In Contact: Pinching, Squeezing and Twisting for Mediated Social Touch

Mediated social touch has the potential to enhance our interactions with machines and with each other. We present three wearable tactile devices that generate affective haptic sensations via three localised skin stretching modalities; pinching, squeezing, and twisting. The Pinch device is adhered to the skin of the forearm, generating pinching sensations in three locations. The Squeeze and Twist devices are wristbands that elicit squeezing and twisting sensations on the skin of the wrist. All of these devices are powered by shape memory alloy actuators, enabling them to be quiet, lightweight and discreet wearable interfaces, unlike their vibrotactile or servo-motor driven counterparts. We investigate the potential for these devices to be used in mediated social touch interactions by conducting preliminary psychometric tests measuring affective response. The Pinch device and Squeeze wristband were found to simulate positive affective touch sensations, particularly in comparison to vibrotactile stimuli.


Introduction
Research on human touch has largely focused on discriminative touch for sensing, but recent studies have recognised the importance of affective touch and its role in social interactions [17,7]. These findings support the growing field of research into haptic devices for mediated social touch [13], allowing people to communicate remotely via touch.
Existing devices generally use vibration or motor-driven force to generate sensations on the skin that either replicate actual social touch (e.g. a handshake [19] or hug [25]) or communicate higher-level symbolic meaning between communicating parties [12]. Recently, however, there has been research into other methods of stimulation in an attempt to generate more natural and localised sensations [13]. To create mediated social touch interactions that are accessible on-the-go throughout a person's daily routine, such devices need to be wearable, comfortable and discreet.
We present three wearable devices and investigate their ability to convey emotive touch sensations. The devices are designed to simulate three forms of human touch; pinch, squeeze and twist, as illustrated in figure 1. The Pinch device (figure 1 top) is based on our previous skin-stretching device [9]. Its triangular configuration provides more degrees of freedom with subsequent different alignments of skin stretching on the forearm. The Squeeze and Twist devices are designed as wristbands, one which contracts to squeeze the wrist (figure 1 middle) and one which has rotating elements to twist the skin (figure 1 bottom). We investigate the emotive responses of wearers to these different sensations.

Related Work
The majority of existing haptic interfaces are vibration based. Efforts have been made to anthropomorphise vibrotactile feedback such as the CheekTouch [21] and ForcePhone [11] which use vibration patterns to signify different social interactions. The TaSST is a vibrotactile sleeve used for mediated social touches [14] such as for conveying squeezing and stroking. Tsetserukou et al. [27] created a number of devices to elicit various human feelings. They used vibration to simulate tickling, shivering and butterflies in the stomach and a speaker on the chest conveyed the other person's heartbeat.
While vibration motors are low cost, compact and effective, they are limited in the range of sensations they can elicit. Vibrotactile stimuli activate the fast-acting mechanoreceptors (the Meissner and Pacinian corpuscles) within the dermis which have a relatively large perceptive field. Skin stretching sensations, on the other hand, can also activate the slow-acting mechanoreceptors (Ruffini endings and Merkel's disks) [5,1] that have smaller receptive fields and process localised force information. Skin stretching is also capable of activating the CT afferents in human hairy skin which process social and affective touch [20]. This evidence suggests that vibrotactile stimuli alone may not be sufficient to simulate diverse and meaningful affective touch sensations.
Existing skin-stretch devices and interfaces can be categorised by their location on the body. Finger-tip displays are commonly used [23,6,18,26,10,4] for providing natural discriminative touch feedback in virtual situations. These devices are generally powered by servo-motors that move linkages or belts to create shear forces on the pad of the finger [23,6,18], or independently move metal pins on the skin surface [10,4]. Some of these devices are wear-LBW055, Page 2  able [18,26], but fingertip displays are generally only practical for specific interaction tasks rather than for wearing throughout the day.
Of these devices, those that were designed for affective touch or mediated social touch have aimed to simulate real human interactions. Stanley and Kuchenbecker [24] created wrist-worn devices to simulate four types of human touch; tapping, dragging, squeezing, and twisting. The sensations were generally found to be comparable to human touch and participants reported that squeezing in particular felt natural and pleasant. Wang et al. created a servo-motor driven device that squeezes a listeners arm at specific times during a story. They found that it increased the listener's sense of connectedness with the storyteller [28]. Knoop and Rossiter [15] created a wearable wristband designed to gently stroke the user's skin as a method of conveying affection and emotion. Hamdan et al. [8] used SMAs attached to pads adhered to the skin in a number of ways to generate six different tactile sensations: pinching, directional stretching, pressing, pulling, dragging, and expanding.

Device Design
The Pinch device (figures 2 and 3) consists of two structural 3D printed parts (Wanhao Duplicator i3). The red part is printed with a flexible filament (TPU) and has six equal legs. At the end of three of those legs are circular rings within which poppers were attached and under which adhesive pads attached the device to the skin. Rigid reinforcement elements were 3D printed from PLA to strengthen the rings (shown in black in figure 2). The three shape memory alloy (SMA) coiled wires (BioMetal Helix, BMX series 15000) were attached to the ends of these rigid elements to create a triangular shaped device.
User Study 10 volunteers (7 males, 3 females) were asked to wear the device on the inside of their forearm orientated as shown in figure 3. Each SMA on the device was separately activated at three voltages (1.67 V, 2.5 V and 5 V), as described in [9], and participants were asked to rate the strength and pleasantness of the sensation on a scale of 1-10, consistent with the Circumplex model of affect [22]. For comparison with existing devices, participants were asked to wear the Pinch device, the previous version [9] and a smart watch which provided vibration stimuli. Participants were asked which sensation(s) they preferred out of these devices.
To determine whether the device could be used to convey information, the three different SMAs were actuated and participants were asked to choose which SMA they thought had been activated. To investigate how subtle the device is, participants were asked to record each time they noticed a sensation from the device while they were doing nothing, reading a book or playing a game. The percentage of correctly noticed activations was recorded.

Results
As shown in figure 5, participants consistently found the Pinch device pleasant with little variation between participants. All participants reported that they preferred the sensations generated by the Pinch device and the previous skin-stretching device in comparison to vibration.
There was no significant difference (Welch's t-test at 10% significance) between the SMA positions for both perceived strength and pleasantness (figure 4 left), but there is a positive Pearson correlation coefficient (r-val=0.866; p-val=0.000) between perceived strength and the voltage of activation (figure 4 middle). Pleasantness is slightly increased by increased voltage (r-val=0.314; p-val=0.003).
Participants were able to correctly determine which SMA was activated with 90% mean accuracy. From the confusion matrix (figure 4 right) it can be seen that all incorrect responses for SMA wires S1 and S2 stimulation were recorded as S3 stimulation (i.e. S3 had the most false pos-itive responses). S3 also generated the fewest incorrect responses. This could be an indication that the location or direction on the forearm that SMA wire S3 stimulates is more sensitive than the areas stimulated by S1 and S2.
When participants were distracted by tasks that took more concentration they were significantly less accurate at noticing pinch sensations (supported by Welch's t-test at 1% significance showing statistical independence); their % correctly noticed sensations was 99% for no task (std 3.2%), 90% when reading (std 8.2%) and 61% when playing a game (std 15%), as shown in figure 6. In the previous study [9] it was found that vibration stimuli were consistently noticed (mean 99%) during all tasks. This suggests that the Pinch device is able to convey subtle alerts where the user is less likely to be disturbed when involved in a task that requires more concentration, but when they are not focusing on a task they notice the sensations.  Participant's comments were generally positive with people stating that the Pinch device felt "like a real human hand" and "like someone is touching me". They also commented on the the fact that the Pinch device "is really quiet and doesn't affect me and distract my attention" and "it's lightweight and doesn't put an extra burden on my arm". There were, however, some users who felt that the device had too many wires for it to be practical and that "the tactile feeling it brings does not appeal to me".

Squeeze and Twist Wristbands
Device Design Both wristbands are based on auxetic structures. The Squeeze wristband (figure 9 top) is made up of seven re-entrant hexagons placed end-to-end. It is 3D printed (Wanhao duplicator i3) from flexible filament (TPU). The coiled SMAs are attached along the centre of each auxetic unit. To avoid buckling of the wristband, PLA printed rods were glued to the structure between each unit (shown in black in figure 9 top). Actuation of the SMAs causes the structure to shorten and to squeeze the wrist. The Twist wristband (figure 9 bottom) is made up of a connection of crosses. Four crosses combine to make one auxetic unit. The wristband consists of nine auxetic units arranged alternating to 90°. Actuation of the SMAs causes each cross to rotate and to twist the skin. Both wristbands have adjustable velcro straps to ensure a secure fastening on the wrist and direct skin contact.
User Study 10 volunteers (8 males, 2 females) participated in this part of the study. The wristbands were worn on the user's wrist and participants were asked to rate on a scale of 1-10 the strength of the device and how natural the generated sensations felt. The participants were then asked to compare these sensations to that of a vibration device such as those found in a mobile phone for alerts. To test the devices as a CHI 2020, April 25-30, 2020, Honolulu, HI, USA means of long-distance interactions, a second participant activated the device remotely and the wearer was asked to rate out of 10 how effective they felt the interaction was in the context of using the device for mediated social interactions. At the end of the experiment participants were asked for feedback on the device.

Results
The results show that the Squeeze wristband felt stronger and more natural than the Twist wristband ( figure 7). When asked to compare the wristbands to vibration, participants found the Squeeze and Twist devices more natural with a mean rating of 9.3 (std 0.8) and 6.8 (std 1.0) respectively. Participants found the Squeeze wristband more effective than the Twist wristband when used as a remote device and activated by another person, with a mean effectiveness rating of 7.2 (std 1.0) for the Squeeze wristband and 4.2 (std 1.0) for the Twist wristband (figure 8). Participant's comments indicated that the wires were impractical and suggested making the wristbands remote controlled. Future iterations of the devices can address this. Some participants commented that the rotation wristband did not generate strong enough sensations for them to really feel or appreciate the twisting motion. We will undertake further design iterations to generate more effective twisting sensations. Aside from these suggested improvements, participants commented that they liked the wristband devices as they were not bulky or noisy.

Discussion
Pleasant sensations were generated by the Pinch device (figure 5) which was preferred to vibrotactile sensations. The location of actuation (S1, S2 or S3) did not correlate to pleasantness. However, figure 4 (right) suggests that the area of the forearm at S3 is most sensitive as participants had the least incorrect responses to this stimulation and all incorrect responses for S1 and S2 were recorded as S3.
The Pinch device was able to provide information of different strengths as participants recorded an increase in perceived strength as voltage increased from 1.67V to 5V (figure 4 middle). This gives the Pinch device the potential to use different levels of voltages on the skin to simulate different sensations or to symbolise emotive meaning. As concentration on a task increased, participants were less accurate at detecting sensations generated by the Pinch device; comparing no task, reading and playing a game ( figure 6). The Pinch device could therefore be used as a means of providing subtle alerts that do not disturb the user when concentrating on work or driving, for example. The Squeeze wristband was also able to provide natural sensations ( figure 7). An advantage of it over the Pinch and previous skin-stretching device is that it is worn as a wrist-CHI 2020, April 25-30, 2020, Honolulu, HI, USA band rather than adhered to the skin and is therefore more practical as a wearable device. The Squeeze wristband was more effective at generating natural sensations compared to the Twist wristband ( figure 7). When used as a communication device between two people, the Squeeze wristband was more effective at providing stimulations compared to the Twist wristband (figure 8).

Conclusion and Future Work
We have demonstrated that affective touch can be achieved with skin-manipulating devices that are preferred over devices using vibrotactile sensations. We have shown that SMA-driven devices are capable of generating Pinch and Squeeze sensations on the skin that are pleasant and natural. The Twist sensation was less effective, however this could potentially be enhanced by adding points of contact to improve skin coupling.
Further development of these devices will predominantly focus on making them untethered and remote controlled so that they can be truly wearable. This will allow us to test their effectiveness in real social interactions, outside of a laboratory environment.
Another area of interest for future work is to combine multiple Pinch units together to form a modular network of skinstretchers as illustrated in figure 10. This network can be embedded in fabric and worn discreetly underneath clothing. The modular format could become the haptic equivalent of a prototyping toolkit [30], allowing people to design personalised touch interactions. Having a network of skinstretchers would greatly increase the capacity for providing information, at different locations, strengths and patterns.
With these further developments, the devices could be discreetly worn in daily life, giving people a means of both sending and receiving affective touch.