Principles from human-human physical interaction may be necessary to design more intuitive and seamless robotic devices to aid human movement. Within a framework of interactive motor behaviors, haptic “collaboration” has been defined as when “both agents jointly try to develop a consensual solution to solve a problem” with symmetric behaviors of both agents seeking to reduce each other’s error and cost. Another class of joint behaviors is “cooperation,” which is an asymmetric relationship where “one agent focuses on itself and the other either obeys in the assistance or accepts to look for the other’s task in the education” (Jarrassé et al., 2012). A robotic device that aids a person’s walking falls under the class of “cooperation,” where the robot seeks either to assist or educate the human. Hand-operated robotic devices have recently been developed to aid walking similar to a powered cane (Suzuki et al., 2009; Di et al., 2016; Nakagawa et al., 2016; Lam and Fujimoto, 2019; Trujillo-León et al., 2020) or walker/rollator (see reviews in (Martins et al., 2012, 2015; Werner et al., 2016)). Robotic walking aids have been designed to aid balance through a variety of methods such as providing mechanical support during falls (Hirata et al., 2006; Suzuki et al., 2009; Mou et al., 2012; Geravand et al., 2015; Di et al., 2016; Lam and Fujimoto, 2019), preventing risky postures (Nakagawa et al., 2016), or providing a proprioceptive cue (Stramel et al., 2019), but it is unclear which strategies are most intuitive and beneficial to the user. Meanwhile, a person can quickly and intuitively aid another person’s balance during walking by holding their hand. Greater understanding of the mechanics of human-human balance aid may establish principles to guide improved control laws for robotic devices.

Previous studies on light touch measured forces at the hands to examine the how physical interactions aid balance. Early work on light touch, originally defined as forces <1 N, showed that postural sway during standing balance was reduced when a person used a fingertip to touch a stationary object; this reduction in postural sway was interpreted as improved balance control due to more sensitive sensory feedback for maintaining upright posture through light touch, and not due to mechanical support (Holden et al., 1994; Jeka and Lackner, 1994). Later studies showed that light touch between two persons can also decrease postural sway (Johannsen et al., 2009, 2012, 2017, 2018; Reynolds and Osler, 2014; Steinl et al., 2018). Few studies have examined light touch during walking, and then only between a human and an object or device. Some results show that light touch with an object/device can improve walking balance either on a treadmill or overground, as measured by reduced body sway (Dickstein and Laufer, 2004; Forero and Misiaszek, 2013), pelvic acceleration (Boonsinsukh et al., 2009), decreased step width (IJmker et al., 2015; Stramel et al., 2019), and lower margin of stability variance (Oates et al., 2017). However, we are not aware of any previous work examining whether light touch between two persons can also improve walking balance.

Perhaps engaging similar sensorimotor mechanisms as light touch, haptic communication has been used to explain the benefits of human-human hand interactions for performing constrained physical tasks, but mostly during upper-limb manipulation. Human dyads have been shown to perform better than the worse partner in reaching (Reed et al., 2006; Reed and Peshkin, 2008), object manipulation (van der Wel et al., 2011; Mojtahedi et al., 2017; Jensen et al., 2021), target tracking (Wegner and Zeaman, 1956; Glynn and Henning, 2000; Feth et al., 2009; Ganesh et al., 2014; Melendez-Calderon et al., 2015), and force tracking (Masumoto and Inui, 2013). In studies that measured each person’s force or torque contribution independently through coupled robotic devices, human partners have been shown to exert opposing forces or torques that cancel each other (Reed et al., 2006; van der Wel et al., 2011; Madan et al., 2015; Melendez-Calderon et al., 2015; Mojtahedi et al., 2017). These opposing forces/torques do not contribute to motion of the object being manipulated, and have been proposed to create a haptic communication channel that facilitates interpersonal collaboration (van der Wel et al., 2011; Madan et al., 2015; Jensen et al., 2021). As it consists primarily of information transfer (van der Wel et al., 2011), haptic communication may describe the same phenomenon as light touch.

Previous work measuring hand forces between two persons during walking provide an estimate of the range of forces involved in passive dynamics and haptic communication during walking. Small ranges of hand contact forces (2–4 N) have been measured between two persons walking side-by-side on separate treadmills (Sylos-Labini et al., 2018), a task that does not involve explicit movement goals, and thus the force magnitudes likely characterize dynamics of arm motion and the hand connection. Larger, but still small peak forces (10–30 N) were sufficient to convey movement goals and timing between partners with assigned roles of leader and follower in the absence of visual or audio feedback (Sawers et al., 2017). Similar peak force levels (up to 8.4 N) were measured for two persons walking and carrying a large object together (Jensen et al., 2021), another physical interaction task with a defined goal. However, no previous work has measured hand forces used to aid balance during walking.

A suitable paradigm for examining physical interactions that aid balance must appropriately challenge balance in order to elicit aid. As individuals without balance impairments are unlikely to experience significant balance challenge during overground walking, beam-walking has been used as an experimental task to challenge balance in unimpaired adults (Speers et al., 1998; Domingo and Ferris, 2009, 2010; Sipp et al., 2013; Sawers and Ting, 2015; Chiovetto et al., 2018). Furthermore, Sawers and Ting 2015 showed that the distance walked on medium and narrow-width beams can distinguish between expert, novice, and balance-impaired populations, leading to the use of this task as a clinical or laboratory tool for measuring dynamic balance. Here we take advantage of a challenging beam-walking task to evaluate whether and how hand forces between two persons aid balance.

The current study examines the effects of human-human physical interaction to aid balance in a beam-walking task. Pairs of unimpaired young adults held the ends of a custom force handle device with one hand while one person walked on a narrow balance beam and the other person walked overground by their side. We compared balance performance during partnered vs. solo beam-walking to examine the effects of haptic interaction, and we compared hand interaction mechanics during partnered beam-walking vs. overground walking to examine how the interaction aided balance. We calculated the torque components on the beam-walker’s body generated by hand forces and the interaction torque, overlapping and opposing torque that does not contribute to motion of the beam-walker’s body. Defining haptic communication as exchange of information through sensory feedback elicited from physical contact (i.e., touch), we hypothesize that balance is aided through haptic communication and predict that interaction torque is greater for partnered beam-walking than partnered overground walking. In order to gain insight into feasible controller designs to emulate human-human physical interactions for aiding walking balance, we further model the relationship between torques and the beam-walker’s angular sway as a second order system. Our approach examines physical partner interactions in a paradigm relevant to assistance and rehabilitation (i.e., the beam-walking task challenges balance in unimpaired young adults while overground walking challenges balance in impaired populations), and our results suggest haptic interaction parameters that can be implemented in robotic devices to aid balance.

This first study quantifying human-human physical interactions in a balance-challenging walking task relevant to physical assistance and rehabilitation suggests that hand interactions aid balance primarily through haptic communication. With a partner, participants were able to walk further on the beam without falling, reduce lateral sway, and decrease angular momentum in the frontal plane. Consistent with haptic communication, hand forces used to aid balance were small and created interaction torques on the beam-walker’s body about the beam axis. Balance improvements with a partner were correlated with the amount of interaction torque, which does not contribute to movement and likely provides haptic communication. Finally, the relationship between torques and motion of the beam-walker’s body can be represented by opposite types of mechanical elements (inertia, damping, and stiffness) that can be implemented in robot controls to aid walking balance.

Our approach examines partner interactions in a paradigm relevant to physical assistance and rehabilitation. The beam-walking task challenges balance in unimpaired young adults while persons with balance impairments may be challenged by overground walking. Notably, there was no correlation between improvement in balance performance and the beam-walker’s solo balance ability, demonstrating that physical interaction benefits participants of varying ability levels. Our results may be relevant to individuals with balance impairments, but the specific disorders may also impair sensorimotor function in different ways and thus the relationships between hand interactions and balance performance must be explicitly tested in impaired populations. For example, proprioceptive and cutaneous sensory acuity and muscular strength may affect the degree to which hand interactions are used for communication or mechanical support.

The small hand force magnitudes (mean of 2.2 N laterally and 3.4 N vertically) we observed during partnered beam-walking may reflect a combination of passive dynamics of moving limbs during walking, light touch, and haptic communication. Although previously a 1 N threshold was established for light touch during standing (Holden et al., 1994; Jeka and Lackner, 1994), there may be additional interaction forces during walking due to the passive dynamics of arm and hand motion without intentional communication or cooperation; hand forces in handholding during treadmill walking (i.e. no balance aid) were previously measured to be within a range of 2–4 N (Sylos-Labini et al., 2018). Therefore, balance improvement during partnered beam-walking as well as during interpersonal light touch in standing balance may arise from similar mechanisms, e.g., increased sensory feedback on the body’s spatial localization (Holden et al., 1994; Jeka and Lackner, 1994; Johannsen et al., 2016). Moreover, the hand forces we observed are slightly lower than during a partnered stepping task that required haptic communication; they found peak interaction forces of ∼7 N between novice partners during an unpredictable forward-backward stepping sequence (Sawers et al., 2017). Although it is arguable what exact amount of force is meaningful for mechanical support in our balance task, the fact that we measured hand forces during balance aid similar to that of handholding during walking without balance aid suggests that the forces primarily served sensory feedback rather than direct mechanical support, and thus can be considered both “light touch” and “haptic communication.”

The opposing torques created by hand forces further support that balance during partnered beam-walking is aided through haptic communication. Prior studies in dyadic object manipulation identified opposing forces that facilitated haptic communication (Reed et al., 2006; Reed and Peshkin, 2008; van der Wel et al., 2011; Sawers et al., 2017; Jensen et al., 2021). Similarly, during partnered beam-walking, we show that orthogonal components of the hand interaction forces cause opposing torques in the plane of body sway during beam-walking, consistent with the rotational dynamics of the body about the beam (Chiovetto et al., 2018). Although each of the torque components (Ty_Fx and Ty_Fz) was higher during partnered beam-walking than partnered overground walking, neither torque magnitude was correlated with balance improvement. There were also non-zero mean net torques that may reflect normal force supporting the weight of the beam-walker’s arm, light touch, and/or haptic communication, but these net torques were not correlated with balance improvement. In contrast, we found that greater interaction torque was correlated with greater reduction in body sway during partnered beam-walking. Our results are consistent with the finding that larger interaction forces result in better performance during a partnered walking task (Sawers et al., 2017). Because interaction torque creates no net torques on the body, it does not contribute directly to body sway, and likely improves balance through haptic communication. The bias we used to account for sensor drift did not qualitatively alter our results. We tested high and low extreme values of sensor drift, for which the difference in opposing torque magnitude between partnered overground vs. beam walking (Supplementary Figure S1A) and the correlation between balance improvement and interaction torque magnitude remain significant (Supplementary Figure S1B). Mass-spring-damper model estimates rely on the variance in the signal and would not be affected by the bias.

However, interaction torque could contribute to increased mechanical stability of the beam-walker’s body by increasing the net impedance in a manner similar to co-contraction (Reed and Peshkin, 2008; Melendez-Calderon et al., 2015). In our setup we were not able to explicitly decouple forces and torques used for communication vs task dynamics (impedance). This decoupling is very difficult to achieve in a walking task, but has been done in upper-limb studies with virtual objects that either isolated force feedback from the object vs. the partner (Groten et al., 2013; Roche and Saint-Bauzel, 2016) or used physical interaction to negotiate a joint non-motor decision (Pezzulo et al., 2019).

The human-human balance-aiding principle of opposing dynamics for lateral and vertical directions is applicable to many robotic walking aids and rehabilitation devices regardless of exact hardware design. Our modeling results suggest that balance aid from a human partner during walking can be represented by two mass-spring-damper systems with opposite types of elements; torques from lateral forces are consistent with active (i.e., energy-injecting) impedance while torques from vertical forces are consistent with passive (i.e., energy-dissipating) impedance. Lateral forces generally create torques acting in the same direction as the beam-walker’s body motion and may function both to communicate information haptically and mechanically alter balance. Vertical forces may reflect normal forces to counter the mass of the beam-walker’s arm, which is consistent with passive impedance. While passive impedance can be implemented with conventional springs, dampers, and masses, “active impedance” (Aguirre-Ollinger et al., 2007) can only be physically realized with an actuated robotic system. Active impedance devices have been shown to improve performance in lower-limb target-acquisition (Aguirre-Ollinger et al., 2007) and upper-limb exploration (Huang et al., 2010) tasks. Several robotic walking aids use an admittance controller to determine the device’s output velocity based on hand forces through the equation: F = M x ¨ + B x ˙ , where F are forces exerted by the user at the handles, x ¨ is acceleration, x ˙ is velocity, and M and B are inertia and damping matrices (Chuy et al., 2005; Spenko et al., 2006; Nakagawa et al., 2016; Jiang et al., 2017). Notably, our model demonstrates the importance of a stiffness coefficient relative to the balance equilibrium point (vertical torso position) that we have not found implemented in any existing robotic walking aids and warrants further exploration. The coefficient values from our modelling results may be implemented directly in admittance controllers of existing robotic devices to emulate the intuitive balance-aiding strategies used in human-human interaction but are likely more appropriate for certain physical setups. For example, our experimental paradigm is more similar to a robotic cane moving beside the person and operated by one hand than a robotic walker/rollator moving in front of the person and operated by two hands. Finally, the fact that haptic balance aid can be modeled by a mass-spring-damper (and thus implemented in robotic systems) does not imply that the human-human interaction is created by mechanical elements; in fact, the haptic signal may communicate information related to acceleration, velocity, and position.

Our model also has limitations in its ability to fully describe human-human haptic interaction to aid balance. The interactions between humans were quite complex, as participants were free to choose how to interact and could vary force/torque magnitude, direction, and timing. The model has several underlying assumptions - such as linearity and time-invariance–that are not true of biological systems, and more complex nonlinear and transient processes may need to be added in the future to create a more comprehensive model. The inertia coefficient (I) of the model may not be reliable as acceleration was not measured directly, and it may represent a combination of the mass of the force handle, force sensor, and the beam-walker’s arm or body. Considerable variability in model coefficients and goodness of fit likely reflect the complexity of the balance task and variations in strategies between partnerships; e.g., some partnerships may communicate more information related to displacement while others may focus more on velocity. Relevant to design of robotic devices to aid human balance, the amount and type of physical interaction should vary according to the ability and needs of the user. Overall, the model provides a useful conceptual framework that can be tested in robotic systems in the future.