A fundamental component of synchrony between actors is an individual’s adaptation in response to a partner’s behavior (Burgoon et al., 1993). Burgoon et al. (1995) argued that an actor’s tendencies to either reciprocate or compensate are affected by the individual’s needs, expectations, and preferences, called their interaction position (IP) in IAT (Burgoon et al., 1995; Burgoon et al., 2017). An interactant’s IP is comprised of three pre-interaction factors. The first is an individual’s requirements, or what they need during the interaction such as basic needs like comfort, or physical proximity (Toma, 2014; Burgoon et al., 2017). Second is the individual’s expectations about their communication partner and the exchange, for example, the verbal and nonverbal communication patterns that people anticipate from others based on what is typical for a given context (Burgoon, 2014). Thirdly, the individual has desires which are the communicator’s goals for the interaction (Burgoon et al., 1995; Burgoon, 2014; Burgoon et al., 2017). If the actor’s IP matches that of the receiver, IAT predicts the receiver will match, or reciprocate, the actor’s behaviors (Toma, 2014; Burgoon et al., 2017). This adaptation happens throughout the course of an interaction, and while there is an implication of intent when observing these interdependent patterns, Burgoon et al. (2017) point out that adjustments can be made automatically. In other words, reciprocity and compensation can be both conscious and subconscious (Burgoon et al., 1993; Guerrero and Burgoon, 1996).

Finally, another factor to consider in the adaptation process is the actor’s perception of the difference between their expectations and the actuality of their partner’s behaviors (Toma, 2014; Burgoon et al., 2017). Violations of expectancies are generally considered negative unless the person performing the violation is particularly rewarding (Burgoon et al., 1995). If their partner’s behavior is considered more positive than the actor expected, the individual will reciprocate with similar behavior. The difference between expectations and actual behaviors, (e.g., under- or over-reliance on automation), can have implications not only for adaptation, but also other factors related to collaboration, including trust and team performance (Lee and See, 2004; Nikolaidis et al., 2016).

It is not always a given that humans will synchronize with another individual as there are several barriers that may prevent it. As synchrony can be either subconscious or unconscious, so too can the barriers to synchrony be intentional or not (Bernieri and Rosenthal, 1991; Burgoon et al., 1995; Fujiwara et al., 2022). Such challenges include factors related to intergroup differences, lack of understanding or comprehension between the interacting individuals, dislike or mistrust of one individual by the other, and contrasting goals or expectations for the interaction, (i.e., misaligned IP; Bernieri and Rosenthal, 1991; Burgoon et al., 1995).

Early evidence on human-machine synchrony indicates that barriers may be similar to human-human interactions (Lorenz et al., 2013). Kirkwood et al. (2021) identified several potential barriers to human-exoskeleton synchrony. They point out that as the exoskeleton is a wearable technology, users could have physical and psychological reactions to the machine on their body—like an invasion of personal space. Additionally, factors such as identification, understanding of the technology, perceptions and attitudes toward new technologies, and cultural norms in the work environment could all influence a user’s synchrony with the exoskeleton (Kirkwood et al., 2021).

Synchrony is particularly important for humans working with powered exoskeletons as the human and machine need to coordinate to become one entity (Kirkwood et al., 2021). While some dyads may be able to synchronize in one or two subcomponents, the challenge for synchronizing with a wearable robot, is that must be achieved across all three domains: body movements, rhythm, and timing. The interface between a human and exoskeleton is physical and influenced by the movements of the wearer and exoskeleton. Rhythm and timing are key to ensuring that the human and machine are not “opposing” each other, or acting out of phase, and can coordinate effectively to achieve the task shared-goal. Sensors in the suit can detect motion and send information to the robot controller, which then triggers a mechanical response in the machine to anticipate and match that of the human motion (Ronsse et al., 2011). However, this technology is still developing and there are many challenges to creating full-body, wearable robots, namely, humans vary not only physically, such as in height, strength, range of motion, etc., but also in learning and adaptability (Ronsse et al., 2011; Sposito et al., 2020). Since the action of one agent (human or wearable robot) can directly influence the action of the other, and the relative mechanical power of the robot (compared to the human) can be adjustable, sensemaking and agency are identified as two essential factors affecting synchrony in this context, and described further.

Sensemaking is an important concept related to adaptation and synchrony, as experiencing a new technology or feature can produce uncertainty and trigger sensemaking in order for the user to adapt (Griffith, 1999; Henfridsson, 1999). According to Weick et al. (2005), Sensemaking is a continuous, retrospective process in which circumstances are interpreted, and then action is taken based on that interpretation. The underlying assumption is that most people are operating within a mental script—not directly thinking about actions or actively interpreting meaning until something comes along out of the ordinary (Weick, 1993; Weick et al., 2005). When faced with a novel situation, individuals often use previous experience to interpret and react. For example, Griffith (1999) explored how members of an organization made sense of a new information technology (IT). During the initial release phase, members experienced high levels of ambiguity as to the meaning of the new technology in their work context, (e.g., role changes). Over time, as workers became more experienced with the technology, they used sensemaking strategies to adapt and create new organizational routines.

Within the domain of HRI, the concept of sensemaking has been utilized in two significant ways: First, from a robot interface-design perspective, there is a rich tradition within the human factors community of exploring how humans form mental models (internal representations or frameworks to understand and predict phenomena or action-consequences) of intelligent robots, and the factors that influence mental model formation (e.g., Kiesler and Goetz, 2002; Phillips et al., 2011; Tabrez et al., 2020). This work is very similar in conceptualization to sensemaking, in that humans are known to initially form primitive and incomplete, but functional, mental models of unfamiliar technologies that will continually improve with increased interaction and adaptations (Norman, 1983). This body of work also highlights that a robot’s physical appearance and language abilities may significantly influence the accuracy of mental models (Lee et al., 2005; Walters et al., 2008) leading to a human’s under- or over-estimation of robot capabilities. In the case of wearable robots, while the robot is not being seen by individuals during the interactions (as they are being worn), whether humans still anthropomorphize the robot/exoskeleton due to its intelligence and adaptive controls, and what design features affect sensemaking and mental model formation are important and intriguing questions to address.

Second, from a robot algorithm-design perspective, explainable AI techniques have focused on describing agent reward functions effectively, to enhance human-robot collaboration and task performance (e.g., Sanneman and Shah, 2022). These have been studied from the perspective of enhancing a human’s situational awareness (e.g., Sanneman and Shah, 2022) while not excessively increasing their mental workload; and included exploring the effects of different information types (Amir and Amir, 2018; Huang et al., 2019) and levels of information complexity Sanneman and Shah, 2022). While this body of work can be applied to the human-wearable robot context as such, it primarily stems from controlling highly complex industrial robots. Whether any such overt or explicit “explanation” of robot behaviors are necessary for sensemaking during training or operations of exoskeletons, is currently unknown.

In general, scholarship has broadly defined the term agency as the “capacity to reflect, adapt, and act” (Gibbs et al., 2021, p. 156). While this definition implies actual capability, it is important to recognize that humans are not always aware of the bounds of their agency. Thus, we incorporate a perception-based alternative sense of agency which is considered as, “the feeling that one is in control over one’s actions and their consequences” (Ciardo et al., 2020, p. 2). Additionally, individuals not only perceive their own agency, but also will attribute agency to others which influences outcomes such as perspective taking (Ciardo et al., 2020). This sense of agency is critical for individuals to understand relationships with their environment and others around them, and in turn, it impacts interactions with another individuals (Kawabe, 2013; Lorenz et al., 2013; Fujiwara and Daibo, 2016). The increasing ability of intelligent machines to both appear humanlike while taking on human roles, such as decision-making, has brought the question of agency front and center to discussions on human-machine interaction (Edwards and Edwards, 2017; Sundar, 2020). Rammert (2008) argued that machine agency lies on a continuum of passive to active technologies and depends on the kinds of autonomous actions the human versus the machine can make. At the lower end of the spectrum are passive tools such as hammers that are operated completely by a human user are less agentic; then semi-active machines that are able to have some self-driven actions after human input, such as a standard vacuum; reactive machines such as those with sensors like washing machines or thermostat-based climate control; proactive machines that self-activate such as a simple customer service chatbot; and cooperative machines such as mobile robots or smart homes that may have equal or more agency than a human user (Rammert, 2008; Sundar, 2020). For users, their sense of agency and their attribution of agency to a machine counterpart can have impacts on decision-making and performance outcomes (Ciardo et al., 2020; Kirkwood et al., 2021). For example, Ciardo et al. (2020) found that humans experience a reduction in their own sense of agency when interacting with robots they perceived as having more agency than passive mechanical devices (such as air pumps).

Previous scholarship on humans and technologies have taken either an anthropocentric or a technocentric perspective, primarily focusing on either the human agent or the machine agent as the source of change or site of impact. Wearable technologies, however, provide a unique opportunity for a middle-ground perspective that looks at both humans and machines as mutually adapting influential agents (Guzman, 2020). To our knowledge, there is very little research exploring agency in the context of collaboration and adaptation between humans and wearable robots. As exoskeletons are a relatively new technology, we are interested in how users make sense of their interactions with the machines when exposed to a complex full-body, powered exoskeleton, and explore the concepts of agency and adaptation as they relate to human-machine synchrony. Thus, we pose the following research question:

RQ: How do the concepts of agency and adaptation manifest in human-exoskeleton synchrony, particularly as human users make sense of operating a complex whole-body powered exoskeleton?

To answer this question, we conducted qualitative interviews with novice and experienced users of an exoskeleton prototype, to gain a deeper understanding of their experiences.

This study used a convenience sample of 13 healthy male participants (5 experts, referred to below as “EXP#,” 8 novices, referred to below as “N#S#” to indicate session number as novices participated in multiple sessions whereas experts only participated in one session). None of the novices had used the exoskeleton prior to the study, while the “experts” were engineers with extensive experience in testing and operating the exoskeleton through its developmental phases (>3 months). Respective mean (SD) stature, body mass, and age were 1.8 (0.04) m, 84.4 (6.8) kg, and 36.8 (15.4) years for the novices; 1.8 (0.03) m, 83.9 (8.2) kg, and 31.2 (7.8) years for the experts. All participants performed short bouts (2-3 trials) of level walking, stepping up over a small obstacle, push/pull, load carriage, stationary load handling, and reach and point tasks. Data from the multiple sessions of the novices and the experts were pooled together and analyzed, to draw out the thematic elements. We did not use the data to try to differentiate between experts and novices or between the multiple sessions of the novices, owing to differences in sample sizes and imbalance in the data. Hence, the question addressed by the paper is what a user’s experience of the exoskeleton is, irrespective of whether they were a novice user, or had had some time to get used to the device. Prior to any data collection, written informed consent was obtained from all participants following procedures approved by the lead University’s Institutional Review Board (IRB).

For this study, we used an early research-prototype version of the Guardian^® XO^® (EXO for brevity) developed by Sarcos Robotics, which is specifically designed for occupational applications (Figure 1). The EXO has a mass of 158 kg and includes 18 active degrees of freedom (DOFs) spanning the shoulders (flexion/extension, abduction/adduction, and internal/external rotation), elbows (flexion/extension), trunk (axial rotation and lateral bending), hips (flexion/extension, abduction/adduction, and axial rotation), and knees (flexion/extension). At the time of assessment, the EXO was still in the development stage, and the calf and ankle joints were not actuated. The EXO uses a patented “Get-Out-Of-The-Way” control scheme with torque sensors at the major body joints that allow the EXO to follow the human movement and amplify human joint torques (Jacobsen et al., 2014). With this technology, users can be aided to freely lift and handle loads up to 90 kg. To understand user movement intent, user input is obtained from embedded 6-DOF force-moment load cells located at the hands, feet, torso, and pelvis locations of the EXO. The EXO also has several tunable parameters, including virtual center of mass, gravity compensation, and actuation gains (magnitude of torque amplification).

Expert participants participated in one 2.5-h session and novice participants performed five 2.5-h sessions of simulated tasks, namely, walking, lifting, and transporting objects, pushing a cart, force-reproduction, and repetitive target-tapping while wearing the exoskeleton (See Park et al., 2023 for a more detailed explanation of study design and tasks). The sessions took place on separate days and the multiple sessions spanned across two to 3 weeks. Semi-structured interviews were conducted after each session.

After each study session, participants accompanied one of the experimenters to a separate location, where the interviews took place. Care was taken to maximize participant comfort and encourage unbiased responses, and their responses have been anonymized in this report. Interviews began with general questions, e.g., “how did today’s session go?,” followed by open-ended questions about participant comfort and confidence in using the exoskeleton, the physical- and cognitive strategies that participants used to learn and operate the exoskeleton, and their general perceptions of safety (Supplementary Appendix SA). Questions were similar between successive interview sessions, with minor modifications to suit the specific physical tasks performed on each day. Upon completing the interview, participants were asked for any final comments, and subsequently thanked for their time and escorted out of the interview location. Each interview typically lasted ∼30–40 min. The interviews were then transcribed for analysis.

In this project, researchers used an iterative approach where prior research on human-machine synchrony and sensemaking guided the coding process while also looking for emergent codes and themes (Tracy, 2020). After interview data were transcribed, the transcripts were uploaded in the computer-assisted qualitative data analysis software Atlas.ti for coding and analysis. After reading each transcript, researchers conducted line by line analysis of in vivo coding (Tracy, 2020), which consists of using the participants’ own language to create codes, to discover how participants achieved synchrony while wearing the exoskeleton and how they made sense of those experiences (Strauss and Corbin, 1990; Lindlof and Taylor, 2019). After in vivo coding, similar codes were grouped together and the constant comparative method was used to conduct as many rounds of axial coding as possible, or grouping codes into categories and subcategories, that were necessary to identify emergent themes (Lindlof and Taylor, 2019). After two rounds of axial coding, the emergent themes were identified as findings.

We found that participants faced a variety of challenges that required adaptation in order to synchronize with the exoskeleton and perform the given tasks. Challenges of synchronizing with the technology fell into three major themes: physical, cognitive, and affective. Physical challenges included fatigue, tension, jerky movements, and limited mobility in the exoskeleton. Affective challenges included expectancy violations while wearing the suit, uncertainty of how the suit was moving while using it, and feelings of fear and intimidation. Lastly, cognitive challenges included the nuanced ways that concentration and mental energy impacted suit use. We found that participants developed and used strategies for synchronizing with the exoskeleton and performing the given tasks that were thematically related to the respective physical, cognitive, or affective challenges. Thus, there is a relationship between the barriers faced and the adaptations needed to address those barriers.

The second major theme that was revealed in our analysis was how participants perceived their own agency as relative to that of the exoskeleton. Within the topic of agency, we found two subthemes: participants engaged in sensemaking to understand the exoskeleton technology; and participants had a preconceived idea of interdependence and who should be in control-the human or the exoskeleton.

There are several practical implications for designers, managers, and users of powered exoskeletons. The sense of agency a human-user has with an intelligent machine, may guide practical decision-making in many situations: for example, i) if a performance error is anticipated, does the human user take ownership of the situation to correct their inputs to the machine or wait for the machine to sense/adaptively correct its outputs; and ii) if human expectations are violated in terms of task performance strategies by the machine, do they trust that the system knows better and acquiesce, or fight to regain control. These are issues that are likely to be determined by what agency the human user attributes to themselves and the machine. Knowing how the human-machine partnership impacts the human’s understanding of their own agency is thus critical for safety and performance when deployed in a work environment. If employees feel they are not fully in control of their own actions, or there is uncertainty regarding which entity is responsible for specific decisions, this could lead to challenges in operating the exoskeleton and an increased likelihood for accidents and further liability concerns. Additionally, reductions in agency can have a negative impact on employee’s job satisfaction (Feldman et al., 2018).

From a designer’s perspective, the ideal human-machine partnership in different scenarios and the agency human users “should” develop with the machine, should be carefully considered and communicated to the users of such technologies, along with training about the machine’s capabilities and limitations in a variety of scenarios. It may also be wise to consider how diverse users may make sense of the technology. Exoskeleton designers/control engineers can then guide sensemaking by how they introduce and train users to elicit more positive responses and help users maintain a higher sense of agency and reduce anxiety. For example, designers can ensure the users understand that while the suit itself may be more physically capable, the human is still in control. Lastly, designers can offer users a basic understanding of how the exoskeleton works and what it does from a mechanical and controls standpoint. In other words, making the user aware of the suit’s state and intention. Comparably, Stirling et al. (2020) suggested that an exoskeleton should enable the user to maintain sufficient cognitive abilities for processing task- and stimulus-related information while using the exoskeleton. When designing control interfaces, researchers need to not only consider what is the most efficient interaction between human and machine, but also what allows the human to maintain the highest sense of agency. This can be achieved by either creating a set of initial interactions that can allow a user to actively explore agency using their own sensemaking strategies, or intentionally program a series of technology interactions that can help create the expected sense of agency.

Our findings also reveal that although participants in our study did attempt to synchronize with the machine in order to accomplish tasks, the cost/benefit ratio, and the need for human adaptation is a barrier to synchrony and may reduce human trust, technology adoption and human-machine performance. It is also currently unclear as to how long participants would have continued to make adjustments to their natural body movements and their mental models of the interaction and expectations from it, to create a sufficient level of synchrony that can sustain the human-machine partnership. Previous studies have shown that individuals have a range of desire and ability to adapt to a robot counterpart (Nikolaidis et al., 2016). As such, it is possible that some users of exoskeleton technologies will be unwilling or unable to adapt to the machine’s movements or experience more challenges when attempting to adapt to this new technology.