1. Introduction
A virtual partner (VP) is a virtual character simulation of a real human, also referred to as a software-generated partner (SGP) [
1]. The interactions between VPs and humans are more interactive than those between humans and ordinary graphics and texts. As virtual simulation technology is becoming increasingly mature, VPs are being developed with diverse and personified characteristics that not only verbally interact with the user, but also achieve nonverbal interactions through body movement (BM) and eye gaze (EG) [
2], thereby providing an interactive experience close to that of a real person. Many sports and fitness software applications have been developed to provide a VP to accompany individuals during exercise, such as the exergames Just Dance, Ring Fit Adventure, and Fitness Boxing [
3]. These programs have been widely adopted by young exercisers.
In addition to enhancing the interactive experience, VPs provide an incentive to work out. Feltz and other scholars have demonstrated that VPs can elicit the Köhler effect [
4,
5], which is exhibited strongest in joint tasks [
6]. In the Köhler effect, when individuals perform effort-based tasks as part of a team, the least capable team members display motivation gains [
7]. There are two key mechanisms for generating the Köhler effect. The first is social indispensability, in which the less capable members perceive their contributions as integral to group success and social evaluation. The second is upward social comparison, involving more collaboration with competent group members [
8,
9]. Some scholars later applied VPs to continuous workouts and found that exercising with better VPs produced outcomes of motivation gain with moderate effectiveness [
10]. However, some studies have drawn inconsistent conclusions, and it has been suggested that in long-term sports training, upward social comparison with the consistently exceptional VP may induce a depressed state of mind in individuals, which can inhibit their self-assessment [
11,
12]. Based on the above findings, we found that the effect of a VP’s SP on exercise motivation is not clear, but the VP’s SP is an important interface design element in the VP-accompanied exercise system. Therefore, the motivational effect of a VP’s SP requires further analysis in specific sports contexts.
Most of the existing research on the motivational effect of VPs has limited the interactions between individual exercisers and the VP. A small number of studies investigating the interactive features of VPs focused primarily on verbal interactions with the VP. However, nonverbal interaction is vital for interpretation and interaction in real human-to-human interactions. Related behavioral studies have indicated that individual behavior can be influenced by observing the behavior of others, especially when the individual intentionally interacts with others. Participants tended to imitate the kinematics of other’s actions [
13], even if such imitation was detrimental to the individual’s success in a competition [
14,
15]. Additionally, studies have shown that pair performance improves when participants can directly experience their partner’s action during cooperative motor interactions [
16]. Therefore, when engaging in sports with a VP, the VP’s BM may enhance the sports performance of exercisers. However, the current knowledge is inconclusive, and whether the presentation of BM from a VP in a sports context influences exercisers remains to be further examined.
Moreover, the EG of a VP is an important feature of nonverbal human–computer interaction. Psychology and behavioral science studies have shown that EG affects human attention, behavior, and other cognitive processes [
17]. Relevant research in computer science and human–robot interaction (HRI) has also demonstrated that EG can capture attention, maintain engagement, and enhance human–robot interactions [
18]. Furthermore, the EG of a VP is likely to influence the exerciser’s perception that another person is present [
19]. According to the social facilitation effect, the presence of others (in person or remotely; real or perceived) will promote the performance of simple tasks and inhibit the performance of complex tasks [
20], and virtual characters can also elicit social facilitation effects [
21]. Thus, as an important interactive feature of VP, EG may impact the human–computer interaction experience or elicit the social facilitation effect. However, this variable was not controlled in previous studies, and the EG direction of the VP varied among the exercise task conditions. For example, in an experimental study involving planking [
22,
23], EG appeared in only one of the five movements of the VP, and the remaining four movements were conducted in a sideways position relative to the exerciser. In another study involving riding stationary bicycles [
11,
24], the VP always had his back toward the exerciser; thus, no EG was present. Although these studies all showed that the participants had a positive attitude toward the VP and considered themselves part of a group, some of the participants performing the bicycle ergometer experiment indicated that longer workouts may be needed to build team awareness [
24]. Therefore, the present study hypothesized that the EG orientation of a VP might influence exercise motivation, results, and perception. Few studies have explored this topic, and it remains to be clarified whether the EG feature of a VP will elicit the social facilitation effect and how EG affects sports performance and exercise perception.
In previous studies of the VP, the subjective exercise perception of exercisers was the focus, and related investigations have assessed the exercisers’ subjective exercise enjoyment, attitudes toward VP, sports-team mindset, and willingness to continue the workout. Previous studies have suggested that the presence of a VP does not significantly improve exercise enjoyment [
5]; boosted motivation was not accompanied by increased exercise enjoyment [
22]. However, the existing studies on individual exercise enjoyment mostly focused on the appearance of the VP and social category features; few have explored the interactive features of VPs, such as BM and EG. In addition, the exercisers’ attitudes toward the VP and their sports-team mindset are important factors in motivation research involving VP-accompanied workouts. Establishing a competitive or cooperative relationship between VPs and exercisers through human–computer interactions stimulated EL [
1,
25], and the participants who viewed the VPs as teammates and collaborated with them attained significant motivation gain in the perseverance of exercise training [
26]. Research has shown that the social category feature setting of a VP could improve its team relationship with the exerciser [
23], whereas it remains unclear how the VP’s interactive features affect exercisers’ perceptions of the VP and attitudes toward the sports team formed with the VP.
Studies have confirmed that VP characteristics, such as presence or absence, verbal interaction, and athletic ability, can have a significant motivational impact on individual sports performance. However, whether VPs are better than cooperative athletes in human–computer interaction design remains unknown. The effects of a VP’s BM and EG on exercisers and what the possible influences are also remain unknown. Additionally, the effects of these design factors on the individual subjective perception and sports performance of real cooperative exercisers require further exploration. Therefore, in this study, a self-designed VP-accompanied exercise system was used to probe the impact of the above interactive VP features on exercise level (EL), exercise perception (subjective exercise enjoyment and the attitude toward the team formed with the VP), and local muscle fatigue of exercisers. The research findings provide a certain theoretical basis for VP-derived human–computer interactions.
3. Results
The descriptive statistics results of the exercisers’ EL, subjective exercise enjoyment, attitude toward the athletic team formed with the VP, and local muscle fatigue under different experimental conditions are shown in
Table 3.
3.1. EL
EL refers to the cumulative squat count at the time of exercise termination when the rating of subjective perceived exertion reached 15 during the VP-accompanied exercise. The RM ANOVA results of the VP’s interactive features regarding EL are shown in
Table 4. The main effect of the presence or absence of the VP’s BM on EL was significant (
p < 0.001), and the number of squats with the VP’s BM (M = 78.01) was markedly higher than that without the VP’s BM (M = 73.76). The main effect of the presence or absence of SP on EL was significant (
p < 0.001), with an evidently greater squat count without the VP’s SP (M = 77.23) than with the VP’s SP (M = 74.54). The presence or absence of the VP’s EG did not have a significant main effect on EL (
p = 0.106), suggesting that the exercisers’ EL did not differ substantially with or without the VP’s EG.
The interaction effect between the VP’s BM and EG on EL was significant (p = 0.021). The simple-effect analysis revealed that under the condition of EG, the presence or absence of BM showed a significant difference, F(1, 39) = 26.328, p < 0.001, with a noticeably higher number of squats with the VP’s BM (M = 77.14) than without the VP’s BM (M = 73.67). Under the condition of no EG, the presence or absence of BM had a significant difference, F(1, 39) = 55.046, p < 0.001, with a greater squat count with the VP’s BM (M = 78.87) than without the VP’s BM (M = 73.84). In addition, under the condition of BM, the presence or absence of EG had a significant difference, F(1, 39) = 6.535, p = 0.011, with a greater squat count with the VP’s EG (M = 78.87) than without the VP’s EG (M = 77.14). However, when no VP BM was present, EG had no significant effect, p = 0.790.
The interaction effect between the VP’s BM and SP on EL was significant (p < 0.001). The simple-effect analysis showed that under the condition of VP BM, there was a significant difference with or without SP, F(1, 39) = 4.526, p = 0.034, with a manifestly higher number of squats without the VP’s SP (M = 78.73) than with the VP’s SP (M = 77.29). When without BM, the presence or absence of SP had a significant difference, F(1, 39) = 33.821, p < 0.001, with a manifestly higher number of squats without the VP’s SP (M = 75.73) than with the VP’s SP (M = 71.79). There was a significant difference between with BM and without BM under the condition of SP, F(1, 39) = 65.921, p < 0.001, with a remarkably greater squat count with the VP’s BM (M = 77.29) than without the VP’s BM (M = 71.79). When there was no SP, the presence or absence of BM had a significant difference, F(1, 39) = 19.635, p < 0.001, and the number of squats was noticeably higher with the VP’s BM (M = 78.73) than without the VP’s BM (M = 75.73).
The interaction effect between the VP’s EG and SP on EL was significant (p < 0.001). The simple-effect analysis showed that in the absence of EG, there was a significant difference with or without SP, F(1, 39) = 95.957, p < 0.001, with an evidently greater squat count without the VP’s SP (M = 79.68) than with the VP’s SP (M = 73.04). In the presence of EG, the presence or absence of SP showed no significant differences, F(1, 39) = 3.432, p = 0.065. Without SP, the presence of EG had a significant effect, F(1, 39) = 52.338, p < 0.001, with the number of squats obviously higher without the VP’s EG (M = 79.68) than with the VP’s EG (M = 74.78). With SP, a significant difference existed with and without EG, F(1, 39) = 19.483, p < 0.001, with the number of squats with the VP’s EG (M = 76.03) prominently higher than without the VP’s EG (M = 73.04).
The interaction effect between the VP’s three variables, BM, EG, and SP, was not significant (p = 0.167).
3.2. Exercise Perception
3.2.1. Subjective Exercise Enjoyment
The subjective exercise enjoyment scale is an 8-item, 7-point scale. The lower the score is, the less the participant enjoys the exercise. RM ANOVA was performed for the subjective exercise enjoyment scores under different experimental conditions (shown in
Table 5). Our results showed that the VP’s BM variable had a significant main effect on the individual’s subjective exercise enjoyment (
p < 0.001), and the exercise enjoyment score with BM (M = 3.97) was significantly higher than that without BM (M = 3.49). Additionally, the VP’s EG variable displayed a significant main effect on the subjective exercise enjoyment (
p < 0.001), with a higher exercise enjoyment score with EG (M = 3.91) than without EG (M = 3.55). The main effect of the VP’s SP on subjective exercise enjoyment was not significant (
p = 0.243). The three variables’ double and triple interaction effects were insignificant (
p > 0.05).
3.2.2. Attitude toward the Sports Team Formed with VP
The RM ANOVA results for the exercisers’ attitude toward the athletic team formed with the VP are shown in
Table 6. The VP’s BM had a significant main effect on the attitude toward the athletic team formed with the VP (
p < 0.001), and the VP’s BM enhanced the exercisers’ attitude toward the sports team formed with the VP. On the contrary, the VP’s EG (
p = 0.837) and SP (
p = 0.089) had a non-significant main effect on the attitude toward the athletic team formed with the VP.
The interaction effect between the VP’s BM and SP was significant (p < 0.001). According to the simple-effect analysis, in the absence of BM, there was no significant difference with or without SP, F(1, 39) = 3.292, p = 0.071. However, in the presence of BM, there was a significant difference with or without the VP’s SP, F(1, 39) = 13.835, p < 0.001, and the exercisers’ attitude toward the athletic team formed with the VP was evidently superior without SP (M = 4.90) compared to that with SP (M = 4.38). Additionally, in the absence of SP, there was a significant difference with or without the VP’s BM, F(1, 39) = 90.544, p < 0.001, and the exercisers’ attitude toward the athletic team formed with the VP was evidently superior with BM (M = 4.90) compared to that without BM (M = 3.58). In the presence of SP, there was a significant difference with or without BM, F(1, 39) = 15.853, p < 0.001, and the exercisers’ attitude score with the VP’s BM (M = 4.38) was significantly higher than that without the VP’s BM (M = 3.83).
3.3. Local Muscle Fatigue
In this experiment, RF was selected as the muscle to be tested, and the difference between the mean MF value within 30 s before exercising and the mean MF value within 30 s after exercising was used as an indicator of local muscle fatigue. The MF difference data of the participants’ RF under eight experimental conditions (shown in
Figure 7) indicated that the local muscles of the exercisers all showed certain degrees of fatigue under each experimental condition, and the changes in the sEMG data of RF were relatively consistent. In the absence of BM, the MF difference without SP was slightly higher than that with SP; in the presence of BM, the MF difference without SP was also slightly higher than that with SP.
The RM ANOVA results for the interactive features of the VP on local muscle fatigue degree are shown in
Table 7. Our results showed that the main and interaction effects of the VP’s three interactive feature variables, BM, EG, and SP, on the local muscle fatigue of the exercisers were not significant (
p > 0.05).
4. Discussion
4.1. The Effect of the Interactive Features of VP on EL
This study used the number of squats when the participant’s RPE rating reached 15 to characterize their EL. Our study results showed that the BM of the VP had a significant main effect on EL (
p < 0.001), and the EL with BM was always higher than that without BM, consistent with previous studies. Automatic imitation of observed behavior is a powerful mechanism [
51]. With BM, the exerciser tends to mimic the BM of the VP and adapt to its rhythm, and this natural social behavioral adaptation process results in the dynamic synchronization of BM between the exerciser and the VP. The dynamic synchronization of behavior between the interaction partners, namely interpersonal synchrony (IPS), is a fundamental behavioral and physiological mechanism [
52], which exhibited in this experiment as the exerciser and VP presenting the same BM simultaneously. IPS likely strengthened the social bond between the exerciser and the VP and was thereby conducive to eliciting the key mechanism of the Köhler effect [
1], promoting the exerciser’s EL.
The main effect of the VP’s EG on EL was not significant (
p = 0.106), and the presence of EG did not substantially raise EL. These results do not parallel existing research findings. The drive theory in the social facilitation effect deems that the presence of a VP may improve the performance of tasks that the participants are good at [
53]. This is because the existence of others increases the level of interest and promotes the individual’s propensity and performance to complete the task. However, when confronted with a complex and novel task, the instinctive responses that people show directly are not necessarily correct or optimal, and at this time, the drive theory will inhibit the task performance of the individual. All participants in this experiment had no regular fitness habits and did not have scientific theory or method guidance. Bodyweight squats may have been difficult for them. Consequently, the drive effect resulted from EG-inhibited EL. Moreover, conflicting conclusions have emerged in the gaming field regarding drive theory. For instance, some studies have suggested that social presence could boost gaming performance in digital game players [
54], while others indicated that this facilitating effect was limited to less challenging games [
55]. However, some studies also showed that regardless of the game’s difficulty, social observation did not affect the players’ gaming performance [
56]. The conflicting results regarding the social facilitation effect produced in the gaming field have been attributed to the “fake reality” created by games, i.e., that real-world theories such as the social facilitation effect may not be effective in virtual reality [
56]. In the present study, in the post-experimental interview, the participants expressed that the presence of the VP made the workout feel more like a game; thus, the standard theory may not be valid in this experiment. In addition, this result may be related to the expression of the VP. Some subjects indicated in post-experimental interviews that the VP’s expressionless stare made them feel nervous. Therefore, we hypothesized that the relatively serious expression of the VP may cause psychological stress to the exerciser, which may lead to a decrease in motivation to exercise, but this hypothesis requires further research.
The main effect of the VP’s SP on EL was significant (
p < 0.001), and the EL without SP was significantly higher than that with SP, which is inconsistent with most studies’ conclusions. There are two possible explanations for this. The first is due to the competitiveness difference of gender, and intrasexual competition theory posits that men are more competitive when facing men [
57]. The participants and the VP in this experiment were all male; thus, the participants may have been more competitive in the face of the VP in the early stage of exercise. However, as the amount of exercise increased, the participants may instead have felt frustrated because their SP was always lagging behind [
12], leading to the suppression of the incentive effect of the VP on EL. Another reason is the lack of feedback on the exerciser’s performance. It is known that varying the capacity gap between the VP and exerciser with the exerciser’s effort could help the exerciser set goals and match up to the VP [
58]. Such informative performance feedback is a potential key factor in eliciting the motivation gain mechanism of the Köhler effect, while insufficient feedback might affect the Köhler mechanism [
24]. In this experiment, the VP’s athletic ability was always moderately ahead of that of the exerciser, and the exerciser may not have perceived how and if his EL was instrumental to team achievement, potentially repressing the motivation gain of the Köhler effect and leading to the different results in this study. Furthermore, the participants stated in the post-experimental interviews that if there was some feedback as the EL increased, such as surpassing the VP, there might be a sense of accomplishment that motivated the exerciser to work harder. These findings suggest that instant information feedback is a potential key to motivating EL in a long-term and dynamic process, such as VP-accompanied sports. When the user’s EL and persistence change, the capacity gap with the VP also changes accordingly, which can effectively motivate the exercisers to improve their benchmark targets and elevate EL.
The interaction effect of the VP’s BM and EG on EL was significant (
p = 0.021). With or without EG, the difference was always significant with or without BM (
p < 0.001). These findings suggest that the prominent effect of BM on EL was not influenced by EG. The interaction effect of the VP’s BM and SP on EL was significant (
p < 0.001). In the absence of BM, SP made a significant difference (
p < 0.001); and in the presence of BM, there was a significant difference with or without SP (
p = 0.034). The interaction effect of the VP’s EG and SP on EL was significant (
p < 0.001). Without EG, SP made a significant difference (
p < 0.001), whereas with EG, there was no significant difference with or without SP (
p = 0.065). The effect of the VP’s SP on EL was influenced by the two variables BM and EG, which can be explained by the relevant research of cognitive load theory (CLT) and the split-attention effect [
59,
60]. The exercisers had limited cognitive resources and could only focus on a portion of the incoming information at any given time. The BM and EG of the VP were more palpable interactive features; thus, in the presence of BM or EG, the attention of the exercisers was more focused on these two features and less on SP, thus weakening the impact of SP.
4.2. The Effect of the Interactive Features of the VP on Exercise Perception
The main effect of the VP’s BM on subjective exercise enjoyment was significant (
p < 0.001), and the main effect on the attitude toward the athletic team formed with the VP was significant (
p < 0.001), with the presence of BM markedly improving the individual’s subjective exercise enjoyment and attitude toward the athletic team formed with the VP. These results are consistent with existing studies, and according to the media equation theory (CASA), the interactions between humans and computers in anthropomorphic states are essentially social [
40]. Therefore, in the presence of BM, the exercisers tended to synchronize movements with the VP, and this natural process of social behavior adaptation could improve the exercisers’ trust in the VP and the VP’s likeability [
61], as well as the entitativity and rapport between them [
62]. The collaboration and rapport relationships between the exercisers and the VP enhanced social interactions. During the interactions with VPs, exercisers will establish important social relationships with the VP (e.g., keeping promises and treating the VP as a teammate) [
40], and the virtual interactions with the VP may also provide mental health benefits [
63].
The main effect of the VP’s EG on subjective exercise enjoyment was significant (
p < 0.001), and having EG evidently enhanced the exercise enjoyment of the exercisers. This outcome can be interpreted through the relevant research of social agency theory [
64]. As a typical social cue, the VP’s EG and body orientation may foster a sense of interaction between the VP and the exerciser [
65], thereby heightening exercise enjoyment. In the post-experimental interview, the participants indicated they paid more attention when the VP had EG. These findings agree with those obtained in some other studies; EG in human–computer interaction enhances the engagement and motivation of the user, helping direct the user’s attention to associated information [
66].
The interaction effect of the VP’s BM and SP on the attitude toward the athletic team formed with the VP was significant (p < 0.001). With BM, the VP’s SP could increase the exercisers’ attitude toward the athletic team formed with the VP. In the condition with BM, the VP’s SP significantly decreased the exercisers’ attitude toward the athletic team formed with the VP, while in the absence of BM, the VP’s SP did not significantly influence the exercisers’ attitude toward the athletic team formed with the VP. The post-experimental interview revealed that with SP, the exercisers were more concerned about the gap between themselves and the VP’s SP. Thus, social comparison may have exerted a greater influence, and the exercisers tried to catch up with or surpass the VP, thereby being more inclined to consider themselves in competition with the VP, resulting in a significant difference in the effect of the VP’s SP on the attitude toward the athletic team formed with the VP in the presence of BM.
The exercise experience results indicate that the individual’s exercise enjoyment can be boosted through two interactive features, BM and EG, helping the exercisers produce more durable persistence. Moreover, BM notably enhances the exercisers’ attitude toward the sports team formed with the VP, and the formation and strengthening of social relations between exercisers and VPs will boost athletic performance through the Köhler effect [
3]. Therefore, in the VP sports motivation software design, longer or more personal interactions with exercisers can be created by virtue of the VP’s BM, such as customized actions or encouraging gestures, to increase the individuals’ exercise enjoyment and sense of identification with the VP, thereby augmenting EL.
4.3. The Effect of the Interactive Features of the VP on Local Muscle Fatigue
The interactive features of the VP can affect the exercisers’ subjective exercise enjoyment, while the increase in exercise enjoyment may distract individual perceptions of subjective effort and fatigue. Thus, this study measured the degree of local muscle fatigue through an EMG experiment combined with RPE to probe the impact of VP interactive features on the incentive for EL. Our results showed no significant effects of the VP’s BM, EG, and SP on the exercisers’ local muscle fatigue (p > 0.05), and the interaction effects between the three variables were also insignificant (p > 0.05). These findings suggest that the interactive features of the VP had no obvious influence on the exercisers’ changes in local muscle fatigue before and after a workout.
Combined with the EL data, in the presence of SP, some exercisers had a reduced number of squats when the RPE level reached 15, whereas the EMG data showed that their local muscle fatigue was lower than that when SP was absent. These findings suggest that the VP’s SP may have increased the exercisers’ subjective perceived exertion. In addition, when both BM and SP were present, some exercisers had intensified local muscle fatigue, and in the post-experimental interview, the participants indicated that this was because when BM was present, they still wanted to keep going despite their RPE rating reaching 15. These results suggest that the VP’s BM did not affect subjective perceived exertion but promoted the motivation of exercisers to continue the workout, thereby generating a motivational effect on EL. This finding agrees with previous studies in which it was found that being immersed in a pleasurable exergame distracts one from the perceptions of subjective effort and affects the exercisers’ motivation to persist in completing the tasks [
67,
68]. Therefore, this study infers that the VP’s interactive features may affect EL by influencing the exercisers’ subjective perceived exertion and exercise adherence.
However, there were still some deficiencies in the experimental results. For instance, the EL of the exercisers showed significant improvements under certain experimental conditions, and in the post-experimental interview, the participants indicated obvious soreness in lower limb muscles after exercising. However, this phenomenon was not measured in the experiment. The reasons for this outcome may be that during the workout, the participants used different ways of exerting force, or systemic muscle compensation occurred with the accumulating exercise amount and deepening of muscle fatigue during the exercise, which indirectly influenced the experimental results. These causes received confirmation in the post-experimental interviews with the participants. Some exercisers said they preferred using gluteus maximus muscle to compensate when squatting to reduce strain on the knee area. It is also speculated in this study that rest time may have affected these experimental results [
69]. In the experiment, the participants could not complete the MF static test without an interval after the exercise, and some local muscle fatigue may have been quickly relieved during the interval.
Overall, the results of the EMG experiment, to a certain extent, verified the purpose of this study; the interactive features of VPs may influence EL by affecting the exercisers’ subjective perceived exertion. However, the insufficient selection of indicators in this experiment design resulted in single experimental data, and measuring the EMG data of the rectus femoris muscle alone was inadequate. Assessing the degree of muscle fatigue using the MF difference before and after exercising had limitations. In the future, more physiological measurements are needed to accurately study muscle fatigue of exercisers.
4.4. Limitations and Future Research
Although the present study derived some conclusions on the effects of the different interactive features of VPs on EL, exercise perception (subjective exercise enjoyment and attitude toward the athletic team formed with the VP), and local muscle fatigue, the research still has limitations. First, the experiment in this study involved a single exercise, and follow-up studies can explore the motivational exercise effect of the VP’s interactive features on other sport types (such as running, planking, and walking). Second, the lack of female subjects is a huge limitation. In order to avoid the influence of gender factors on the experimental results, the participants were all young men. Future studies can use gender as an independent variable to discuss the impact of the VP’s interactive features on EL and can also extend the research cohort to all age groups, exploring more general conclusions. Third, the selection of detection indicators was lacking. It has been pointed out that self-efficacy is an important factor affecting EL; with the increase in EL, the self-efficacy of the exerciser may play a motivating role. Thus, the self-efficacy of exercisers before and after a workout deserves further examination. The effect of the VP’s facial expression on individual exercise performance also deserves to be studied. The present study chose the indicators, including the MF, subjective exercise enjoyment, and attitude toward the athletic team formed with the VP, and provided certain conclusions on which future research can be based.