forces on human, i.e., \({{{\varvec{f}}}_{e}}_{{\text{assist}}_{(\text{thigh})}}\), \({{{\varvec{f}}}_{e}}_{{\text{assist}}_{(\text{shank})}},\) at the respective contact supports are to be measured in the support end-effector frames using the respective force sensors as discussed in detail in Sect. 2.3. To ensure proper control, it is proposed that \({{\varvec{x}}}_{{h}_{(\text{thigh})}}^{*}\) is estimated using \({q}_{{h}_{(hip)}}^{*}\) while \({{\varvec{x}}}_{{h}_{(\text{shank})}}^{*}\) is estimated using the actual current position of the hip joint \({q}_{(hip)}\) (instead of \({q}_{{h}_{(\text{hip})}}^{*}\)) and the desired human knee joint position \({q}_{{h}_{(\text{knee})}}^{*}\). Gains \({{\varvec{K}}}_{{f}_{(\text{thigh})}}\) and \({{\varvec{K}}}_{{f}_{(\text{shank})}}\) in (9), respectively, represent the force error gain matrices for thigh and shank supports. Since both the desired and actual assistive forces are considered as the forces on human, the terms (\({{\varvec{K}}}_{{f}_{(\text{thigh})}}{{\varvec{e}}}_{{f}_{(\text{thigh})}})\) and (\({{\varvec{K}}}_{{f}_{(\text{shank})}}{{\varvec{e}}}_{{f}_{(\text{shank})}})\) have been considered negative in the impedance control laws in (9). The reference velocity for the joint space velocity control is found from \({\ddot{{\varvec{q}}}}_{r}\) as \({\dot{{\varvec{q}}}}_{r}=\left[\begin{array}{c}{\dot{q}}_{{r}_{(\text{hip})}}\\ {\dot{q}}_{{r}_{(\text{knee})}}\end{array}\right]=\int {\ddot{{\varvec{q}}}}_{r} dt\).
The control law for the DOB-based-DLTC for the 2-CCDC drives is given in terms of the reference current control input \({{\varvec{i}}}_{r}\) and follows from [10] as
$${{\varvec{i}}}_{{\varvec{r}}}=\left[\begin{array}{c}{i}_{{r}_{(\text{hip})}}\\ {i}_{{r}_{(\text{knee})}}\end{array}\right]{={{\varvec{N}}}_{{\varvec{D}}{\varvec{c}}}}_{f}\left(s\right)\boldsymbol{ }{{\varvec{\eta}}}^{-1}{{\varvec{\tau}}}_{{\varvec{L}}}+{{\varvec{H}}}_{{\varvec{c}}}{{\varvec{K}}}_{{{\varvec{t}}}_{n}}^{-1}{{\varvec{\tau}}}_{{\varvec{e}}}^{\boldsymbol{*}},$$ (10)where
$${{\varvec{\tau}}}_{{\varvec{e}}}^{\boldsymbol{*}}=\left[\begin{array}{c}{\tau }_{{e}_{(\text{hip})}}^{*}\\ {\tau }_{{e(}_{\text{knee})}}^{*}\end{array}\right]={\left({\varvec{I}}-{{\varvec{Q}}}_{{\varvec{o}}}(s)\right)}^{-1}\left({{\varvec{\tau}}}_{{\varvec{r}}}-{{\varvec{Q}}}_{{\varvec{o}}}(s){{\varvec{G}}}_{{{\varvec{T}}}_{n}}^{-1}(s){\varvec{\eta}}\dot{{\varvec{q}}}\right),$$$${{\varvec{\tau}}}_{{\varvec{r}}}=\left[\begin{array}{c}{\tau }_{{r}_{(\text{hip})}}\\ {\tau }_{{r}_{(\text{knee})}}\end{array}\right]={\varvec{C}}\left(s\right){\varvec{\eta}}\left({\dot{{\varvec{q}}}}_{r}-\dot{{\varvec{q}}}\right).$$Here \({\varvec{\eta}}= \text{diag}\{{\eta }_{j}\}\) is the gear ratio matrix, \({{{\varvec{N}}}_{{\varvec{D}}{\varvec{c}}}}_{f}\left(s\right)={{\varvec{N}}}_{{\varvec{D}}{\varvec{c}}}\left(s\right){{\varvec{Q}}}_{{\varvec{D}}{\varvec{c}}}\left(s\right)=\text{diag}\left\{{{n}_{DC}\left(s\right)}_{j}{{q}_{DC}\left(s\right)}_{j}\right\}\) is the realizable DLTC matrix, \({{\varvec{Q}}}_{{\varvec{D}}{\varvec{c}}}\left(s\right)=diag\left\{{{q}_{Dc}\left(s\right)}_{j}\right\}\) is the cascaded low-pass filter matrix for proper implementation of DLTC, and \({{\varvec{Q}}}_{{\varvec{o}}}\left(s\right)=diag\left\{{{q}_{o}\left(s\right)}_{j}\right\}\) is the cascaded low-pass filter matrix for proper implementation of DOB. Matrix \({{\varvec{H}}}_{{\varvec{c}}}=\text{diag}\{{{h}_{c}}_{j}\}\) is the current feedback gain matrix while \({{\varvec{K}}}_{{{\varvec{t}}}_{n}}\)=\(\text{diag}\{{{k}_{t}}_{j}\}\) is the constant torque matrix for the CCDC drives of the exoskeleton. Vector \({{\varvec{\tau}}}_{{\varvec{L}}}={[{\tau }_{{L}_{\left(\text{hip}\right)}} {\tau }_{{L}_{\left(\text{hip}\right)}}]}^{T}\) is the sensed load torque using the torque sensor as discussed in Sect. 2.2. Transfer function \({{{\varvec{G}}}_{{\varvec{T}}}}_{n}\left(s\right)=\text{diag}\left\{{{{ g}_{T}}_{n}\left(s\right)}_{j}\right\}\) is the nominal forward torque dynamics matrix of the brushless motors from \({{\varvec{\tau}}}_{e}^{*}\) to \(\dot{{\varvec{q}}}\) while \({\varvec{C}}\left({\varvec{s}}\right)=\text{diag}\left\{{c\left(s\right)}_{j}\right\}\) is joint space velocity controller matrix. Here j = 1 to 2. Detailed description of transfer functions \({{{\varvec{N}}}_{{\varvec{D}}{\varvec{c}}}}_{f}\left(s\right)\), \({{\varvec{Q}}}_{{\varvec{o}}}\left(s\right)\), \({{\varvec{Q}}}_{{\varvec{D}}{\varvec{c}}}\left(s\right)\) and \({{{\varvec{G}}}_{{\varvec{T}}}}_{n}\left(s\right)\) is given in detail in [10]. The complete control law for the proposed control strategy for the lower body exoskeleton employing the DOB-based-DLTC is therefore completely defined by Eqs. (9) and (10).
The comparative simulated results for assistive force tracking of the designed lower body exoskeleton are shown in Fig. 19a–c. The found mean desired assistive force curves (shown in Fig. 11) for each contact support (based on the 18 elderly people experimental gait data [28] as discussed in Sect. 4) were used as desired assistive force inputs for both the contact supports. All the practical saturation limits for the selected actuators and the power drivers were considered during the simulation.
Contact us to discuss your requirements of exoskeleton joint actuator. Our experienced sales team can help you identify the options that best suit your needs.
These results are presented for a single leg of the exoskeleton, for increasing values of the passive compliant element stiffness \({k}_{s}\) (of the contact supports). These comparative results have been found for three different gait speeds, i.e., medium, slow and fast speeds, while using the DOB-based-DLTC at the joint space level. The detailed kinematic and dynamic model of the exoskeleton leg (presented in Sect. 3.2) along with a detailed model of DOB-based-DLTC CCDC drive in [10] was used in tandem with the proposed control strategy (presented in Fig. 18) to obtain the results for required assistance at each contact support. The desired impedance parameters for the task space impedance control law in (9) were selected to be \({{m}_{d}}_{j}=0.1\), \({{b}_{d}}_{j}=147\) and \({{k}_{d}}_{j}=320\). The force error gains were selected to be \({{k}_{{f}_{(\text{thigh})}}}_{j}\)=1.2, \({{k}_{{f}_{(\text{shank})}}}_{j}\)=1.8. The designed bandwidths of the velocity controllers \({c\left(s\right)}_{j}\), DOB filters \({{q}_{DC}\left(s\right)}_{j}\) and DLTC filters \({{q}_{o}\left(s\right)}_{j}\) for both the joints are listed in Table 6.
The \({k_{{s_{th}}}}\) and \({k_{{s_{sh}}}},\) respectively, represent the stiffness of the compliant elements for the proposed thigh and shank supports of the exoskeleton. Increasing the stiffness of the support implies that the interface between the human and the exoskeleton is made stiffer. From a pure robotic perspective where it is desired to accurately track the desired trajectories in time, it is desired that the exoskeleton is stiff, as this ensures small steady-state errors. But high stiffness of the exoskeleton is not a desired feature when force feedback control of interactive forces is required (as discussed in detail in Sect. 1). It is well known that the stiffer the interface is between two dynamically interacting bodies, the more difficult it is to control and realize a desired low contact point impedance, primarily due to the limited bandwidth of actuators [11, 12]. Hence, controlling the interacting force between these bodies is further difficult to regulate, which in turn results in large overshoots and settling times for the force response.
Therefore, it is seen from the results in Fig. 19 that with a high value of support stiffness \({k_{{s_{th}}}}\) and \({k_{{s_{sh}}}},\) (i.e., with a stiffer interface between the human and the exoskeleton), the desired assistive force feedback control is less stable with large overshoots and hence large settling time. This, hence, results in a decreased pHRI performance of the exoskeleton, even though an impedance control strategy is used to control the contact point impedance. But, on the other hand, it also exhibits a low steady-state force tracking error, which is expected, as the interface between the human and exoskeleton is now stiffer.
For a lower value of support stiffness \({k_{{s_{th}}}}\), \({k_{{s_{sh}}}},\), the interface between the human and the exoskeleton is less stiff, and hence, it is seen from the results shown in Fig. 19 that the desired assistive force feedback control (for both the contact supports) is much more stable with lower overshoots and hence fast settling time (for all the gait speeds), therefore ensuring an improved and stable pHRI performance of the exoskeleton, as the desired low contact point impedance is better achieved and realized. On the other, it has a slightly higher steady-state force tracking error, which is also expected, as the interface between the human and exoskeleton is now less stiff. This has been the main motivation and hypothesis for proposing the compliant element supports in the first place, i.e., to further improve the pHRI performance of an impedance-controlled exoskeleton, as given in Sect. 1.
Therefore, it is seen that the proposed control strategy can successfully impart simultaneous impedance-based assistive force tracking control of the exoskeleton at both contact supports with the selected actuators (as discussed in Sect. 3). Furthermore, as compared to the classical ridged contact supports, using passive, compliant arm supports (as discussed in Sect. 2.3 with reduced passive stiffness) in combination with impedance-based force control not only results in improved and stable force tracking performance of the exoskeleton but also allows lower values of desired impedance to be effectively realized with stability.
To clearly see the improvement afforded by the DOB-based-DLTC, comparative velocity tracking error simulation results for both hip and knee joint actuators are explicitly shown in Fig. 19d. These results are presented for a slow-gait speed force tracking with task space impedance parameters set as \({m}_{{d}_{j}}=0.2, {b}_{{d}_{j}}=112, {k}_{{d}_{j}}=320\). Hybrid switching strategy as proposed in [18] has been used to prevent the actuators from going into saturation. It is seen that the average velocity tracking error for hip and knee joint actuators with the proposed compensator (DOB-based-DLTC) is reduced by −7.1 dB (max) and −1.16 dB (max), respectively. This implies that the load torques on the respective joint actuators are better compensated by the use of DOB-based-DLTC, which in turn results in better performance of the joint actuators with lower vibrations and better force tracking performance.
In the UK, 6.8 million people live with mobility-related disabilities; the leading causes of which are musculoskeletal conditions and stroke [1, 2]. Persons with stroke are living longer due to reductions in risk factors and improvements of treatments [2]. The population overall is also aging; the number of people living over 85 is expected to increase from 1.8 million in to 3 million by [3]. Musculoskeletal impairments are associated with older age, therefore both those with musculoskeletal impairments and stroke survivors are living longer with disabilities that require assistance [2, 4].
Impaired mobility can have widespread effects on an individual’s quality of life as participation challenges impact their work, social life and activities of daily living (ADLs) [5]. Mobility impairments are also a risk factor for falls which reduce an individual’s confidence and self-belief in their own mobility, and can lead to activity avoidance, social isolation and depression, which in turn increases frailty and the ‘fear of falling’ cycle [6,7,8]. Thus, the paramount goal for physiotherapy rehabilitation is to ensure the continued mobility of individuals, with evidence demonstrating that, for neurological patients, repetitive movements are crucial to re-learn motor functions [9]. This is not without its challenges; in the UK, persons with stroke typically receive only 35 minutes of inpatient physiotherapy per day, despite the guidance of 45 minutes minimum [10, 11]. Increasing rehabilitation time may not be achievable as traditional rehabilitation frequently requires body weight support of the patient, which can be physically demanding for the physiotherapist who may require assistance from others [12, 13]. Consequently, therapist fatigue and staffing capacity limits what a patient is able to achieve in a session [13]. Assistive devices such as walkers are commonly provided to patients with mobility impairments [14]. These devices fall under the umbrella of ’assistive technology’, which describes products or systems that assist individuals with disabilities, restricted mobility or other impairments to perform functions that might otherwise be impossible or challenging [15]. Although assistive devices can improve rehabilitation of muscle and neural processing, they have limitations that prevent individuals from carrying out their ADLs as normal [14]. Reported challenges include opening of doors or getting on to public transport when using four-wheeled walkers, and issues carrying items, food and drink when using a walking stick [14, 16, 17].
Development in wearable powered exoskeletons offers a potential solution to traditional rehabilitation challenges [18, 19]. An exoskeleton, also known as a wearable robot, is a mechanical system worn by humans to augment, complement or substitute the function of the wearer’s limbs [20]. Early developed exoskeletons were stationary devices used to train patients on a treadmill with body-weight support, reducing loads on lower limbs for rehabilitation, such as DGO [21], LOPES [22] and ALEX [23]. Later, commercially available, portable assistive exoskeletons were developed, including Ekso, Rewalk [24], Indego and Exo H2 with an increasing number in development [12, 25]. Although not identical, their principles and designs are similar, consisting of an external actuator(s) fitted in parallel with weak or paralysed lower limbs to assist with mobilising and activities of daily living [26].
Many of these existing rigid exoskeletons were initially developed to provide maximal assistance to those with complete paralysis resulting from spinal cord injury. Interest has increased in the exoskeletons that can provide sensory-guided motorised lower limb assistance for person’s with stroke [27, 28]. These devices provide partial assistance during mobility tasks, allowing persons with stroke to actively participate through practising postural control and locomotion patterns [12]. A systematic review with a meta-analysis demonstrated that rigid exoskeletons are safe, with no reported adverse events, with falls only reported in a study using an early prototype [26, 29]. Further, rigid exoskeletons have widespread benefits including increased walking time, number of steps and improved strength and postural control in stroke survivors [30]. Studies have only recently explored patients and physiotherapists’ perspectives of the use of exoskeletons [18, 25, 31]. A key advantage of existing rigid exoskeletons was their ability to reduce the physical strain on therapists, therefore fewer members of staff would be needed to assist a patient, increasing the service’s capacity [18, 31]. The ways in which exoskeletons may have psychosocial benefit to individuals was also highlighted, including the potential improvement to a patient’s confidence and feeling of independence [18, 31].
Despite these proposed advantages, rigid exoskeletons have not been widely adopted clinically [32]. Although a systematic literature review on user perspective of rigid exoskeletons has been undertaken previously by Hill et al. [33], the review only included three papers which had limited reporting of qualitative data and their methods were predominantly quantitative components [29, 34, 35]. The review was inconclusive on user perspectives of rigid exoskeletons due to the minimal amount of evidence that has been undertaken; nevertheless, they concluded that users are able to offer their opinions, which may facilitate the design process. Since the publication of the Hill et al. [33] review, there has been further research into patient and physiotherapists’ perspectives; papers highlighted a range of rigid exoskeleton limitations, and they also recognised their novelty and potential [18, 25, 31]. Common perceived limitations or concerns regarding rigid exoskeletons across studies included: safety issues such as joint misalignment; creation of only one device to fit all patients; difficulty of use, including donning and doffing; weight and cost; and device appearance [18, 25, 31, 36]. User perspectives for traditional rigid exoskeletons demonstrated a disconnect between those with clinical knowledge who understand the requirements of assistive devices, and the engineers with the technical knowledge to create such devices [25].
Soft-robotics is an emerging field with capabilities to address the limitations of bulky, rigid and heavy exoskeletons through designing wearable, soft devices that are lightweight, compliant and flexible, resulting in safe human interaction, suitable for body assistance [37, 38]. Soft exoskeletons are different from rigid exoskeletons as they have an interface with the wearer that is a non-rigid structure, for instance, textiles, velcro or straps [39]. A device may also be a hybrid of both, and would therefore not technically be a fully soft exoskeleton (despite sometimes being described as a soft exoskeleton). Examples include an exoskeleton with compliant/soft actuators but a rigid structure-based body attachment, or a system with rigid actuators mounted on the body using a soft structure. A review of 52 lower-limb exoskeletons found that only 11% were fully soft exoskeletons [40]. A recent review on soft wearable robots reported an exponential growth of using pneumatic artificial muscles (PAMs), electrically-driven actuators, and textiles/fabric-based actuators over the past 10 years [41].
If the advantages of soft robotics are to be realised and implemented effectively and transitioned rapidly into clinical and community settings, the engineering and clinical gap must be reduced. Consequently, this state-of-the-art review aims to identify the limitations of existing exoskeletons as perceived by clinicians and patients, and discuss the potential soft-robotic solutions. The review informs FREEHAB, a project which aims to design wearable, assistive soft-robotic devices for people with impaired mobility (project website - The Right Trousers [42]). This collaborative review was undertaken by Freehab researchers with both clinical (LM) and engineering (RSD, NR) expertise, reducing the clinical-engineering disconnect.
The "Methods" section will outline the methods of the review. The "Patient and physiotherapist perceived limitations of exoskeletons" section will discuss the patient and physiotherapist perceived limitations of existing rigid exoskeletons based upon a review of the existing literature. The "Patient and physiotherapist perceived limitations of exoskeletons" section will also introduce potential solutions to these limitations. The "Discussion" section will summarise the main findings in relation to wider literature, and has a wider discussion on future assistive suits and identifies how soft-robotic technologies may provide solutions.
The findings’ framework is based upon six themes: safety, one size fits all, ease of device use, weight and placement of device, cost of device, and appearance. Themes did not change from LM’s consultation of MC and AT, however, the presentation of themes were ordered to reflect the weighting of the literature for the themes. Themes are presented below with their relevant findings. Five papers were included that evaluated patient and/or physiotherapist perspectives of exoskeletons.
Across the sources, safety concerns regarding use of rigid exoskeletons can be divided into: primary harm incidents, secondary harm, and concerns of infection control.
Several concerns were expressed regarding primary harms that could be caused by a lack of device sensitivity and sophistication. In one qualitative study, physiotherapy student views on the H2 rigid exoskeleton were explored; they had not used the device but had only seen a participant don/doff the device or had the process described to them (two out of three focus groups were online) and viewed videos of people walking with assistance from the device [18]. The students felt that the device may force a patient to go past their joint range of movement, causing injury [18]. Similarly, a physiotherapist in Vaughan et al.’s study [25] (who had actually used the device for one year) expressed concerns for how the device would respond to atypical muscle tone, providing an imagined example of a patient with hyper-reflexia and the device misinterpreting this as sitting. Physiotherapists in Read et al.’s study [31] had undertaken training with the Ekso rigid exoskeleton and used it as part of their interventions. The physiotherapists highlighted that this device could cause a larger degree of spasticity, due to patient anxiety for use of the equipment. Thus, it has been perceived by physiotherapists that the device could not only cause injury in its response to muscle tone, but it could exacerbate the problem. It should be noted that although these concerns were not based on actual experiences, nonetheless, they remained present even after training and use of the devices for several months (training spring/summer and data collection July ).
Primary concerns also included the device being inappropriate for particular patient groups. Physiotherapists were apprehensive that, for patients with limited core strength, the device may impact their balance and cause injury, with a physiotherapist referring to it as ‘throwing them through the motions’ [25] (p0.11). Concerns regarding falls were shared by several users of wheelchairs [34]. Consequently, a physiotherapist (with no experience of using the device) stated that they would only feel comfortable using the device if they maintained close proximity when mobilising patients [25]. There were reservations for using the device with patients with cognitive and communication deficits; however, discussion was limited [18, 25].
A secondary safety concern, with expression limited to one physiotherapy student, but with particular relevance in light of the COVID-19 pandemic, was the ease of cleaning the device [18]. Predominantly, secondary safety concerns were in relation to creating one device to fit all.
Compliance as a physical property is one of the most important required features in developed exoskeletons, facilitating adaptability, comfort and safe interaction with human body [46]. Series Elastic Actuators (SEAs) and Variable-stiffness Actuators (VSAs) are commonly used in conventional rigid spring-based exoskeletons because of their ability to change their stiffness [40]. Despite comprising rigid elements, SEAs and VSAs are considered compliant actuators with the capabilities of transmitting high forces and providing smooth assistance and avoiding restriction of natural body motions and injury. However, they have certain disadvantages such as mechanical friction and hysteresis (actuation delay). Often, the empirical behaviour such as the device’s practical stiffness does not match with prediction [40]; the fitting of the rigid exoskeletons determines the expected rotation axis of the joints and, therefore, it dictates the constraints and an undesired range of movement applied to the person. In addition, traditional rigid exoskeletons exploit self-adjustable body attachments to compensate joint misalignment [47]. For example, a rigid full-DoF hip exoskeleton containing rotational hinges and perpendicular sliders has been shown to passively adapt to align the exoskeleton’s components with the user’s specific body biometrics, reducing undesired interaction forces [48]. Similarly, iT-Knee is a self-aligning rigid knee exoskeleton, delivering autonomous adaptability, while assisting lower limbs, and pure knee assistance decoupled from other joint movements [49]. However, inclusion of these joint-aligning mechanisms increases the complexity, weight and rigidity of exoskeletons.
The decrease in rigid components can be observed throughout the history of developments in exoskeletons using a variety of soft robotics technology to pursue totally soft exoskeletons, which is safer, more comfortable and friendly to users (Additional file 1: Fig. S1). Pneumatic Artificial Muscles (PAMs) are common soft actuators that change shape and exert forces when pressurised by air [50, 51]. PAMs have been used for rehabilitating soft exoskeletons over several decades [52,53,54,55,56,57,58,59,60,61,62,63,64,65,66]. Cable-driven soft exoskeletons consist of minor rigid components (e.g. motors, gears and cables) which use textile or soft attachments to deliver direct force transmission and comfort to a user’s body during assistance. These actuators are predominantly off-board (not attached to the suit, but tethered to the suit through cables), reducing the suit weight loaded on a user’s body and making them suitable for use with a therapist who is able to assist carrying the actuator [43, 67,68,69,70,71]. Moreover, more recent PVC gel electroactive polymer actuators have been used to build soft exoskeletons [72]. When assisting the human body, it can decrease the activity and energy spent by the targeted muscles [73]. However, there may be safety concerns due to the high voltage supply required.
Although rigid exoskeletons are designed with the intention to facilitate motor learning of a typical gait pattern, physiotherapists felt that the device may cause unnatural movement. A pre-fixed gait pattern in the sagittal plane was perceived as imposing a gait pattern on a patient that did not correlate with real life walking [18]. Moreover, it was highlighted that a one-size-fits-all device had unnatural hip/pelvic alignment and knee alignment, and subsequently impacted their base of support and caused new compensatory movements [25]. This was the experiences of several stroke survivors who used a rigid exoskeleton and stated that the device ‘felt unnatural’ and made it harder to transfer their weight between legs [25] (p0.7). Furthermore, there were concerns that providing too much assistance to an individual could cause passivity or increased dependence. A physiotherapy student voiced concerns that patients may be unaware of the feedback they are receiving from the rigid exoskeleton, which can lead to device dependence [18]. This was the experience of a stroke survivor; they felt the rigid exoskeleton’s pre-programmed gait pattern was walking for them, rather than providing the assistance they required [25].
Consequently, physiotherapists requested an exoskeleton that they could tailor to the needs of each specific patient. For instance, they wanted to adjust joint angles, length of the femur and the swing pattern [25]. Physiotherapists highlighted the issue of time consumed from altering an exoskeleton for each patient if it were in a rehabilitation facility clinic, and suggested that the device have a function to retrieve patient specific settings [25]. Rather than a one-size-fits-all device, both physiotherapists and several stroke survivors felt that a range of exoskeletons may be required, to meet the differing needs of early and later stages of rehabilitation, in both acute care and community rehabilitation [18, 25].
There were concerns that it might not be possible to have an exoskeleton that is suitable for all [31]. Pragmatically, physiotherapists expressed issues in having one device that can physically fit every patient, and they felt skills may be required to fit the device so that it was comfortable for the patient for an entire session [31]. In a survey exploring how users of wheelchairs perceive exoskeletons, some were concerned that the device would not be suitable for their impairment, with examples provided include: hemiplegia, quadriplegia, low bone density, contractures, lack of arm/hand use, poor balance, amputation, obesity, muscular dystrophy and asymmetrical lower extremities [34]. However, as this was an open-ended survey question, there was limited depth into why patients perceived these impairments as preventing their use of exoskeletons.
From the engineering perspective, the goal of soft assistive devices is to generate predefined trajectories that train patients and recover their body motions, while simultaneously adapting to each patient’s specific needs. Therefore, rehabilitative devices must determine the assistive conditions, estimating the amount of required assistance, and timing of activation and withdrawal of assistance [74]. However, these parameters vary significantly between patients, causing time-consuming adjustment.
A sophisticated control algorithm, called human-in-the-loop optimisation was developed for cable-driven exoskeletons to solve these issues. The algorithm is able to adapt its assistance strategy based on the individual’s walking performance [69, 70]. Although starting with a standard walk-assisting force profile, this algorithm enables the exoskeletons to rapidly adjust the assisting force and provide precise actuation to match the cyclic walking motions, improving and maintaining optimal walking performance at all time [69, 70].
Flexibility and adaptability, which may be offered by novel textiles, are the key solution to the problems of rigidity and low adaptability with existing exoskeletons (See the “Safety” section). Flexible textile materials, such as soft braces, straps and garments, are currently the most effective solution for the compliant connection between an exoskeleton and the user’s body. They are used to transfer assisting forces and hold the suit in an appropriate configuration, providing compatibility with natural motions and safety [40, 75]. Moreover, semi-soft exoskeletons, containing rigid components and soft attachments, were designed to include adaptability by changing the length of their rigid linkages to fit various body sizes. However, this component-adjustment process consumes considerable time for each patient. Alternatively, variable-stiffness materials may be integrated into body attachments. For example, a 3D-printed variable-stiffness structure stimulated by heat and electricity may be incorporated into adaptive body attachments [76].
The ease by which an exoskeleton can be used was a common theme across several studies [18, 25, 31, 34, 35]. Wollf et al. [34] highlighted that, out of 17 criteria for importance of exoskeleton design features, ease of device use was number 4 for users of wheelchairs. Bortole et al. [35] highlighted that, for their sample of stroke patients, ease of exoskeleton use was ranked as 7.2 on a Likert scale, with 10 being ‘extremely easy to use’; one patient expressed that ‘wearing it is fast and simple’ (p0.11). Time was a common rationale for wanting the exoskeleton to be easy to use, which was frequently related to donning and doffing the device [25, 31, 34]. One stroke survivor stated that it took 30-40 minutes to fit the device which they felt was ‘lengthy’ (p0.5), while another stroke survivor did not want half their therapy time being absorbed in this way [25] Read et al. [31] stated that sessions took 60 minutes using the Ekso rigid exoskeleton and 90-120 minutes for the initial assessment, and physiotherapists highlighted the time-consuming nature of the device.
Time management was also discussed in relation to the training required for physiotherapists to use the device. Physiotherapists felt this training had to be multi-faceted to address the complexities of the device, which included: device-specific technical know-how; maintaining patient safety; the necessity of working within time constraints; deciding the appropriateness of the technology in a given situation; accurately acquiring the measurements required for proper fit; a need to check and recheck that all is operating as intended; and an understanding of patient needs [31]. The physiotherapists perceived the training sessions as ‘challenging’, and it was stated that they felt the skills had to be maintained (p0.5). Physiotherapists in Vaughan-Graham et al. [25] perceived it as essential for ongoing support and training, while physiotherapy students wanted a contactable technical expert in case of exoskeleton issues [18]. Physiotherapy students felt that training may be a deterrent for the exoskeleton’s use, particularly for less experienced therapists who find time management more challenging [18].
There were also ease of use factors that were specific to certain patient groups. This included the ability to use exoskeletons without crutches, which was ranked as 3.71 out of 5 on the Likert scale in terms of importance of design feature (ranked 14 out of 17) [34]. This was expressed as important to participants with hemiplegia, Muscular Dystrophy and users of crutches who wanted to be able to have free use of their arms while carrying out activities such as cooking [34]. Physiotherapists also expressed concerns for ease of use for patients with cognitive, perceptual, and communication impairments, and the potential for harm while using the device [18, 25].
In terms of the design considerations for engineers, it is essential that the device is easy to put on and take off. A soft trouser, which integrates soft robotic actuators, (rather than a mechanical assembly that constrains and anchors certain areas in the body) is easier to wear and reduces unproductive time during therapy. The challenge remains for the soft structure to effectively interface with the body and to provide sufficient torque/force and tension/compression to deliver mechanical assistance as required [46]. A lightweight, portable, active undressing trouser [77] is a good example of a compliant exoskeleton integrating a soft pneumatic adaptive belt inside a regular trouser. It can expand and loosen due to the pressure input supplied by a compressed gas cartridge, allowing ease of donning and doffing. As mentioned previously (See the “One size fits all” section), the autonomous control-optimising algorithm can also vitally decrease time and effort therapists spent to manually adjust the exoskeleton setting for each patient [70].
Studies commonly referenced weight, size and placement of exoskeletons. In the Vaughan-Graham et al. study [25], both stroke surivors and physiotherapists were concerned about the weight of the rigid exoskeleton; in particular, therapists felt the device’s pelvic placement could result in the patient’s centre of mass being shifted backwards, disturbing their gait pattern. Wollf et al. [34] survey demonstrated that, for users of wheelchairs, portability of the device was 4.09 out of 5 in importance of exoskeleton features (ranked 10 out of 17 features). In the same study, battery life of the device was ranked number 6 - a consideration that was shared by physiotherapists - suggesting that a potential trade-off exists between operating time and device weight/portability [25, 34]. A further design consideration highlighted was the potential for skin breakdown due to pressure sores; however, this response was limited to one patient [34].
Reducing the weight of exoskeletons has been a key focus for engineers designing rehabilitative and assistive devices. Existing rigid portable exoskeletons aim to restore mobility of disabled patients, examples include Exo-H2, Indego, ReWalk and EksoNR, have masses between 11 and 25 kg [78]. They can support and move patients with mass over 100 kg. In contrast, compliant exoskeletons using off-board tethered cable-driven actuators and light textile attachments to transmit assisting force to a user’s body have significantly lower masses of around 0.9 kg (weight of suit alone), dramatically reducing parasitic or undesirable loads on the body [43, 68,69,70]. Although extremely lightweight compliant assistance devices have been shown to improve walking, their operation has been limited to treadmill training because of difficulty in carrying off-board, heavy actuation units and energy source. Myosuit [71] and XoSoft [79] are examples of portable cable-driven exoskeletons including on-board actuators with entire masses of 4.1 and 4.6 kg, respectively, significantly lighter than a contrasting cable-driven exosuit (12.15 kg) [67]. The mass of pneumatic-driven exoskeletons can be even lower (less than 0.16 kg); however, this excludes the weight of the heavy pumps and compressors required for the air supply [63, 66].
Placement of body attachments is another critical part of exoskeletons. Key anchors, such as the shoulder, the iliac crest of the hip and the plantar surface of the feet, are defined as effective body locations onto which to attach assistive devices [57]. These specific areas have the thinnest skin above the bone compared to other surrounding areas of the body, which can prevent misalignment, pressure sore, skin damages and muscle injuries while transmitting assisting forces to the skeleton.
Cost of exoskeletons was a consideration shared by both physiotherapists and patients [18, 25, 34]. The survey of wheelchair users highlighted that the exoskeleton cost was second in importance only to design features (2/17), with the open-ended responses demonstrating several patient concerns of affordability [34]. Physiotherapists likewise discussed the concerns for the cost of any devices to patients [25]. Furthermore, physiotherapists and physiotherapy students perceived the difficulty in funding this across a facility budget, and a private clinic, respectively [18, 25]. A physiotherapist felt that, if the device were to be shared across entire programs, there could be added complications regarding bookings, and another physiotherapist remarked that there had to be a significant benefit in order to justify the cost [25]. It was perceived that the cost of exoskeletons could hinder clinical uptake [18, 25].
To the authors’ knowledge, there is no literature on cost evaluation of exoskeletons or use of cost reducing materials. However, the soft exoskeletons can generally separate into two major parts: a suit with integrated soft actuators, and a power-supplying unit. Presently, there is an increasing trend to use more fabrics to create soft exoskeletons [80]. These soft devices are low cost and affordable, and they have the additional benefit of minimising the required power-supply unit, thus increasing its portability.
Appearance of exoskeletons was a patient consideration; however, it was ranked as the least important device criteria for users of wheelchairs (3.23 mean importance out of 5) [25, 34]. There appeared to be variation in expectations of the exoskeleton appearance; one participant remarked that they did not want to look like a robot and the device needed to be wearable under clothes, while another participant expressed accepting the limitations of the device appearance in order to improve their mobility. The minimal level of importance of device appearance in the Wolff et al. [34] survey only provides an average of mean importance and it may not reflect the strength and variety of opinions of patients.
Existing soft exoskeletons have been designed with aesthetics and discretion as important considerations, while maintaining high body-assisting performance [46]. Textile materials and garments have been used as the main components of the soft exoskeletons to safely interact with a user’s body, delivering assistance in a discreet manner. One excellent example is integrating soft actuators inside a normal trouser for undressing assistance [77].
This state-of-the-art review has highlighted a range of patient and physiotherapist perceived concerns and design considerations regarding rigid and soft exoskeletons. Patient and physiotherapists had reservations regarding rigid exoskeletons’ safety, such as falls, despite not having an experience that would warrant this or having seen wider evidence that demonstrates that falls were only present in testing of early exoskeleton prototypes [81]. The significance placed upon falling is unsurprising; falls have widespread physical and psychological impact on individuals and a survey of wheelchair users demonstrated that minimising the risk of falling was their top priority when evaluating exoskeleton function [34]. Other concerns shared by the users of rigid exoskeletons were that the device may force patients into unnatural movement patterns, or even erroneously detecting and forcing movements, for instance, interpreting the patient as trying to sit when they are in fact walking. This paper then explored the use of soft technologies with greater sophistication, which are more sensitive to the users’ movements and are thus able to more precisely apply forces as required.
From an engineering perspective, the first challenge is developing new soft actuators that are lightweight, compliant and sufficiently powerful enough to deliver smooth and safe assistance for natural mobility [82, 83]. They must be capable of varying their stiffness and deforming when exposed to external forces generated by patients, reducing unexpected harms. Furthermore, they must be suitable to fit inside a garment-like suit that is wearable and comfortable. The level of stiffness is directly related to the amount of force that can be transmitted to the human body; therefore, the balance between stiffness and assistive effectiveness must be considered. For example, compliant actuators can deliver safety and comfort together with portability due to their low weight, but may have decreased bandwidth or peak force when compared to conventional heavier rigid actuation technologies such as motors and gears [46]. Additionally, actuation speed must be considered when delivering assistance that merges seamlessly with natural body movements, since asynchronous actuation can result in increasing muscle power consumption and fatigue of users and negative changes in mobility patterns.
Both physiotherapists and patients were concerned that if there were only one exoskeleton device that was made to fit all patients, it may result in unnatural movements [18, 25], which was the experience of users in the recent Vaughan et al. study, where there was unnatural hip/pelvic alignment [25]. A longitudinal study of healthy participants demonstrated the individual nature of gait characteristics, and concluded that assessment and therapy must therefore take into account the patient’s unique differences [84]. This review highlighted that physiotherapists wanted a device that they could tailor to an individual [25]. However, they did note that altering a device between patients may be time-consuming [25], and they were concerned that it may not be possible to have a device that can physically fit all patients [31]. It was important for the device to be easy to use, both in terms of donning/doffing the device and training. However, rigid exoskeletons were not easy to use, with one study citing 30-40 minutes simply to fit the device [25]. As previously highlighted, typical therapy time for stroke survivors is already below the recommended 45 minutes per day [10]. A systematic review exploring the clinical applications of the HAL exoskeleton [28] found that, of the three papers that reported the breakdown of the time using the devices, all the sessions took 90 minutes, with one study reporting up to 60 minutes effective time, while the other two reported only 20-30 minute effective therapy time [85, 86]. Devices can be initially time-consuming; National Institute for Health and Care Excellence (NICE) [87] states that training physiotherapists to use an Ekso exoskeleton takes one week and the therapist must initially only use the device under supervision of a physiotherapist who is familiar with the Ekso. It is evident that future devices must be more efficient in relation to the required training and time taken to use the device.
We perceive that the future exoskeletons will become totally soft and naturally integrate with normal clothing. Their actuators and body attachments will be combined to create multifunctional assistive clothing with capabilities of stiffness variability and morphology deformation on every areas of the suit. For example, at no assistance, the entire suit turns soft, improving ease of donning and doffing. Moreover, it is perceived to be able to harvest energy gained from passive shape deformation and heat loss released from the body surface during daily activity. Also, when activated, certain areas of the suit can be controlled to become stiffer to support loads on lower limbs or to prevent undesired movements and joint misalignment. In contrast, other areas can actively vary their stiffness to deform, e.g. contract, elongate, bend and twist, and enable actuators to effectively transmit forces to assist required body movements and also prevent joint misalignment. In addition, active morphology deformation can enable suit adaptability to specific users and self-fitting for comfortable usage by controlling shrinking and loosening of the suit to fit the user’s body. A variety of developed textile actuators was classified depending on their purpose and reviewed in [75].
Physiotherapists and patients all highlighted the importance of having a device that was lightweight, to prevent the user’s gait pattern being altered [25] and to allow it to be portable [34]. However, a long battery life was also desired, suggesting the need for a balance between a lightweight portable device that can operate for a short period and a heavier device that has longevity of use [34, 88]. As therapy time is limited to 35 minutes for stroke survivors [10], this would be the minimal time accepted, however, if the device is to be used by multiple patients in a day, greater device battery capacity or a short recharge time may be required. The requirements for a lighter assistive device are soft and efficient actuators (high torque-power and torque-weight ratio) and power sources that have short recharge time. However, both these requirements urge development in the material science. Within the current technology, wireless charging during users’ session can be adopted to optimise standalone charging time.
The cost of the device was important to both patients and physiotherapists, in both terms of cost for the patient to buy for personal use, and also the purchasing by clinics [18, 25, 31]. Physiotherapists would only want to purchase a device if the device benefits clearly outweighed the cost [25]. In the UK, there are several bodies involved in making budget decisions, passing from the Department of Health and Social Care, to National Health Service then to Integrated Care systems (previously Core Clinical Commissioning Groups) [89, 90]. It can be expected that approval from commissioners for high-cost services or products, such as soft exoskeletons, will be challenging. An economic evaluation explored the cost-effectiveness of rigid exoskeletons in improving quality of life and preventing secondary hip fractures in an imagined population of people with dementia or cardiovascular diseases [91]. The multiple scenarios demonstrated that a significant improvement in reducing hip fractures was not essential, however it was essential to improve quality of life in order to justify the cost of the exoskeleton (with the cost under £17,500) [91]. In , Ekso was provided to one NHS Trust in a package costing £98,000 (excluding VAT), which included the Ekso GT robotic exoskeleton with the SmartAssist software, training for up to four physiotherapists, a two?year warranty, supporting equipment [92]. This cost is significantly greater than the value outlined in the cost-benefit analysis, and it is therefore evident that future exoskeletons must have greater affordability.
Although the appearance of rigid exoskeletons was commonly discussed by patients, their views varied, with some feeling appearance was irrelevant if the device allowed them to live their life, and others not wanting to feel stigmatised by a visible device [34]. This demonstrates the individuality of patient needs when designing future exoskeletons.
To build the next generation rehabilitative device, we propose that developments are required in i) building efficient soft robotic actuators, ii) fabricating comfortable body attachments, iii) improving sensor technology and iv) developing robust adaptive control strategies.
Similar to the key metric performance of human skeletal muscles, soft fluidic and electrically-driven actuators have high potential to become future smart artificial muscles due to their high stress, strain and bandwidth, addressing the requirements to create the future soft exoskeletons [93].
Soft fluidic actuators have backdrivability (behaving like a spring), low cost and high specific force and power, despite low efficiency around 30% because of losses in the fluidic-to-mechanical energy conversion process [40, 93]. They can be divided into hydraulic and pneumatic actuators. Soft pneumatic actuators have lower bandwidth than hydraulic actuators due to gas compression, but are significantly lighter, making them more suitable to create soft exoskeletons (see Additional file 1: Fig. S2 for the vision for future assist devices). Many prototype soft lightweight exoskeletons were built using PAMs [53], for example, McKibben muscles [51, 94], straight-fibre muscles [95,96,97,98], Pouch Motor actuators [99] and pleated PAMs [100] including Bubble Artificial Muscles [64, 65]. However, one drawback of pneumatic artificial muscles is the requirement for large, heavy, noisy pumps or compressors, which limits their portability for ambulatory applications.
Soft electrically-driven actuators are another potential candidate for future soft exoskeletons due to their high actuation performance, fast and quiet operation, and high efficiency [46, 93]. For example, dielectric elastomer actuators (DEA) and ionic polymer-metal composites (IPMC) are active polymers which deform when electrically charged. DEAs can deliver high strain, bandwidth and efficiency, and are widely used in robotic applications [101,102,103]. However, the major disadvantage of DEAs are the requirement for high voltage actuation (typically in the order of thousands of volts). Alternatively, low voltage actuation is achieved in IPMC (in the order of a few volts) but at the cost of lower stress and power density [93]. Dielectric fluid electrostatic actuators have been developed as a new soft robotics actuation technology. Examples include dielectrophoretic liquid zipping actuator [104,105,106], hydraulically amplified self-healing electrostatic (HASEL) actuators [107,108,109], soft fluidic pumps, such as a stretc.hable pump [110] and an electro-pneumatic pump [111, 112]. The recent stretc.hable pump and electro-pneumatic pump demonstrated important capability to integrate with soft fluidic actuators, e.g. hydraulic McKibben muscles [113] and pneumatic Bubble Artificial Muscles, respectively. Together with the development of lightweight battery technologies and on-board controllers, these soft electrostatic actuators promise a new generation of entirely soft, lightweight, flexible and portable exoskeletons. Although their stress-strain performances are lower than human muscle requirement, they combine the benefits and address the drawbacks of both fluidic and electrically-driven actuators.
Auxetic or transformable structures has been explored to develop ankle and knee braces that are lightweight and comfortable, similar to a textile or garment [114]. Exploiting the advancement of additive manufacturing and 3D printing, the transformable structure can be actuated using micro pneumatic actuators or by pulling threads. There are numerous auxetic structures available, and their potential has not been fully investigated in actuation. Many of the transformable structures can bend, twist and contract, and their characteristics can be altered depending on the way they are placed, such as unit type (e.g. block size, number of edges, number of hinges/constraints or layers) and connection type (e.g. series or parallel). These structures can be easily incorporated within the garment and their ability to transform in shape can be used to move joints or alter the local stiffness of parts of a suit. A combination of different auxetic families in a garment could be helpful to obtain desired strains in different activities such as sit-to-stand or walking. These make active auxetic structures another potential solution to develop future soft exoskeletons. Additional file 1: Fig. S1 demonstrates the perceptual concept of the future assist device, illustrating soft artificial muscles, sensors and a power supplying unit.
The conditions of body attachment, such as size, shape and attachment point, play a crucial role in patient comfort that form an essential focus of evaluation for the performance of soft wearable exoskeletons. As previously discussed, the safe placement for attachments of the soft exoskeletons on the user’s body is a shoulder, hip and feet [57]. In addition, a guideline for designing aesthetic and inconspicuous exoskeletons is provided in Veale et al. [46]. That is, the loads from the suit mounted on trunk and each foot are recommended to be less than 15% and 1.25% compared to the user’s body weight, respectively. The thickness of the suit along lower limbs should be less than 30 mm. The power source can be located on the user’s back with the total volume limited to 0.023 m\(^{3}\).
Although not considered by the patients and physiotherapists, we perceive that breathability is an important feature of body attachment and can significantly enhance comfort while wearing the soft exoskelton. This can be achieved by temperature and humidity exchanges on the skin surface. For example, the suit can possess self-deformation, which is sensitive to skin surface temperature and humidity and autonomously deforms its structure to inhale/exhale surrounding air for cooling/warming of the suit [115].
Typically, exoskeletons require a sensing system to be highly accurate, fast, and robust to disturbances. The sensing system is accountable for three fundamental features: (1) measuring force generation, (2) monitoring body motions, and (3) evaluating assistance efficiency. First, the soft exoskeletons require sensors to measure their actuators’ outputs, which are then fed back to their controller to deliver accurate assisting force and strain. Sensors are also used to prevent undesired damage, for example, stopping actuation when high forces and out-of-range motions are detected.
Second, IMU (inertial measurement unit) sensors are commonly used to track and simulate body movements by their attachment across body joints. With the development in a real-time data-analysing system, IMUs were able to precisely predict walking and sit-to-stand activities [116, 117]. Alternatively, soft flat sensors, such as DEA stretc.h sensors [118] and a multi-bend/shape sensor [119], which can contract, extend, bend and twist, may be integrated as one layer of the soft exoskeleton. These soft sensors can be used to measure and simulate the entire 3D configuration of the suit in order to predict mobility activities and to detect incorrect mobility symptoms or patterns. Soft force sensors can be intergrated in the the soft exoskeleton to estimate the pressure distribution on the user’s skin and thereby to prevent skin damage, especially around the attachments, and adjusting the assisting strategy or altering the exoskeleon morphology for more user comfort and safety.
We envision that all of these functional sensors need to be soft and aesthetically embedded within soft exoskeletons to comprehensively monitor the user’s body. The future of assistive soft robotic clothing may include multiple layers that contain actuating, sensing and control units, operating synchronously to facilitate a person’s improved mobility.
As a result of the body motion information acquired from suit sensors, a virtual mobility simulation can be simultaneously created. Additionally, an open source platform that can capture the motion from the suit with plug-and-play capabilities may facilitate rapid analysis of patient mobility patterns. The integration of the above technologies with an online data communication and service has the potential to track mobility improvements and enable remote patient consultations with physiotherapists.
Exoskeletons must always generate the appropriate amount of assisting forces at the right time for effective assistance without negatively impacting a user’s mobility. As previously shown, human-in-the-loop optimisation was developed as an advanced control algorithm which can automatically adapt the assisting strategy based on the user’s mobility, consistently delivering effective, harmless and optimal assistance [69, 70]. Although based on the same predetermined assisting profile, this algorithm can rapidly adjust itself to suit the mobility needs of a variety of different patients, conserving therapy time. In order to deliver such sophistication, future exoskeletons require a soft, high-speed, micro-scale, computing controllers that can be unobtrusively distributed across the suit and which communciate and integrate to deliver low-level mechanical assistance to deliver a high-level mobility goal.
Implementation will only be successful if devices are themselves effective and that this can be proven in clinical and usability trials. A recent Cochrane review [120] of 62 trials (totalling participants) was carried out to determine whether electromechanical-and-robot-assisted gait training versus normal care after stroke improved walking. It concluded that stroke survivors who received electromechanical-and-robot-assisted gait training alongside physiotherapy were more likely to achieve independent walking than those who carried out gait training without these devices. However, the study concluded that questions remained regarding the most effective frequency and duration of the training and which design characteristics were important for bringing about the improvement. This was in part due to differences in device design, for instance, some devices having FES. Further, the review concluded that the variation in time since stroke needs to be considered in future testing as the training may not benefit those who were ambulatory at the beginning of the intervention. Consequently, it is necessary to understand specifically what about devices make them ‘effective’ and for who, and what outcome measures class this ‘effectiveness’.
However, measuring effectiveness is a problem that is widespread across the robotics field. A systematic review [121] explored the effectiveness of platform-based robotic rehabilitative devices (devices which solely improve ankle performance) for use with people with musculoskeletal or neurological impairment. They could only conclude on the effectiveness of two devices due to the availability of evidence. However, they found both devices to be effective in improving ankle range of movement and stability. They proposed that, to reach further conclusions for these devices, there must be future work into creating universally accepted evaluation criteria which are able to standardise the devices’ outcome evaluations. But this is not without its challenges. Flynn et al. [122] explored the sustainability of upper limb robotic therapy for stroke survivors in an inpatient rehabilitation setting. They highlighted that robotic devices can create very subtle improvements or change in quality and degree of movement, which may be missed by less sensitive commonly used outcome measures. They underlined the need for an understanding of the clinical reasoning that underpins the prescription of robotics for upper-limb therapy by physiotherapists, as well as the most effective time to incorporate it into rehabilitation. This study is based upon upper-limb therapy, however, it is still relevant to the development of lower-limb therapy. Evidence demonstrates the need for greater guidelines for robotics in rehabilitation in general, which may encourage a ‘cultural shift’ for robotics acceptance [123].
Fundamentally, there must be an understanding of how therapists use devices and why therapists use devices in the way they do; this is essential to not only design robotic assistive devices that meet therapists’ needs, but also to create guidelines that can facilitate successful implementation of such devices. Notably, the need to understand the therapist’s clinical reasoning and the devices outcomes are interconnected. As Vaughn et al. [25] study exploring exoskeleton use in stroke rehabilitation concluded, unless client selection criteria and goals of device (both clinically reasoned) are established, there is a risk of exoskeletons lacking statistically significant effect. Future research should use interviews and focus groups with therapists not only to explore their clinical reasoning when using devices, but also to understand how devices might be tested meaningfully for their purpose, in a standardised way.
This review identified that physiotherapists felt that training professionals in exoskeleton use had to be multi-faceted, and that the training was often negatively perceived to be draining. Stephenson and Stephens [123] explored physiotherapists’ experiences of robotic therapy in upper limb rehabilitation within a stroke rehabilitation centre. Although initial starting up times for robotic devices may be greater than conventional methods, it was perceived that the devices could free up more experienced therapists’ time. However, they acknowledged that, without a greater evidence base, it was challenging to set the parameters for training. This reinforces our conclusion for the need to create clear clinical guidelines for device usage (clear device functions, recommended usage time and appropriate patient groups). Additionally, they concluded that close partnerships between the technical device manufacturers and the professionals using the devices were essential for successful training of staff, problem-solving and maintenance of equipment [123]. This further highlights the requirement for inter-disciplinary collaboration for robotic device uptake.
If you want to learn more, please visit our website Robot Joint Actuator.