Firstly, the wearable has to be easy for a user to don and doff. In tendon-based wearables, since the wire routing has to be precise, multiple straps and anchor points are needed to keep the tendon sheaths in place. Pneumatic and hydraulic actuators, on the other hand, can be imprecisely placed on the wearable. Nonetheless, none of these wearables enable users to wear the device independently; healthy individuals generally have no problem doing so, but impaired users will require assistance. Moreover, the benefits in muscular or metabolic activity of healthy individuals are rather minuscule considering the hassle needed to put on an assistive wearable, achieving a reduction of 15% at best [60,64], with many others below 10% only [38,47,53,69,73,88]. If more assistance is needed, external structures should be introduced to reroute the tendon above the skin so that the effective moment arm applied over a joint can be increased. An alternative is to use a stronger motor to generate greater tension in the cables. Either way, the wearable will get bulkier and require design changes to allow users to don it independently.

Secondly, the wearable needs to be portable. Wearability and portability are two distinct features. The wearable component may be low-profile, lightweight, and easy to don, but the control system may not be portable. For example, some cable-driven wearables have the motor located externally [78,103], with the control system being wheeled around the user. Pneumatic- and hydraulic-based wearables also need to be tethered to a compressor or a tank, which limits their portability and applicability to rehabilitative settings. Unless the entire package can be made smaller and lighter, users will be less willing to use the device on a daily basis, as they will have difficulty transporting it while carrying out their daily tasks [124].

Thirdly, the wearable should be practical for its intended use while not interfering with other daily activities. Despite the remarkable experimental results in a controlled environment, some reported questionable outcomes in real-life situations. For instance, Shi et al. [48] developed a cable-driven wearable that assists knee motion and has a unique energy harvesting function during walking. However, the cadence is limited to a maximum of 2 km/h and may not cater to the average person’s walking speed. In another work, Hennig et al. [65] developed an assistive hand exoskeleton with an intuitive EMG control scheme that had an activation accuracy of 94.8%. However, the hand closure was slow at 1.2 s, making it more suitable for rehabilitative training instead of assisting with activities of daily living (ADLs). Schmidt et al. [96] devised a cable-driven suit that reduces up to 30% of muscular activity in the hips and knees during sit-to-stand transitions. Unfortunately, the suit was heavy, with two actuator units each weighing 1 kg on each leg, making other ADLs strenuous. Similarly, Zhang et al. [121] developed a vacuum-actuated rotary actuator that assists knee flexion during walking, but the wearable was bulky and hindered activities such as sitting at a desk for an extended period of time. In light of the given examples, although the actuation mechanism is sound and the experimental results are affirmative in terms of assisting strenuous tasks, the caveats in the design would render the wearable impractical for real-life usage.

Cable-driven wearables are made from off-the-shelf metal wires and electric motors. Similarly, SMA-based wearables are fabricated using commercially available SMA wires. Other types of soft wearables, however, are made of unusual materials. Pneumatically- and hydraulically-driven, PVC-based wearables primarily consist of elastomeric actuators that mimic living tissues. These elastomers are blended with other polymers in precise ratios so that their stiffness can be adjusted. In some cases, composite materials are used to adjust stiffness. For instance, Polygerinos et al. [139] described their work on a hand-wearable that used an elastomeric tubular construct reinforced with anisotropic fibers to program motions such as bending, extending, extending-twisting, and bending-twisting.

The fabrication process is multiphasic and requires manual winding of the fiber around a cured elastomer. Such manual processes make it difficult to manufacture soft actuators at scale, as they are labor-intensive and costly. Furthermore, substitutes for elastomers may not be commercially available. In the event that the substitute does exist, its material properties and mechanical behavior would be different from those created manually in the laboratory, compromising the performance of the wearable. Moreover, should the soft wearable be embedded with sensors [47,75,101,118,120], the challenge in material selection and the complexity of manufacturing would increase accordingly. That said, it is paramount that materials used to fabricate soft wearables be obtainable on the market. If commercial equivalents are unavailable or specially treated materials are indispensable to preserve the niche functions, the materials have to make economic sense [146]. Otherwise, the soft wearable may be inadequate for mass production and lose its marketability. Furthermore, the difficulty in procuring these basic materials may complicate the replacement and maintenance of the parts.

3D printing has emerged as a viable option to fabricate complex elastomers. Yeow’s research group has done much work on developing 3D printed soft actuators integrated into hand [126], wrist [122] and elbow [114] soft wearables. The filaments used are commercially available, ranging from 60 A to 85 A in Shore hardness. Fabrication is automated, and since the actuators are identical to one another, performance is more consistent than if they were handmade. However, even though the actuator design remains unchanged, printing settings tend to vary among consumer-grade printers. Although an industrial-grade 3D printer would provide more accurate and reliable calibration, it may not be available to most laboratories. Material choices are also limited, with the softest material available at 60 A shore hardness. Some soft wearables are silicone-based and possess much lower shore hardness, making them more capable due to their larger strain response. As such, more commercial 3D printing mechanisms and filament options need to be established before they can be adopted as a mainstream option for fabricating diverse elastomer-based wearables.

To date, few studies have been conducted to investigate the durability of soft wearables and their actuators. Technically, cable-driven wearables are as durable as their motors, whose actuation cycles are rated in the millions. However, other failure modes. such as wear and tear of the textile and anchor points, have yet to be considered and investigated [147]. A couple of studies have shown that elastomeric actuators are prone to fatigue due to the large stress and strain applied cyclically [148,149]. In the case of SMA-based wearables, rapid heating and cooling in an electric field can result in rapid breakdown of the wires. While failure modes and methods to extend the longevity of SMA wires have been reported [150,151], durability studies of SMA-based wearables should be carried out more extensively [141,142,143].

Soft wearables are structurally vulnerable compared with rigid ones, so regular maintenance is needed to maintain their performance. The frequency of maintenance and replacement needs to be optimized to keep operational costs low. Aside from using more durable materials, fatigue optimization should be done during the design process to ensure the longevity of these actuators [152]. Results from fatigue studies and durability tests can provide useful information on how to preempt system failure and reduce operational downtime.

Efficient control of soft wearables is vital in operation because it directly affects user-machine interaction. Basic control schemes include the use of buttons [84,134], a smartphone [51], or a joystick [80] to conduct preprogrammed actions. Such control methodologies are mostly unintuitive and require a free hand to operate. These soft wearables are therefore more suited for rehabilitative applications than assistive ones that prioritize task efficiency. Rehabilitative exercises can be carried out repetitively without the constraints of time and portability under these touch-based controls. These preprogrammed actions are simple and have a predetermined path, which does not require the precise modeling of a soft actuator.

More intuitive and complex control algorithms involve event-driven torque compensation or intent detection. For example, Cao et al. [47] recently presented a hardware circuit design that utilizes sensor fusion to allow the wearable to decipher motion intention and perform power-assisted control in a lower limb exoskeleton. The control system uses a combination of EMG sensors, inertial measurement units mounted on the lower limb, and force-resistive sensors embedded in the insoles to detect gait phases. Power delivery is efficient with a 1% delay time. Similarly, Natali et al. [75] used embedded force sensors in the insoles of an exosuit, XoSoft, to detect gait phases and assist hip and knee movements simultaneously. The resultant assistance was approximately 10% for hip actuation and 9% for knee actuation.

Upper-limb rehabilitation can also be improved by involving EMG sensors for intent detection during mirror therapy [62,120,129]. One comparative study [76] showed that, by using a neural network-based strategy, a difference of more than 50% in torque compensation could be observed as opposed to several other control methods. Therefore, applying forces inappropriately will cause premature muscular contraction that eventually leads to increased metabolic expenditure. That said, further research should be directed towards determining the most effective control strategy for a particular soft wearable.

However, difficulty arises due to the nonlinear behavior of elastomeric actuators, which is hard to model. Precise control along the actuator’s continuum may not be achievable in the control of these wearables. Researchers thus simplify their control by using one input to drive multiple DOFs in one or more joints [110,111,113,115,117,120,122,124,125,126,127,128,129,131,132,134,135,136,138,139]. In addition to modeling the behavior of the wearable, it should be underscored that human tissues are highly deformable and the interface needs to better conform to the skin [46] to better correlate with the motion of the body.

Artificial Intelligence (AI) has been a hot research topic in recent years. It is a broad term that generally refers to the ability of a machine or computer to perform tasks that require human intelligence. Based on the results of this systematic review, AI capabilities are scarcely embedded in software wearables. Most of the contemporary software wearables are not equipped with the appropriate sensors, computing power for AI augmentation, or intuitive control algorithms. Some effort has been made towards developing soft sensors, but none has been actively applied to a soft wearable with the aim of imbuing it with AI capabilities. In an article by Wang et al. [153], several prevalent sensors (i.e., resistive, piezoresistive, capacitive, optical, magnetic, and inductive sensors) have been shown to be promising for inventing sensorized soft robots. Despite such developments in soft robotics proprioception, it is recognized that not much work is being done on interpreting the data from these sensing systems. Advanced algorithms and new frameworks are needed to interpret the raw data collected and construct sensible information.

The importance of AI augmentation is obvious when discussing the ability to conduct data analysis for performance tracking. These tracking data are helpful for collecting clinical feedback and informatics firsthand. More advanced algorithms may make it feasible for the wearable to adjust the workout intensity according to a patient’s progress, thus allowing for a bespoke training regimen to accelerate the patient’s functional recovery. Secondly, at the forefront of assistive wearables, AI can reflect the state and monitor the performance of the actuators [154]. Repair and maintenance can then be prompted automatically upon detecting a malfunction.

It is noteworthy that the few soft wearables programmed with algorithms [76,82,83,84,95] are not portable, as they often require a tether to an external computing source. Given that portability of the overall system is one of the technical challenges of AI augmentation of soft wearables, wireless data processing should be taken into account.

After the soft actuators have been fabricated and integrated into wearables, they undergo a series of experiments and tests to demonstrate their actual performance. One metric involves assessing the ability of the soft wearable to replicate or assist natural biomechanical motions. To evaluate this, a kinematic analysis of joint movements is commonly conducted. Another method involves estimating or measuring the assistive torque and force outputs from the wearable. However, none of the reviewed articles cover a comprehensive list of experiments; some have been done while others have been arbitrarily omitted. For instance, a study by Otálora et al. [45] used only gait timings to assess the biomechanics of lower limb motion but did not evaluate gait patterns or the assistive torque rendered by the use of the wearable. On the other hand, Natali et al. [116] not only tracked the joint angle throughout a gait cycle but also recorded joint moments. Another widely accepted metric for assessing a soft wearable involves measuring muscular or metabolic activity, which was used as the primary indicator in the evaluation of the wearable [45,60]. Various research groups have used these values to claim superiority over one another. As mentioned earlier, researchers have reported reductions of up to 15% in metabolic or muscular activity [38,47,53,60,64,69,73,88] by applying their soft wearables.

However, the discrepancy in sample size, demographics of human subjects, experimental protocols, and metrics for assessment may render these claims fallible. An important metric, net metabolic savings, is rarely reported. For example, emphasis is placed on the difference in metabolic activity between the powered-on state and the powered-off state when a wearable is already worn during an exercise session. Since wearing a soft wearable and keeping it on would inevitably raise metabolic activity, net metabolic activity should be calculated without wearing the device as well. Asbeck et al. [38] found that users carrying their backpack-sized control system, which weighed a total of 10.1 kg, experienced an increase in metabolism from 16% to 17.5%. Additional steps were needed thereafter to optimize the weight distribution of the system on the body, eventually attaining an average metabolic reduction of 6.4%.

A reduction in metabolic or muscular activity undoubtedly implies effective assistive force provided by a soft wearable. Nevertheless, the distinction has not yet been made between the two, and studies have arbitrarily chosen one over the other to gauge the performance of a soft wearable. Chen et al. [63] measured both metabolic and muscular activity to substantiate the effectiveness of their wearables in assisting hip flexion and claimed that it lowered metabolic consumption by 11.52% when walking on the treadmill at 5 km per hour. Other studies excluded these tests entirely in their evaluation of the wearables [127,128,139,142]. As there is currently no testing methodology that can be used universally, academia and industry should come to an agreement to ensure fair assessments and comparisons of soft wearable technologies.

The adoption rate of any technology is greatly influenced by public perception. This ultimately drives demand for a product. The same rule applies to soft wearables, which are a relatively novel technology. Despite its recent emergence, little effort has been made thus far to improve the public’s perception of soft wearables. To this end, some recommendations can be made as follows:

Although the experimental results described in these works are commendable, the values reported are irrelevant to the user. These have to be translated into tangible and economic benefits in terms of saved man-hours, reduced risk when completing strenuous tasks [155], and positive clinical evaluation [156], for example. As mentioned in Section 4.1.1, the current assessment of assistive soft wearables is discrete and specific to an intended use. Moreover, most studies have been conducted at the proof-of-concept stage thus far, which may not provide strong evidence to change clinical practice. Undeniably, the paucity of high-profile large-scale clinical trials adopting universal evaluation standards has hampered building confidence in device performance and clinical benefits.

Additionally, more thought should be put into the design to integrate the wearable into daily life without hindering movement, as user-centric design approaches can foster the adoption of soft wearables [157]. Moreover, global sentiment towards AI augmentation remains largely positive. An effort to incorporate smart sensors is therefore required, despite the technical challenges. When developed and used responsibly, such technologies will appear futuristic and have the potential to make a significant impact on society [158].

With the bulkiness of some soft wearable designs, it is noteworthy that little feedback has been collected regarding comfort and ease of use. Only a few studies have been done regarding optimization of the interface to improve comfort, alleviate unnecessary stresses on the body [159], and reinforce functionality [160,161]. In some wearable designs, there is a trade-off between functionality and comfort, and a balance should be sought. More work needs to be done on pneumatic and hydraulic systems, in particular, to upgrade the portability of control systems by reducing their size and the weight of the power source [162].

Although the façade of a soft wearable may seem unessential to its adoption, social acceptance is an important aspect of consumer behavior [163]. The aesthetic elements of soft wearables have to reflect psychological and social factors [164]—wearable soft robots need to be unnoticeable and unobtrusive when worn [165]. Especially for disabled individuals, successful adoption of an assistive wearable requires them to explore the meaning of these devices in their daily lives, tailor their expectations of the technology, weigh the social costs, and adapt to their disability [166]. Therefore, if these requirements are not met, users are more likely to discard the device, regardless of its technical capabilities.