Technological advancements in robotic devices have the potential to transform human mobility through gait assistance. However, the integration of physical hardware and software control algorithms with users to assist with impaired gait poses several challenges, such as allowing the user to adopt a variety of gaits and the process for evaluating the efficacy and performance of these assistive devices. Here, I discuss some of the challenges in the development of assistive devices and the use of biomechanical concepts and tools for control and test validation. Several potential solutions are proposed through the case study of one project that aimed to provide gait assistance for individuals with a spinal cord injury. Further challenges and future directions are discussed, with emphasis that diverse perspectives and approaches in gait assistance will accelerate engineering solutions towards regaining mobility.

1. Introduction

As technological advances imbue our world with machine capabilities never before imagined, robotics could become the solution to overcoming human limitations. Mobile robots, especially legged robots, could enable navigation and exploration in complex natural and human-built environments. With wearable exoskeletons, we hope that workers can perform their duties with reduced risk to injury or that we can run faster or farther than our biological legs alone could carry us. However, decades of research in these areas reveal that it is not trivial to design, build and test robotic devices that must interface with human users. Emerging ‘bio-mimetic’ or ‘bioinspired’ research looks to nature, such as our own bodies, for inspiration, but the translation of human biomechanics to robotics remains a significant challenge.

Lower-limb robotic devices for gait assistance, such as exoskeletons, require overcoming technical challenges and the difficulties of integration with humans. Current actuators, such as electric motors or pneumatic actuators, may be magnitudes more powerful than human muscles [1], yet their lower power-to-weight ratios preclude lower-limb exoskeletons from generating the dynamic physical behaviours observed in natural human walking. Even if dynamic motion could be achieved, it is unclear how the user’s desires and remaining volitional function should interface with the exoskeleton. For instance, how and when the robot should assist gait and how the user can signal his or her intentions to change gait are open questions. A wearable assistive device designed without properly integrated control and interaction with human users could lead to high rates of user abandonment, perhaps at higher rates than for unpowered orthoses (e.g. 26% abandonment rate for knee-ankle foot orthoses [2]), or at worst could harm the user or their recovery. Regardless of the hardware performance capabilities of current or future robot technologies, wearable robotic devices will not work or be used without supporting natural interaction and shared control between the user and the machine.

Bio-inspiration can take multiple forms, and understanding human biomechanics during locomotion could enable the design and build of better robotic devices for gait assistance. Early robotic devices focused on recreating gait kinematics for treadmill rehabilitation training [3]. Acquisition of appropriate trajectories could come from recording gait behaviour of healthy subjects and, beyond kinematics, include ground reaction forces to derive joint torques through inverse dynamics, muscle activity through electromyography and metabolic expenditure. How these data should be used and which aspects are relevant to informing the design of assistive exoskeletons and their controllers are unclear. Perhaps simply imposing gait kinematics or dynamics is already suitable for providing assistance, and there is no need to invoke underlying biological principles that may be complex to model. On the other hand, a biologically inspired controller that can recreate gait without pre-defined trajectories could be more adaptable to the user’s changing needs. Multiple challenges exist in understanding the principles of biological movement, an active area of research, and the translation and application of these principles to robot hardware and software.

Here, I provide a biomechanics perspective on the challenges of translating biomechanical theories and tools for robotic gait assistance with a focus on powered lower-limb exoskeleton devices. As a case study to contextualize these challenges and their potential solutions, I will also discuss how one particular project on assistive devices for individuals with spinal cord injuries sought to address and overcome them. This article is not meant to be an exhaustive overview of exoskeleton research for gait assistance but rather seeks to provide an in-depth look at understanding and resolving some of the ongoing challenges of the field through the lens of biomechanics.

2. From biomechanical models to exoskeleton control

Gait is complex, as the exact mechanisms within the musculoskeletal system which govern it are difficult to precisely characterize. Whether for scientific understanding of normative gait, developing interventions for impaired gait or recreating locomotion in robotic devices, various biomechanical theories and tools have allowed us to distill and conceptualize aspects of locomotor behaviour. These theories may include, for example, mechanical considerations from inverted pendulum models and neural coordination from central pattern generators. Tools such as motion capture and electromyography enable the capture of gait kinematics, kinetics and electrical activity of relevant muscles. These measurements could be used to test locomotor models or might form the foundation of desired behaviour to emulate. However, leveraging bio-inspiration by applying biomechanical results and tools to exoskeletons is not straightforward. Bio-inspiration could involve, for example, choosing different gait models as root principles, gait objectives to recreate and the number of degrees of freedom to actuate, each ranging with its own level of complexity (figure 1).

Figure 1.
Figure 1. Leveraging biomechanical concepts and tools is challenging, including defining and applying different complexity levels of bio-inspiration, providing intuitive shared control between the user and the exoskeleton, and evaluating the exoskeleton with appropriate efficacy measures. (Online version in colour.)

Mechanical models provide one tool to capture and understand the main determinants of gait behaviour. During steady-state walking, no mechanical work is performed overall, yet energetic costs are exacted because our interactions with the ground cost energy, specifically to recover the energy lost to heelstrike. As a simple model, the principles of locomotion can be distilled into inverted pendulum concepts. The model is a point mass on top of massless legs, and it may be passively driven by gravity walking down a ramp [4], or active in the form of impulsive push-offs [5,6] or forces from telescopic legs [7]. These inverted pendulum models, combined with positive or negative work minimization, are able to reproduce multiple aspects of gait, such as the mechanical consequences of changing step length and width [8,9] and the walk–run transition [7]. They also show that ankle push-off may be more efficient than hip work during stance to recover energy losses from heelstrike [10].

Simple models and their implications are not necessarily reflected in exoskeleton hardware, partially for practical reasons. Until recently, few full lower-limb exoskeletons had ankle actuation [11,12] and instead more commonly used passive spring-loaded ankle joints [1315]. Added mass at the ankle requires more energetic cost to swing the leg [16] and may necessitate stronger, perhaps heavier, hip motors to overcome the inertia of ankle actuators during leg swing. An inadequately quick leg swing could hinder corrective stepping manoeuvres against perturbations. Without ankle actuation, gait can still be generated through hip actuation [17], and a motorized knee can provide adequate foot-to-ground clearance, compliance for energy absorption and weight-bearing capabilities. The benefits of a fully actuated lower-limb exoskeleton spanning ankle, knee and hip joints are still to be determined.

Inverted pendulum models also demonstrate that walking can be generated dynamically rather than prescriptively because no work needs to be performed during the stance phase to produce walking behaviour. Healthy humans also rely on a combination of powered movement and passive dynamics to exhibit dynamic behaviour. By contrast, the overall motion of current exoskeletons is not dynamic, which seems sensible for maintaining user safety. Step generation is deliberate (e.g. from a user-enabled push-button), and the motion is restricted and generally keeps the body centre of mass over the base of support for stability. Prescriptive control may be preferred, more stable, and repeatable because the gait movements the exoskeleton will produce are known. Exhibiting dynamic behaviour may be an unsuitable goal for individuals with impaired gait, especially in early stages of rehabilitation or recovery.

There are other forms of bio-inspiration aside from simple inverted pendulum models with limited means to investigate muscular coordination or neural control. Central pattern generators to reproduce the rhythmic behaviour of human gait offer advantages of speed modulation by tuning only one parameter [18] and in exoskeletons, have produced gait patterns that can adjust to different slopes and uneven terrain [19,20]. Motor primitives could also reduce the complexity of gait coordination into a few basic components, generating smooth transitions among different locomotor tasks in response to changing environments [21], and have been successfully applied to exoskeleton control [2224]. While more complex to simulate, neuromechanical models integrate neural control, sensory information and muscle dynamics [25,26], and offer additional opportunities for bioinspired control of assistive devices [22,27,28].

3. Challenges

With the difficulties of applying biomechanical principles and methods to robotic assistive devices, it is unsurprising that facilitating gait for individuals with mobility impairments is not straightforward and poses multiple ongoing challenges. Here, I discuss three challenges—generating gait for various walking scenarios, facilitating human–machine interaction, and evaluating the efficacy and performance of the exoskeleton with its intended population (figure 1). All three are linked to differing levels of bio-inspiration discussed in the previous section. The choice of underlying model or gait objective has implications for how the exoskeleton can generate a range of gait behaviour. Efficacy asks whether biological data from healthy subjects are appropriate standards for comparison. Shared control seeks natural human interactions with mechanical devices, similar to the way humans interact with their biological limbs. For the sake of isolating these challenges with a biomechanics-related perspective, we assume that mechanical hardware challenges have already been met. In principle, we have a lightweight exoskeleton with actuators that are capable of fast motion, high torques and backdrivability (i.e. unpowered motors move freely with little resistance).

(a) Generation of different gaits

Healthy gait behaviour recorded with biomechanical tools can be used to instruct exoskeleton controllers, but generating a wide breadth of movements is an open challenge. It is unclear whether the robot should be controlled for a desired kinematic state, torque or impedance, or for other biological information. The desired state could be kinematic information (e.g. joint angles, foot trajectories), using position control to command actuators to follow set position or velocity trajectories as faithfully as possible [3,29]. High interaction loads between the user and the machine may exist, however, if the trajectories are too restrictive. Gait could also be generated by following desired joint torques instead of joint positions, but this control method does not guarantee that the produced kinematics will be reasonable or within feasible joint range of motion. A library or bank of trajectories can capture a range of desired gait behaviours, such as generated scaled joint trajectories for walking at different speeds [30], but probably cannot encompass the vast types of movements healthy humans can make during activities of daily living, such as walking on stairs, accelerating or decelerating, and performing sit-to-stand manoeuvres.

(b) Shared control between user and machine

Another key challenge is the interaction and shared control between the user and the exoskeleton, which can affect gait training or rehabilitation. In early devices for robotic assistance, the robot would play pre-defined lower limb angles, and the user, strapped to the robotic device, would effectively follow along these movements [3]. While these gait trainers were promising, it is unclear how these therapies aid human motor recovery. The user could be a passive bystander along for the ride, or perhaps the gait movements themselves could retrain leg muscles. Gait training has been shown to increase lower-limb muscle activity [28,31,32], but whether that increase in activity is meaningful or not is unclear. There is also the possibility for ‘slacking’, where the user increasingly relies on the robotic device to provide movement and support, and the little volitional contribution from the user could reduce or prevent the capacity for gait recovery [33].

Prescribing kinematic trajectories to each actuated joint seems safe for the user because the exoskeleton will move as commanded, assuming the trajectories are valid and any closed-loop tracking algorithms have been designed adequately. However, position control may restrict the user from positively engaging with the machine. Two classes of controllers—assist-as-needed and myoelectric control—may provide more shared control than simple playback of gait kinematics (see [34] for a comprehensive review of exoskeleton controllers). Assist-as-needed controllers allow users to deviate from defined trajectories and only act when the deviation is too large [3537], but these methods also require known trajectories for comparison. Myoelectric control is based on electromyographic recordings of target muscles of the involved limb and scales the amount of joint torque assistance by the user’s level of muscle activity [38,39]. However, this approach requires clean muscle activity signals, which may be difficult to obtain for certain populations.

The criteria for assigning shared control between the user and the machine for maximizing gait recovery or assistance is difficult to determine. The exoskeleton needs to provide the user with reasonable gait movement (prescriptive control), but the user also needs to be able to easily, ideally intuitively, influence or adjust the robot gait as necessary (interactive control, figure 1). For example, the user may want to indicate a desire to walk faster or simply that they would like to take a step. Current solutions include using push buttons to change modes or generate a step [40] or kinematic measures like forward trunk lean to initiate gait [41]. Push buttons are not a natural method for gait initiation, but erroneous trunk lean detection may also be problematic. Employing intuitive and robust methods to adjust gait is an ongoing challenge.

(c) Evaluating the efficacy of an exoskeleton

How does one evaluate the performance or efficacy of an assistive device, especially within a heterogeneous population, each with their own remaining gait function? In a controlled experiment with healthy subjects, one could compare walking with and without the exoskeleton, different exoskeleton controllers, and evaluate the changes in gait kinematics and kinetics and metabolic cost. Standard experimental procedures may be difficult to perform for ambulatory gait impaired individuals, who likely cannot walk as easily as non-impaired subjects. Others, such as individuals with complete paraplegia, may be unable to walk at all without the exoskeleton or other external assistance. Body weight support, which has been known to influence gait [42], may be needed and at differing capacities. Fatigue may limit the number of trials and thus the combination of controller parameters to be tested. Experimental conditions may also need to be adjusted as requested by clinicians if safety issues arise.

It is unclear if healthy gait should be the standard for comparison, and if so, which aspect to compare is appropriate. Past research has mostly used broad gait measures (spatio-temporal measures, kinematics, kinetics) and muscular effort to evaluate performance [43], perhaps due to the relative ease of acquiring these measurements through the robotic device itself or through standard gait equipment (see [43] for a review of performance criteria). Age-matched controls may allow for fairer comparisons of these measures, but individuals with gait impairments may walk at much slower speeds than preferred by their controls. Measured joint angles that are similar to healthy gait during flat walking also might not translate to comparable behaviour on uneven terrain. Additionally, it is unclear whether biomechanical measures are compatible with standard clinical measures, making reliable and cohesive conclusions difficult to draw.

For healthy individuals, exoskeleton developers seek to measurably decrease user’s effort (e.g. metabolic expenditure or muscle activity [38,44]) through the addition of mechanical work by the robotic device. While the assistive device should not tax the user, it is unclear if effort reductions are appropriate goals for individuals with gait impairments. Greater muscular effort may be desired for recovery, especially if there was little muscle activity prior to exoskeleton use. Still, challenges remain in measuring metabolic cost in less functional populations, and effort through electromyography may also be difficult to compare if the measured activation signals are noisy or abnormal.

4. Potential solutions from one assistive exoskeleton program

In this section, I discuss one exoskeleton project which strived to address the challenges outlined above, as a case study of how biomechanics can be leveraged to elucidate potential solutions. The goal of the Symbitron (‘symbiotic man–machine interaction’) project was to design and test a new wearable exoskeleton with a modular controller to enable gait for individuals with a spinal cord injury. As the assistive device was a lower-limb exoskeleton, we recruited test pilots (preferred term for our paraplegic study participants) that had good control of their upper body with lesion levels at C7 or below. The studies discussed below are from preliminary studies with several powered gait assistive devices. While I primarily focus on applying a biomechanics lens to these challenges, I will also briefly convey the importance of clinical measures and reconciling the gaps that remain between biomechanical and clinical perspectives.

(a) Bioinspired controller for gait assistance

To generate gait and allow for shared control, we applied a neuromuscular controller based on the sagittal plane neuromechanical model developed by Geyer & Herr [25]. The model has seven muscles per leg to actuate the ankle, knee and hip joints, and each muscle is approximated by a Hill-type model (figure 2). The model is governed by two main stance and swing reflex loops. During the stance reflex loop, the stance leg acts to bear weight, while during the swing reflex loop, the swing leg allows flexion to clear the ground. To achieve walking, the model requires few sensors, namely joint angles and foot contact, to determine virtual muscle length and the reflex loops to activate. Weights on reflex parameters were optimized for a human of mass 80 kg and height 1.8 m to walk at 1.3 m s−1 (see [45] for optimization details). The neuromechanical model has previously been applied to prostheses [46], and we were the first to apply the controller to lower-limb exoskeletons.

Figure 2.
Figure 2. The exoskeleton controller is based on a reflex model of gait with Hill-type muscles for actuating the ankle, knee and hip joints. Contact sensors determine whether to activate the stance or swing reflexes, and the muscle forces from the activated reflexes and joint angle measurements combine to produce joint torque commands to the exoskeleton. The seven muscles in the sagittal plane reflex model are the tibialis anterior (TA), soleus (SOL), gastrocnemius (GAS), vasti (VAS), hamstring (HAM), hip flexor (HFL) and gluteus (GLU). (Online version in colour.)


For the exoskeleton neuromuscular controller, the muscle models were organized into control modules for each joint. With these modules, targeted assistance could be specified for each leg, joint, and muscle separately (see [28] for additional details). For example, the controller could give the right ankle more plantarflexor torque and the left knee more torque support for both flexion and extension. The level of torque assistance at each joint was adjusted through a gain multiplier. A gain of 100% represents the nominal joint torque provided by the neuromuscular model (steady-state walking at 1.3 m s−1). A gain of 0% represents zero impedance mode, where the exoskeleton torques generated make the device feel transparent with no input from the neuromuscular controller.

The modular structure of the neuromuscular controller allowed its application on multiple assistive devices (figure 3), including the Achilles ankle exoskeleton [47], fixed gait trainer LOPES [14] with actuated hip and knee joints, and the full lower-limb Symbitron wearable exoskeleton [48]. Here, I discuss tests with the Achilles, LOPES and the knee–ankle configuration of the Symbitron exoskeleton (WE-KA) conducted on healthy subjects and test pilots with incomplete and complete paraplegia. The reflex-based gait generation allowed users to initiate gait as desired, yielding a measure of shared control. The commanded joint torques to the device actuators were not predefined and emerged from users’ joint kinematics and footfall patterns. Therefore, the controller does not require a library of trajectories and could potentially lead to dynamic walking behaviour.

Figure 3.
Figure 3. Several different assistive devices were tested with the neuromuscular controller. Achilles is a wearable exoskeleton with ankle actuation [47], LOPES is a fixed gait trainer with hip and knee actuation [14], and WE-KA is the knee-ankle actuated configuration of Symbitron wearable exoskeleton [48]. (Online version in colour.)


(b) Biomechanical approach to assess efficacy with healthy subjects

Initial tests with the neuromuscular controller used a powered ankle exoskeleton, Achilles, on two healthy adult subjects walking on a treadmill (figure 3; study details in [27]). Only the ankle module was used to provide bilateral torque assistance at gains of 50% and 100%. The biarticular gastrocnemius muscle was removed from the ankle model because there was no knee joint, and the soleus and tibialis anterior muscles remained. Metabolic cost was measured, along with gait kinematics and kinetics. The ankle torques provided by the exoskeleton were qualitatively similar to biological profiles. Promisingly, these preliminary tests showed reduced metabolic cost and muscle activity of the soleus and tibialis anterior. Inverse dynamics from one subject indicated that biological peak ankle moment and power decreased as the level of ankle assistance increased. Although subjects had differing anthropometric measurements and walked at speeds (0.58 m s−1 to 1.07 m s−1 on average) lower than the model’s nominal speed of 1.3 m s−1, using the same optimized parameters for both subjects had no apparent negative effects on their gait. This study was the first indication that the neuromuscular controller could robustly generate gait for various speeds without hindering users.

While the study with healthy subjects had encouraging results, it was difficult to translate them for our test pilots, individuals with incomplete or complete spinal cord injuries. Recreating the biomechanical testing procedure with healthy subjects was not possible for this population. Although some test pilots could walk with the remaining function in their lower limbs, we did not ask them to continuously walk the 4–6 min per trial on an instrumented treadmill to measure metabolic cost and ground reaction forces. Lesion levels, remaining ambulatory function, and any compensatory strategies also differed by individual. The performance measures of lowered metabolic cost and biological torques for healthy subjects were difficult to apply, and we attempted an indirect measure of energetic cost in the following study with test pilots.

(c) Biomechanical approach to assess efficacy with test pilots

We tested the neuromuscular controller with test pilots, both incomplete and complete, on a fixed gait trainer with body weight support (figure 3; study details in [28]). The gait trainer, LOPES, provided knee and hip actuation, and the unactuated ankles were held in a nominal position by a spring [14]. For this study, we used the knee and hip modules of the controller, including the gluteus, hip flexor, hamstring and vasti muscle models. Each test pilot walked at their own speed and with their own specifically tailored gains, both of which were adjusted based on their subjective needs, comfort level, and walking ability combined with their clinicians’ subjective evaluations of safety, gait quality, and perceived exertion. To evaluate efficacy, we investigated gait kinematics and kinetics, muscle activity, and the speed-step length relation. Joint kinematics and torques demonstrated that the controller was creating reasonable gait trajectories, with positive joint work increasing as walking speed increased. One notable exception was that knee flexion torques were not present during late stance. This divergent behaviour was due to the missing gastrocneumius muscle because the ankle module was not used, resulting in torque differences that had little effect on the resulting joint angles overall.

As alluded to earlier, quantitative analysis was challenging with this population, especially for determining controller performance. We used the optimal speed–step length relation as an approximate metric for energetic economy, which also enabled us to evaluate the controller over all recorded strides. For healthy walking, preferred step length increases as walking speed increases according to a power law, and deviations from this curve incur higher energetic costs [10,49,50] (shaded region in figure 4). It was uncertain if the same relation would hold for the controller with test pilots. We found that four out of five test pilots exhibited a power-like behaviour, but their step length increases were greater than normal, suggesting healthy-like but sub-optimal behaviour (LOPES test pilots in figure 4). It is difficult to ascertain if the increased step length was due solely to the controller because body weight support could have played a role [42].

Figure 4.
Figure 4. Changes in step length with walking speed from test pilots with the neuromuscular controller. Recorded steps are from test pilots (symbols) with LOPES [28] and WE-KA [48]. Healthy data (shaded region based on [51]) included for comparison, from which deviations (such as by test pilots at faster speeds) may lead to increased energetic effort [10,50]. Step length shown in both dimensionless (left axes, normalized by leg length) and SI form (right axes); speed shown in both dimensionless (lower axes, normalized by the square root of gravity and leg length) and SI form (upper axes). Test pilot names were kept consistent with previous publications [28,48]. (Online version in colour.)

(d) Combined biomechanical and clinical approach to assess efficacy with test pilots

Both biomechanical and clinical tools were applied for exoskeleton evaluation when the ankle exoskeleton with the neuromuscular controller was tested on four test pilots with incomplete spinal cord injuries. To measure efficacy, gait speed was evaluated before and after training with the ankle exoskeleton, enabling each test pilot to be compared against oneself. Test pilots underwent ten training sessions overground (study details in [52]). These tests were carried out under the expertise of clinicians from a rehabilitation hospital and research centre, who employed biomechanical measurements (walking speed and distance, ground reaction forces, ankle angles and exoskeleton torques), clinical measures (e.g. muscle force and spasticity tests), and subjective measures such as user perspectives, attitudes towards technology and perceived workload. While subjective measures, such as level of confidence, are not typically included in biomechanical tests, they have been known to affect gait [53]. Prior to the experiment, controller gains for each test pilot were determined using a combined user-specified and walking performance metric, with separate gains for the left and right ankles as needed to reflect the test pilots’ residual function and subjective preferences. The process of gain selection was similar but more rigorous than for the LOPES experiments.

While it was not straightforward to fully compare biomechanical, clinical and subjective measures together, some interesting comparisons were made. Higher torque was delivered to the ankle with the lowest Manual Muscle Test score (test for muscle strength and function [54]). While this could be expected because higher gains were assigned to the more impaired ankle, controller inputs include ankle angles, and the test pilots’ own gait behaviour did not guarantee that higher torques would be delivered. Abnormal angle patterns were observed for the ankle with the lowest Manual Muscle Test score, with some improvement over the training period. With the neuromuscular controller, test pilots were also able to walk at a range of speeds (approx. between 0.28 m s−1 to 0.90 m s−1). Walking speed increased after training with the exoskeleton, and workload assessments suggested that the controller did not negatively impact users. A follow-up study that compared Achilles-aided training and conventional gait training without the device found similar benefits of faster walking speeds and reduced workload only when training with the neuromuscular controller [55].

In addition to the three gait studies discussed here, further studies with test pilots were conducted using the Symbitron wearable exoskeleton with full ankle, knee and hip actuation. Like its modular controller, the exoskeleton was designed to be modular as well and could be transformed into a knee–ankle configuration if users had adequate control of their hip joints [48]. With the knee–ankle configuration, we found a range of approximately 0.6 m s−1 in overground walking speed achieved by two subjects with incomplete injuries, with accompanying step length changes that remained mostly within the healthy region (WE-KA test pilots in figure 4, study details in [48]). Further investigations with this new exoskeleton is ongoing and the subject of future publications.

5. Discussion and future directions

Starting from a human biomechanics perspective, I have outlined some of the main challenges in applying bioinspired approaches to the design, control, and evaluation of exoskeleton devices for robotic gait assistance. I explored potential approaches to address these challenges in the context of one exoskeleton program, which sought to provide mobility for individuals with spinal cord injuries, as a case study. I discussed a bioinspired neuromuscular controller that was capable of generating healthy-like gait at various walking speeds for both test pilots and healthy subjects. In addition to allowing targeted assistance to produce assistive torques that strived to reflect the needs of the test pilots, the controller provided shared control by responding to the user’s joint kinematics and footfall patterns, enabling natural walking speed variations. Efficacy measures from healthy subject studies were difficult to apply to test pilots. Instead of measuring metabolic cost, comparisons of walking speed and gait behaviour before and after training with assistive devices or indirect assessments from speed-related changes were made. Biomechanical data in conjunction with clinical measures, both quantitative and subjective, could also provide a more holistic assessment of exoskeleton performance. Although only one case study was considered, the challenges and proposed solutions presented serve as representative examples of ongoing and future work in leveraging gait biomechanics to assist in the development of exoskeleton devices.

We tried to indirectly determine energetic economy through the speed–step length relationship, instead of directly measuring metabolic cost. Although only weak conclusions could be drawn, the approach allowed us to assess gait over numerous strides covering multiple walking speeds. Metabolic cost through oxygen uptake has been successfully measured in prior studies on test pilots with complete impairments but only after around 40 training sessions [56]. Metabolic expenditure has also been recently used as a performance metric for tuning exoskeleton control parameters. Known as human-in-the-loop or body-in-the-loop optimization, control parameters are adjusted based on online measurements of physiological signals to achieve optimal behaviour, such as minimum metabolic cost [57]. Metabolic cost, however, requires several minutes to reach steady-state behaviour. To test multiple control parameters and avoid lengthy measurements, estimated steady-state cost has been derived from transient behaviour. Successful tests have been performed on healthy subjects with ankle [57,58] and hip exoskeletons [59], where optimized control parameters have reduced metabolic cost. While decreasing the high energetic cost of impaired gait would be beneficial, it is uncertain if energy consumption is a major factor for individuals with gait impairments, and whether other factors, like stability, could represent competing interests.

Ambulatory individuals with impaired gait generally walk at very slow speeds (e.g. 0.27 m s−1 to 0.83 m s−1 for individuals with spinal cord injuries) with gait assistive devices operating at similar speeds [42]. While slow speeds may seem safer and more stable for exoskeleton users, there may be biomechanical consequences to slow walking. While our neuromuscular controller allowed users to walk at very slow speeds (0.6 m s−1 or less), its underlying neuromechanical model is unable to produce gait at those speeds. Without understanding healthy gait at very slow speeds, it was also difficult to determine if the gait patterns the controller produced were appropriate. We investigated the mechanics of very slow walking from healthy subjects and found that spatiotemporal, centre of mass work and joint work measures scaled at slow speeds in a similar manner to how they scale at faster speeds [60]. Discrete changes in behaviour that distinguish between slow speeds and normal walking speeds, like between walking and running, were not observed. As speed slowed, no increase in step width, step width variability or step length variability were observed, but we did find an increase in stance time variability, which could have implications for gait stability.

Balance control is another unresolved challenge that is currently mitigated in exoskeletons by the use of external aids, such as crutches, therapist assistance or overhead harnesses, to counteract small unexpected perturbations. To maintain stability, the main balance strategies include control of step placement, centre of pressure modulation and upper body control [61]. For walking, stepping is the primary means for stabilization, as it allows for the largest corrective maneuver [62,63]. At slow walking speeds for healthy subjects, we found a decrease in lateral stability, demonstrated by a reduced margin of stability, with the ankle and hip strategies compensating over stepping strategies [64]. These results imply that current exoskeleton capabilities should extend beyond step placement to include stabilization control with the ankle or upper body in the sagittal and frontal planes. Recent work on assistive devices include evaluating how balance control algorithms for the ankle [65,66] or for stepping [67] assist recovery after anterior–posterior perturbations during standing.

Enabling individuals with impaired gait to regain-independent mobility through powered assistive devices requires the integration of multiple perspectives and approaches. The Symbitron project discussed here exemplifies a collaborative effort among mechanical design, control, biomechanics and clinical researchers which endeavoured to address some of the difficulties facing robotic gait assistance. Assistive robots need to provide appropriate gait patterns while remaining adaptable to user movements and intentions. The understanding and utilization of the fundamental principles underlying gait behaviour may provide the key to enabling more natural locomotor behaviour and human–robot interactions. Combining interdisciplinary knowledge, techniques and tools may lead to more inclusive metrics that reflect the intricate facets of human gait to evaluate exoskeleton efficacy. While numerous challenges still remain, the integration of diverse perspectives in the search for engineering solutions to the complex challenges of gait assistance offers promising, if not vital, potential towards the development of robotic technologies to afford independent mobility for all.


No new human studies were performed. Each study discussed was previously published with its own ethics approval.


Leave a Reply

Your email address will not be published. Required fields are marked *

You May Also Like

Impact of technology on Finance industry

Financial services companies are looking to improve efficacy, service speed, and customer…

Upstream Oil and Gas Technology Innovations

Upstream Oil and Gas Technology is a spic and span, an online-just…

The Ever-Evolving Landscape of Oil and Gas

The oil and gas industry, a cornerstone of the global economy for…

Transforming Education: The Digital Revolution: 2024

In an era marked by technological advancements, the education industry stands at…