1 Introduction

Haptic interactions, such as touching, hugging, and kissing, play an essential role in affective interaction with others [1,2,3]. The effects of haptic interactions are not limited to building close relationships [4, 5]. Such interactions with intimates offer both physical and mental support, such as lowering blood pressure and the levels of proinflammatory cytokines [5, 6] and stress-buffering effects. Touching is important for children in the context of their developmental support [7, 8].

Although such haptic interaction is effective for humans from various perspectives, not everyone benefits from them due to several difficulties. For instance, the COVID-19 pandemic created physical separations [9]. To solve this problem, robotics researchers focused on social robots that can physically interact with people to "supplement" human–human haptic interaction [10]. Researchers developed human-sized robots with sufficient capabilities to touch people [9,10,11] and investigated the effectiveness of haptic interaction with robots [11, 12].

The above studies provided two major achievements in human–robot touch interaction. One is to develop robots that provide comfortable touching to people. These robots enabled researchers to investigate the effects of human–robot haptic interaction. Another one is to identify the positive effects of being touched by these robots. Such effects showed the advantages of using physical robots in interaction compared to non-physical agents such as computer-graphics-based characters. In particular, whole-body hug interaction is useful for robots that possess actual physical bodies. Researchers reported similar positive effects of human–human hug interaction [4,5,6] from both such physical and mental benefits of hugging interaction with robots as stress buffering and encouraging self-disclosures [11, 12].

Compared to human-hug interaction, human–robot interaction studies focus less on touching the head [13]. We thought this factor might increase the effects of hugging interaction. The reason is that people who have intimate relationships touch the faces and heads of others as intra-hug gestures, while hugging conveys intimate emotions [14, 15]. In fact, according to studies in the UK and Japan [16, 17], the head is only an acceptable place to touch between close relationships.

Although past studies in human–robot interaction investigated the effects of intra-hug gestures, they mainly focused on touching the back, not the head [10, 18]. Therefore, the effects of touching the head as well as how to do so, remain unknown during hug interactions with robots. Based on these considerations, we address the following research question:

What kind of intra-hug gesture is effective when a robot touches a person's head during a hug?

Based on our past work [11, 12], we developed a large huggable teddy-bear type robot named Moffuly-II (Fig. 1). It has enough capability to both hug and touch the head. To investigate an appropriate behavior design for touching the head during a hug, we experimented with thirty participants using Moffuly-II.

Fig. 1
figure 1

Snapshot of hugging with Moffuly-II

Full size image

Our paper is structured as follows. After introducing related work in Sect. 2, we present Moffuly-II's detailed systems in Sect. 3. We describe the experimental design in Sect. 4 and the experiment results in Sect. 5. We discuss its implementations and limitations in Sect. 6 and conclude by describing our main findings in Sect. 7.

2 Related Work

2.1 Touch Interaction Between Humans

The effects of touch interaction between humans have been broadly investigated from wider research perspectives: neural mechanisms, brain science, physiological changes, and psychological effects [19,20,21,22,23,24,25]. These studies reported how people react to touching behaviors, in particular, affective touches.

In the context of affective touching, the relationships between emotions and touch characteristics are essential research topics for understanding and reproducing such behaviors. For example, researchers investigated which body area is socially acceptable for touching based on its special characteristics, e.g., location and area [16, 17]. Other studies investigated the relationships between the temporal characteristics of touching behaviors like the duration and the velocity of the perceived impressions [26, 27]. Other characteristics, including force intensity and body temperature, also have important roles in conveying different emotions by touching [28,29,30]. Other studies reported the importance of affective touches for bonding [2, 31].

However, although human–human touch interaction has been broadly investigated, touching the head during a hug has received less attention. A past study reported that C-tactile fibers, which contribute to the perception of pleasantness [32], can be activated by the movement of scalp hairs [33]. This study supports the importance of head-touching behavior from the viewpoint of neurophysiology, although the effects of intra-hug gestures were beyond the scope of its examination.

Some studies qualitatively analyzed such touching behaviors by observing haptic interactions between people (and between animals), although they did not investigate the effects of touching head behaviors [14, 15, 34]. Other work investigated how caregivers or therapists touched their patients' heads and stroked them for therapeutic purposes [35,36,37]. These studies concentrated less on what kinds of head-touching behaviors are effective; they investigated the interaction procedures and styles of professionals in care contexts. A few studies investigated how people touched the heads of others and the importance of expressing affection through such touch interaction, particularly between parents and children, although they emphasized the perspectives of cultural differences and manners [38,39,40].

One difficulty of quantitatively investigating the effects of types of head-touching behaviors may lie in its manner since head-touching is only acceptable between those who share close relationships. When head-touching is done as an intra-hug gesture, the hug effects must be separated from the head-touching effects. Therefore, exclusively investigating the effects of head-touching behaviors is difficult in human–human haptic interactions.

Compared to existing studies on touch interaction between humans, the unique point of our study is its investigation of the effects of head-touching behaviors using a huggable robot to avoid relationship effects. This approach will provide knowledge for designing appropriate haptic interaction with social robots.

2.2 Touch Interaction Between Humans and Robots

Following the positive effects of touch interaction between humans, robotics researchers first developed various devices that people can touch [41,42,43,44,45,46]. Touching baby seal robots provides mental therapy effects [42] and pain reduction [47]. Researchers also investigated people's touch interaction using robots' sensors to identify interacting individuals [48], behaviors [49], and attitudes [50].

Due to the advances in sensing and actuator hard-ware, researchers also developed robots that can actively touch people. These studies focused on touch interaction by robot hands because the hand obviously has an essential role in touch interaction between humans. Past studies reported how being touched by robot hands effectively conveys emotions [46, 51,52,53], friendly impressions [54, 55], and encouragement [56, 57].

Step by step, the interest of researchers has grown in more human-like, intimate touch interactions compared to hand-touching behaviors: hug interaction. Several studies developed devices that can achieve hug interactions [58,59,60,61] and reported how robots could actively hug people and described the positive effects of hug interaction with robots, e.g., encouraging self-disclosures and stress-buffering [11, 12]. Researchers investigated the effectiveness of such intra-hug gestures as squeezing and rubbing, which were identified in relatively long-hug interactions, and reported their importance for greater comfortable hug interactions [10, 18].

These studies provided useful knowledge to create robots that support people through hug interactions. However, although past studies in human–human touch interaction identified the importance of head-touching [35,36,37], robotics researchers have not yet sufficiently investigated it in the context of human–robot interaction. Some studies reported that people rub or stroke a robot's head to express positive impressions [62,63,64], resembling a parental touch that expresses affection [65]. Unfortunately, these studies only investigated how people touched robots, not the touch designs for the robots themselves.

Creating comfortable head-touching behaviors from robots to people enables us to reproduce human-like affective touching during hugs [14, 15] and to quantitatively investigate its effects. Compared to existing studies on touch interaction between humans and robots, our study has two unique points: (1) developing a robot with sufficient capability to touch heads during hugs and (2) investigating the effectiveness of our head-touching behavior design.

3 Hug Behavior, Device, and System

3.1 Hug Behavior Design

This study investigates the effects of head-touching behaviors during hug interactions. However, since touch gestures can vary widely from hitting to stroking and patting to tapping, selecting promising candidates for intra-hug gestures for comparisons is important.

We employed two kinds of intra-hug gestures (rubbing and squeezing) based on the results of a past human–robot interaction study [10], which compared the effects of four intra-hug gestures on the back, and showed the advantages of squeezing over patting and holding. The experiment results also described the possibility of rubbing, which also outperformed patting and holding. Past human science literature also reported that people stroke or rub the heads of close or intimate others, although they don't squeeze [35,36,37].

Based on these considerations, we investigated the effectiveness of different types of head-touching behaviors and touched area by preparing two factors for hug behavior design: gesture (rubbing and squeezing) and area (head and back).

3.2 Huggable Device

To rub and squeeze a head during a hug, we needed to determine how to design a robot that could function as a huggable device. For this purpose, the robot requires the capability to make intra-hug gestures, similar to past studies that reported the effectiveness of hug interaction [10, 18]. Based on past studies, we broke down the essential factors of huggable robots that provide whole-body, comfortable hug interactions [58,59,60,61]. We focused on three characteristics of hardware design: body size, materials, and arm length. These studies developed adult-size, soft materials, and enough arm-length to hug people for comfortable whole-body interactions. Following these results, we developed Moffuly-II (Fig. 2), a large teddy-bear type robot, which is a modified version of a previous iteration of Moffuly [12].

Fig. 2
figure 2

Moffuly-II (left) and its fabric-based touch sensors

Full size image

Moffuly-II is 200 cm tall with 110 cm arms, designed for hugging adults and touching their heads during such interactions. For intra-hug gestures, Moffuly-II has two degrees of freedom (DOFs) in each elbow and one in each wrist. Its body consists of a metal frame, cotton, and fabric skins. Cotton functions as a buffer between the frame and the skins. Moffuly-II's mouth area is removable because it directly contacts people's faces during hugs.

We installed a Raspberry Pi 4, six motors (Dynamixel MX-106R, 8.4 Nm), a speaker, and fabric-based touch sensors in Moffuly-II (Fig. 2). The Raspberry Pi 4 autonomously controls each joint angle and speed for safe hug behavior using its torque information and the touch sensor outputs. The details of the control mechanism are described in the next subsection.

3.3 System Overview and Software

Figure 3 shows an overview of our developed system. We prepared functions for both autonomous and semiautonomous hug behaviors.

Fig. 3
figure 3

System overview

Full size image

The hug controller manages the robot's motions using sensor outputs, torque information, and the target touch area. When the touch sensors or the torque values exceed thresholds, the robot controllers stop the hug motions for safety.

The autonomous hug function also uses each joint's sensor outputs and angle information for giving hugs. When a person initiates a hug, the robot autonomously closes its arms to return the hug. When a person finishes a hug, i.e., dropping her arms from the robot's body and stepping back, the robot also opens its arms to finish the hug.

Because such human physical characteristics as height differ between individuals, we also prepared a calibration function. The robot can record appropriate arm and hand angles, i.e., the angles of three joints in each arm, for rubbing/squeezing the interacting person's head and back with touch sensor outputs.

We also prepared a teleoperation controller to enable operators to control the robot's hugs. Similar to the autonomous hug function, the operator can control a hug's timing and intra-hug gestures by observing the touch sensors' outputs.

3.4 Intra-Hug Gestures

We implemented two intra-hug gestures (squeezing and rubbing) for two different touch areas (head and back). We used the robot's left hand for touching the head or the upper back and its right hand to touch the lower back.

Following past studies [10, 66], we designed squeezing gestures to hold the users for three seconds tightly and rubbing gestures to move the robot's left hand for three seconds vertically.

Figure 4 shows a squeezing intra-hug gesture of the back and the head. The elbow flexion joints of the left arms were adjusted inward by 35°, and the left wrist flexion joints were simultaneously adjusted inward by 25°; these two joints were then returned to their original values.

Fig. 4
figure 4

Squeezing gesture movements

Full size image

Figure 5 shows the rubbing intra-hug gesture of the back and the head. For this gesture, we adopted only a rubbing-down motion to prevent disturbances of hair and clothing from the rubbing-up motion. At the beginning of the rubbing gesture, the left arm is placed above the user's head. The elbow flexion joints of the left arms were adjusted downward by 35° when touching the back and by 25° when touching the head. When the robot's left hand is completely lowered, the left arm returns to the starting point in a semicircle. The rubbing speed is about 10 cm/sec to stimulate C-tactile fibers [27] efficiently.

Fig. 5
figure 5

Rubbing gesture movements

Full size image

In both intra-hug gestures, the right arm's elbow flexion joints were adjusted inward by 30°, and the right wrist's flexion joint was simultaneously adjusted inward by 25°. These two joints moved only at the start and the end of the intra-hugs. Each movement is controlled in joint space relative to the current hug's embracing pose around the user.

4 Experiment

4.1 Hypotheses and Predictions About Intra-Hug Gesture

We assume that the rubbing gestures will have a positive effect while hugging a robot because past studies that analyzed touch interaction between humans reported the importance of rubbing gestures [35,36,37]. However, another past study that developed a huggable robot described greater advantages of squeezing gestures compared to rubbing gestures [10]. Since the phenomena related to intra-hug gestures remain basically unexplored, we made two predictions based on different hypotheses.

Our first hypothesis is related to the positive effects of rubbing gestures. As described above, rubbing (or stroking) gestures are often observed when the toucher expresses positive attitudes or supports the mental/physical perspective of a receiver: humans, animals, and robots [14, 15, 34,35,36,37, 62,63,64]. Therefore, we hypothesized that people will prefer rubbing gestures more than squeezing gestures when a robot is hugging them. Based on these considerations, we prepared a prediction related to the advantage of rubbing gestures:

Prediction 1-a people will positively evaluate the hug impressions and accept the robot when it uses more rubbing gestures than squeezing gestures.

Our next hypothesis advocates a contrary position, i.e., related to the positive effects of squeezing gestures. Although the above studies related to touch interactions reported the advantages of rubbing gestures, they less focused on hug interactions between humans and robots, particularly when the latter actively touches the former. A past study, which actively touched people during a hug, reported the advantages of squeezing gestures compared to rubbing gestures [10]. Therefore, another possible hypothesis is preferring squeezing gestures over rubbing gestures during a robot's hug. If we follow this hypothesis, the following prediction is also reasonable:

Prediction 1-b people will positively evaluate the robot's hug impressions and accept it when it uses squeezing gestures more than rubbing gestures.

4.2 Hypothesis and Prediction About Touch Area

Compared to the squeezing and rubbing gestures, past studies consistently emphasized the importance of head-touching. For example, related studies about touch interaction between humans showed that people touch others' heads in close relationships [35,36,37]. In human–robot interaction contexts, several studies showed that people touched robots' heads when they expressed positive relationships [62,63,64].

Although these studies did not investigate the effects of people being touched by robots, such an interaction style would be effective when a robot touches people. Therefore, we hypothesized that people will prefer head-touching over back-touching during hug interactions with a robot. Based on these considerations, we prepared a prediction related to the advantage of head-touching gestures:

Prediction 2 people will positively evaluate hug impressions and accept the robot when it touches their heads more than when it touches their backs.

4.3 Conditions

We employed a within-participant design for our experiment, which investigated what kind of intra-hug gestures are effective when the robot touches a person's head during a hug. We compared two different intra-hug gestures (squeezing and rubbing) with different area (back and head). Thus, the participants experienced four different hug iterations with our robots. The order of four conditions was counterbalanced to avoid the order effects.

4.4 Participants

Thirty Japanese people participated in this experiment: 15 women and 15 men whose ages averaged 38.33 with a standard deviation (S.D.) of 12.23 and a range from 20 to 59. We took significant precautions to protect their health during the COVID-19 pandemic.

4.5 Procedure

Before the experiment, the experimenter explained its aims. After obtaining informed consent, participants watched a video that summarized the experiment's explanation. It showed how to hug the robot (no explanation about being squeezed and rubbed), the calibration process, and how to answer the questionnaires/interviews. After watching the video, the experimenter also physically demonstrated how to hug the robot. Then the participants hugged it to confirm the head and back positions.

After the calibration process, participants experienced hug interactions with the robot under each condition. The duration of each was about ten seconds, including a three-second hug interaction. During the interaction, the participants and the robot silently hugged each other to avoid any influence of conversations. After each hug interaction, the participants filled out a questionnaire on a tablet near the robot. We conducted interviews at the end of the experiment.

4.6 Measurements

We prepared two different questionnaire items to evaluate the participants' hug impressions and their acceptance of the robot through the interactions. For the former, we employed the same questionnaire items to evaluate the hug impressions following a past study, which also evaluated human–robot hug interactions [18]. The first item consisted of five questions about impressions of the robot. Participants answered them on a 7-point response scale (where 1 is the most negative and 7 is the most positive). The entire questionnaire is shown in Table 1.

Table 1 Statements about robot's hug impressions
Full size table

For the latter, we employed the modified version of the Unified Theory of Acceptance and Use of Technology (UTAUT) [10, 18]. This measurement was often employed in human–robot interaction studies with different modifications [68,69,70]; note that we used the same version of [10] because this study also investigated the hug effects of human–robot interaction. It evaluates acceptance by investigating performance expectancy, effort expectancy, social influence, facilitating conditions, reciprocity, and attachment.

Participants answered the questions shown below on a 10-point Likert scale, where 1 is the most negative (completely disagree), and 10 is the most positive (completely agree). The entire questionnaire is shown in Table 2.

Table 2 Statements about modified versions of UTAUT
Full size table

At the end of the experiment, we also asked for their preferences for the combinations of intra-hug gestures and touched areas that allow multiple answers.

5 Results

5.1 Impression of Robot

Figure 6 shows the questionnaire results of the impressions of the robot. We conducted a two-way factorial ANOVA on the questionnaire items and found significant differences in the gesture factor of all items: friendly (F (1,29) = 13.899, p < 0.001, partial η2 = 0.324), safe (F (1,29) = 11.966, p = 0.002, partial η2 = 0.292), intelligence (F (1,29) = 10.875, p = 0.003, partial η2 = 0.273), enjoyment (F (1,29) = 10.333, p = 0.003, partial η2 = 0.263) and naturalness (F (1, 29) = 17.069, p < 0.001, partial η2 = 0.371).

Fig. 6
figure 6

Average and standard error of questionnaire results about impressions of robot

Full size image

We did not find any significant differences in the area factor (friendly: p = 0.271, safe: p = 0.079, intelligent: p = 0.580, enjoyment: p = 0.750, naturalness: p = 0.666), and the interaction effects of all the questions (friendly: p = 0.250, safe: p = 0.114, intelligence: p = 0.526, enjoyment: p = 0.240, naturalness: p = 0.489).

5.2 UTAUT

We calculated Cronbach's alpha to test the factor's reliability. For all the factors, alpha was higher than 0.7 (A: α = 0.929, AT: α = 0.766, C: α = 0.901, G: α = 0.839, PE: α = 0.966, R: α = 0.925).

Figure 7 shows the questionnaire results of the modified versions of UTAUT. We conducted a two-way factorial ANOVA on the questionnaire items and found significant differences in the gesture factor for all items: A (F (1,29) = 7.600, p = 0.009, partial η2 = 0.208), AT (F (1,29) = 7.548, p = 0.010, partial η2 = 0.207), C (F (1,29) = 4.873, p = 0.035, partial η2 = 0.144), EE (F (1,29) = 6.062, p = 0.020, partial η2 = 0.173), G (F (1,29) = 10.821, p = 0.003, partial η2 = 0.272), PE (F (1,29) = 4.477, p = 0.043, partial η2 = 0.134), R (F (1,29) = 6.990, p = 0.013, partial η2 = 0.194).

Fig. 7
figure 7

Average and standard error of modified versions of UTAUT

Full size image

In addition, we found interaction effects of the items about C (F (1,29) = 4.302, p = 0.047, partial η2 = 0.144), G (F (1,29) = 4.700, p = 0.039, partial η2 = 0.139) and PE (F (1,29) = 5.413, p = 0.027, partial η2 = 0.157). The simple main effects showed a significant difference in the head condition (C: rubbing > squeezing, p = 0.003, G: rubbing > squeezing, p < 0.001, PE: rubbing > squeezing, p = 0.004) and in the rubbing condition (PE: head > back, p = 0.048).

We did not find any significant differences in the area factor for any of the items (A: p = 0.523, AT: p = 0.571, C: p = 1.000, EE: p = 0.211, G: p = 0.580, PE: p = 0.485, R: p = 0.333) or the interaction effects of the items about A: p = 0.137, AT: p = 0.546, EE: p = 0.092, R: p = 0.084.

5.3 Preferences of Gestures and Area

Figure 8 shows the participants' preferences of hugs based on the results gathered at the end of the experiment. Each box represents the number of participants who choose each condition. We conducted a Cochran's Q test and found a significant difference (Q (3) = 24.871, p < 0.001). Multiple comparisons with the Bonferroni method showed significant differences between head-squeezing and head-rubbing (p < 0.001) and back-squeezing and head-rubbing (p < 0.001). We did not find any significant differences between head-squeezing and back-squeezing (p = 1.000), back-squeezing and back-rubbing (p = 0.067), head-squeezing and back-rubbing (p = 0.452), or back-rubbing and head-rubbing (p = 0.253).

Fig. 8
figure 8

Preferences of hugs

Full size image

5.4 Verifications of Predictions and Summary

Our analysis of the robot's impressions and UTAUT showed the advantages of rubbing gestures compared to squeezing gestures in the impressions of the robot. Prediction 1-a was supported, i.e., not the opposite, prediction 1-b. Thus, people evaluated the rubbing gestures more highly than the squeezing gestures, which is the opposite phenomenon from previous work [10].

Although the area factor is insignificant in the analysis of the robot's impressions, the UTAUT analysis also showed the advantages of the combined effects in the head area with a rubbing gesture in the C/G/PE scales. Analysis of the preferences showed that the participants significantly preferred head-rubbing over head-squeezing and back-squeezing. These results show that prediction 2 was partially supported, i.e., since people evaluated the head-rubbing gestures in some of the evaluation criteria more than back-touching behaviors. In the next section, we discuss the possible reasons for this phenomenon.

6 Discussion

6.1 Design Implications

Our experiment results showed the advantages of rubbing gestures during hug interaction. Interestingly, these are opposite results from previous work [10], perhaps due to the differences in the rubbing gestures. The past study's rubbing gestures included both up and down strokes. On the other hand, our study just adopted a downstroke to avoid disturbing hair and clothing and used the same motions for its back-touching. Although the difference seems slight, it might explain the various feelings of the participants.

Another possibility is cultural differences. Past studies and ours did not directly compare the effectiveness of rubbing and squeezing gestures during hug interactions with people from different countries/cultures. Although the results of these studies provide basic knowledge to understand/compare the effect of intra-hug gestures, the effects of cultural differences are unknown.

Cultural differences may also influence a possible reason for the advantages of head-touching behavior compared to back-touching behaviors during hugs because we conducted the experiment in Japan. Past studies reported that head-touching behaviors are more often observed in Eastern Asia than in Europe and the U.S. [38,39,40]. Therefore, Japanese participants might perceive head-touching behaviors more positively.

6.2 User's Comments

During the interviews, several participants explicitly praised the robot's rubbing gestures and its head-touching behaviors. Four felt that the rubbing gesture was closer to a human-like behavior than the squeezing gesture. Three others described the squeezing gestures as rather mechanical because the robot only touched one point. These comments suggest that touching a larger area during a hug might provide positive impressions to those getting hugged.

Regarding the touch area, five participants gave positive impressions about the head-touching behaviors, regardless of the gestures: “I felt close to the robot,” “I felt friendly toward the robot,” “I felt more protected,” “I felt affection,” and “I felt comforted.” Five others reported positive impressions of the head-rubbing gestures: “I was impressed because I thought that only people could rub my head,” “I was greatly impressed when the robot rubbed my head,” “I felt so relieved when the robot rubbed my head,” “The robot seemed to understand me,” “When he rubbed my head, I felt he accepted me.” These comments provide additional evidence that supports why our implemented head-rubbing gestures were preferred.

6.3 Limitations

This study has several limitations. First, since we only experimented with a specific robot (Moffuly-II), appearance effects, the touch-feeling strength of the hug interactions, and the robot's size remain uninvestigated. For example, touching a head by both a big and a small robot may provide different impressions. Moreover, the experiment did not deal with any relationship effects between the people and the robot. We need to investigate how long-term interaction affects the perceived impressions of hugs.

However, we believe that developing a huggable robot capable of hug interactions with intra-hug gestures and investigating their effects on head-touching context will provide rich knowledge for researchers who are working on affective interaction by physical contact with robots.

7 Conclusion

Whole-body haptic interaction, e.g., hugging, provides physical/mental support to people in both human–human and human–robot interaction. Past studies reported that such intra-hug gestures as head-touching behaviors might increase positive effects in hug interaction between humans. On the other hand, researchers were less focused on head-touching behaviors and hug interaction between humans and robots.

To investigate the effects of such intra-hug gestures as squeezing and rubbing, we developed a huggable robot that hugs people and can touch their heads and backs. We conducted an experiment with human participants and evaluated the effectiveness of combining intra-hug gestures and the touch area. The experiment results showed the greater advantages of using rubbing gestures than squeezing gestures. Our results also showed some advantages of the head-rubbing gesture. We expect that these findings will provide rich knowledge for further work on affective interaction by physical contact with robots.