The different backgrounds of the respondents hamper the achievement of specific conclusions and observations for a single device category. Respondents may have referred to one specific device but also to exoskeletons in general. We can still suppose that the result is an average, general estimate in the field, which cannot apply to any device and we shall avoid misunderstandings that might lead to read the results as a general RA. However, the result of this survey also represents a picture collected by an audience of real users operating outside the laboratory or clinical conditions. Such a result can favor a more concrete view of how exoskeletons are perceived, not only limited to research and scientific literature. The items presented as a list of hazards that usually/typically apply to exoskeletons can only be taken as a reference for RAs. Frequency and severity feedback shall be read with the knowledge of the variety of devices and conditions considered by the respondents, and cannot be related to an accepted frequency/severity classification of a specific RA.

The frequency analysis presented low variability among the responses. All the items, except misalignments, received a “rare” occurrence score from at least the 50% of the participants while “recurrent” was less than 10% of the answers. Severity evaluation also did not show significant trends with the consequences of the events averagely rated as “minor”. Misalignments were rated as “recurrent” by 25% of the respondents. A recurrent event was proposed as something happening from once a day to several times per session, meaning that misalignments are often daily problems for the users. Our definition for misalignments was “an offset between the human and the device joint”. This event is impossible to avoid considering that robot kinematics is just an approximation of the human body. Conversely, misalignments also received a 24% of “rare” occurrence rate. Misalignments might be considered negligible for some applications. One comment underlined how they were observed for paraplegic users (specifically taller users) and not for healthy users, suggesting that misalignments were noticed only for remarkable offsets. The controlled conditions from clinical environments might here influence the result since improper fitting and unprecise positioning are more easily avoided. None of the remaining events were frequently experienced with electrical malfunctioning collecting the lowest frequency score. This is understandable for devices in a commercial or pre-commercial stage since requirements for electrical safety are far clearer than for all the other event types. IEC-60601 is indeed very detailed on specific rules and safe limit values. This can perhaps be considered a simple field to comply with, in terms of knowing what can and needs to be done to make the system as (electrically) robust and safe as possible.

While the frequency evaluation might be based on a quantitative scale (one can record the occurrence rate of an event on a determined time unit), severity mostly relies on qualitative evaluations. The absence of clear and measurable criteria able to define severity makes the results more dependent on the specific device and application considered. Events leading to a severe injury are relatively rare in exoskeletons, for this reason, respondents might have less experience in rating severe AEs. In a recent review on exoskeleton risk management [16], two bone fractures were reported due to the occurrence of misalignments while the remaining listed events led to no injuries or skin damages that were resolved in a few days. Another review on AEs in stationary robots (e.g., Lokomat) collected 3 severe AEs out of 169 reported events although 43 remained unrated. However, the reported number of AEs could also be an underestimation, since reviews are limited to published reports. From the results, more than 10% of the participants experienced severe outcomes after unintended/unexpected motion (19%) and use error (12.7%). Unintended and unexpected motions can indeed lead to falls, having then higher hazardous potential compared to other hazards. The 19% of event’s consequences classified as severe can anyway raise a flag since “severe” was associated with incapacitating outcomes. From some collected comments, the participants pointed out how their severity score was relatively low because they were testing the device under the supervision of clinical staff, prevention measures, or under very limited and constrained conditions, which reduced the overall risk level. As we stated for misalignment, frequency scores are normally influenced by the presence of clinical staff supervision and controlled conditions. The fact that occurrence and severity are based on tests performed in a clinical environment may indeed affect the perception of the users in evaluating the events. Both severity and frequency can indeed be strongly affected by studies performed in presence of staff members monitoring the user and the system. What the community can communicate about their perceived experience in exoskeleton safety can be far from reality. This point is stressed by the poor knowledge that technical personnel may have about safety aspects. Additionally, safety tests without intervention are difficult to perform when they include humans. Part of the respondents might also have provided a more technical experience rather than a knowledge of medical or physical outcomes.

A lack of awareness from users and technicians might also explain the low severity score of the experienced outcomes. For example, the appearance of skin injuries can even occur days after exoskeleton use, and the user may not perceive the injury when doffing the device. In some cases, the supervisor might not pay attention to the harm the user is experiencing. Clinical studies normally focus on gait or task-related parameters without considering AEs, which can be of difficult detection. The combination of severity and frequency scores as indicated in Table 3 (and Fig. 4) produced a generally low relevance score for the proposed events. This can be interpreted as a matured confidence of the users towards these devices, but also as a lack of experience regarding the possible hazards and risks in these applications. Existing literature on exoskeleton safety is indeed much more limited when compared to the literature on electromechanical features, design, or control strategies. AEs are also poorly published, although investigators are obliged to list the ones occurring, with the obligation to take action in case of serious AEs. Further analysis of events’ causes showed some more recognizable trends. 60% of the participants identified electrical faults and loss of communications as causes for unintended shutdowns. However, electrical faults also received the lowest risk score in the evaluation, in contrast with the unintended shutdown risk results. Electrical faults were on their side mainly attributed to issues with cables and precarious connections, although the frequency of these events was one of the lowest recorded. As a consequence, unintended shutdowns also received a low-frequency score.

Skin damages were associated with mechanical contacts (73.8%) and misalignments (60%). This result shows how the design of the device plays a key role in defining safe shapes, surfaces, and attachment designs, which can otherwise lead to harmful and uncomfortable use. Misalignments can be related to the design phase of the device. Attachment strategies and materials directly influence the capability of the device to be well aligned and maintain the desired configuration. Misalignments were linked to three main aspects, namely cuff design, incorrect cuff positioning, and oversimplified kinematics. As previously said, the simplicity of the device could reduce the overall complexity but increase deviations from real human kinematics. Cuffs and interfaces shall be a developers’ priority to ensure safe human–machine contact and communication. However, incorrect cuff positioning could be also considered use error and not a design error (too high or low, rotated, etc.). More than 50% of the participants identified unintended triggers (exoskeleton incorrectly reacting to body movements) and sensor failure in reading as potential causes for unintended and unexpected motions, whereas still half of the participants also identified use errors as an important cause. If we analyze use errors, respondents highly agree on insufficient training as the major cause (67.7%). Training is one of the mitigation measures suggested by the FDA to decrease the level of risks. The other two major items for use errors are “wrong use of exoskeleton interface” and “wrong settings”, which might be also related to a lack of training and practice of the device, equally contributing to an AE occurrence. Together with training, user errors can be mitigated by improving usability testing where the user can provide useful feedback on the device’s design and user interface. From this point of view, pilot tests could be of great importance for a deep and practical knowledge of device safety.