Competition and cooperation with virtual players in an exergame

Methods

A cross-sectional within-subjects study was conducted to examine the effects of competition and cooperation using different representations of another player in an immersive virtual reality (VR) exergame. Specifically, the study examined exergame performance in three conditions: (1) solitary play in the exergame with no virtual player “Default Condition,” (2) a ghost condition whereby participants played the exergame while competing against a replay of their performance in the first condition “Ghost condition,” and (3) playing the exergame with a virtual trainer present “Trainer condition.” The main outcome variables were distance travelled on the Exercycle, calories expended on the Exercycle, and RPE. In addition, the study explored participants’ responses to self-report measures of enjoyment and motivation, following the completion of each condition.

A total of 22 individuals participated in the study. Three participants withdrew from the study due to suffering from discomfort related to the use of the Oculus Rift during the session. The remaining 19 (15 male, four female, mean age: 31.5, standard deviation: 9.2) were able to complete it. Informed consent was obtained from all participants, and the study was approved by the University of Auckland Human Participants Ethics Committee (reference number: 8450). The study took place between the 7th and the 21st of January, 2015.

All participants completed the control condition first in order to provide data to be used during the Ghost condition. The order of the Ghost and Trainer conditions were then randomized so as to counterbalance potential order effects.

Design

For this research, we extended an existing exergame described in Shaw et al. (2015). This exergame was chosen as it elicited high intensity exercise from the users, and was rated by the users as enjoyable. In this exergame, the user cycles along a procedurally generated track, avoiding obstacles and collecting bonuses, in an effort to maximise their score. The speed at which the user moves is governed by the rate at which they pedal on the exercycle. A 3D camera tracks their movements, allowing them to steer by leaning from side to side. The game is presented to the user via an Oculus Rift Head Mounted Display (HMD), providing them with an immersive experience.

We extended this exergame, adding a replay system and a simple virtual trainer system. The base gameplay was also slightly modified, changing obstacles to slow the player and penalize their score, rather than causing them to replay a section. This was necessary in order to avoid divergence between the ghost replay system described below, and the user’s current play session.

Figure 1 shows a screenshot of the exergame, and illustrates some of the gameplay elements.

The exergame allows for playback of a participant’s previous attempts through a “ghost racer” system, in which the participant sees a non-interactive replay of the past attempt on the track as they play. This offers encouragement to exercise harder in order to beat their previous attempt. When users are lagging behind their “ghost,” they are also able to see points where their previous run failed to avoid obstacles, and thus they may be able to adapt their play to avoid more obstacles.

The player’s ghost has the same appearance as the trainers (shown in Fig. 2): a simplified figure on a bike. When close to the player, the ghost and trainers are semi-transparent, increasing in opacity as they move further away. This is to prevent them from blocking the players view of obstacles or bonuses and becoming a potential source of frustration.

The simple virtual trainer system behaves in a similar fashion to the ghost system. When the player begins to move, another ‘player’ appears and travels along the track with them. However, rather than showing previous behaviour, the trainer attempts to show optimal behaviour, both in terms of gameplay and exercise. With regard to gameplay, the trainer chooses an optimal path through the track, avoiding all obstacles. With regard to exercise, the trainer adjusts its speed to guide the user towards an ideal exercise heart rate, as explained below.

First, the current heart rate of the user, as measured by the handlebar sensors, is used to estimate a relative heart rate, i.e., a percentage of the user’s expected maximum heart rate based on their age. This is done using Tanaka, Monahan & Seals (2001) regression equation 208 − 0.7 × age. The trainer attempts to set a speed suitable for keeping the user’s heart rate at the level associated with moderate to vigorous exertion, i.e., 64–90% of their expected maximum heart rate (Garber et al., 2011). If the user’s heart rate is below 64% (“low heart rate”), the trainer increases its speed, requiring the user to work harder to catch up. If their heart rate exceeds 90% of their maximum (“high heart rate”), it decreases its speed, allowing them to exert less effort to keep pace. While the user’s heart rate is in the target zone (“average heart rate”), the trainer stays a short distance in front of the user providing a target to follow in order to motivate them.

Procedure

Participants completed a pre-test questionnaire to provide general demographic data: their age, gender, and baseline self-report measures of the typical number of hours spent exercising and playing video games each week. Participants were then given a written outline of the test procedure and conditions, and written instructions on how to play the exergame. They were assisted with adjusting the exercycle and motion tracking equipment.

Participants first completed the ten-minute “Control” condition, followed by the “Ghost” and “Trainer” conditions in a counterbalanced order, separated by five-minute breaks. In the control condition, the participant’s play attempt was recorded to provide the ghost for the later Ghost condition. During the break following each condition, participants were given a sheet listing the different exertion levels on the Borg RPE scale (Borg, 1982), and asked to rate their level of exertion. They were also given a post-condition questionnaire in which they were asked to rate how enjoyable and motivating they found the condition, and were invited to give feedback about the Ghost and Trainer systems, and about the exergame in general.

Measures

Data analyses

The normality and sphericity assumptions of repeated measures analysis of variance (RM-ANOVA) were tested with the Shapiro-Wilk test and Mauchly’s sphericity test, respectively. With a p-value threshold of 0.05, the normality assumption holds for the measures of distance travelled, calories expended, and RPE, but does not hold for the measures of enjoyment and motivation. The sphericity assumption holds for all measures except motivation.

RM-ANOVA with post-hoc Bonferroni tests were conducted to examine the effects of the Control, Ghost, and Trainer conditions on distances travelled, calories expended, and RPE.

Due to the non-normally distributed data, the effects of the three conditions on enjoyment and motivation were examined with Friedman tests.

Pearson correlation analyses were used to examine the association between the participant information gathered during the pre-test, and the measures listed above.

Results and discussion

Table 1 shows the means and standard deviations of the various measures across the three conditions.

The results of a RM-ANOVA showed that there is a significant main effect Condition on distance travelled (F = 15.28, p < 0.001). The results of the post-hoc Bonferroni tests showed that distance travelled in the Trainer condition was significantly lower than in the Control condition (p = 0.001), and the Ghost condition (p < 0.001). There was no significant difference between the Control and Ghost conditions.

There was also a significant main effect Condition on calories expended (F = 4.64, p = 0.16). However, the results of the post-hoc Bonferroni tests did not show any pairwise significances.

There was a significant main effect of Condition on RPE (F = 3.79, p = 0.032). The results of the post-hoc Bonferroni tests showed that the RPE rating for the Ghost condition is significantly higher than for the Control condition (p = 0.01). There were no significant difference between the Trainer condition and either of the other two conditions.

The results of a Friedman test showed a significant main effect Condition on enjoyment across the three conditions (p = 0.016). The Ghost condition was significantly more enjoyable than the Control and Trainer conditions.

The results of a Friedman test showed no significant main effect Condition on motivation across the three conditions (p = 0.370).

The results of a Pearson correlation showed that enjoyment of a condition, and level of motivation in that condition have no significant correlation with distance travelled in the condition.

The use of player recordings of past performance to encourage self-competition shows significant promise to encourage users to exercise via an exergame, particularly if they enjoy competition. Verbal and qualitative feedback from the participants indicated that being able to see and beat their previous attempt was highly enjoyable during the Ghost condition. This study failed to show benefits for the use of a multiplayer-style virtual trainer system, however that may be due to flaws in the trainer system discussed further below.

It is not too surprising to see that the Ghost condition did not encourage players to exercise significantly harder than in the Control condition. Unless turning around to look directly backwards, players would only see their ghost when it was ahead. Thus they would only receive motivation from the ghost to speed up when doing worse than it.

Attitudes towards the trainer system were less positive. Overall, it was not significantly more enjoyable than playing in the absence of another player. From verbal and written feedback, several of the participants who did not find the trainer system motivating found that it reduced their enjoyment of the exergame, citing unrealistic behaviour: “The trainer system moved strange.” We suspect this may be related to the framing of the trainer system. While the ghost system was clearly competitive, the trainer system had no particular framing as either competitive or cooperative. If the trainer was ahead, it would show participants an optimal performance, but participants were able to push themselves above the target heart rate and pull ahead; “competing” with it.

The trainer system was extremely effective at avoiding obstacles, and often navigated through obstacles with superhuman dexterity. When this occurred, participants tended to react negatively, stating that they felt that the trainer was “cheating,” and was not helping them as it was not showing them an optimal path that they were capable of following.

However, it should be noted that while the Ghost condition was better received than both the Control and Trainer conditions, the overall participant response to all three of the conditions was generally positive, with the mean enjoyability and motivation ratings still being high. The exergame in general was regarded as enjoyable and motivating, and the Trainer system did not detract from that.

Methods

A cross-sectional within-subjects study was conducted to examine the effects of competition and cooperation with a virtual trainer in the exergame environment. The effects of: (1) solitary play in the exergame with no virtual player (Default Condition), (2) competition whereby participants played the exergame while competing against a virtual trainer with a competitive behaviour profile, and (3) cooperation whereby participants played the exergame working with a virtual trainer with a cooperative behaviour profile on distance travelled and calories expended while cycling. Participants were recruited through open advertisement. A total of 28 individuals participated in the study, of which 25 (21 male, four female, mean age: 24.3, standard deviation: 9.2) were able to complete it. Informed consent was obtained from all participants, and the study was approved by the University of Auckland Human Participants Ethics Committee (reference number: 8450). The study took place between the 20th of July and the 21st of September, 2015.

Design

The virtual trainer system described in Study 1 was fairly limited in its capability to interact with the user. For the second study, we designed and implemented a more advanced virtual trainer (shown in Fig. 2) based on the evaluation of the initial trainer design. The full design of the advanced trainer system is detailed in Shaw et al. (2016).

As research discussed earlier in this paper indicates, competition as part of an exergame can affect different users very differently depending on how competitive they are. The behaviour of the first trainer system was not clearly framed as either competitive or cooperative. The advanced trainer system was designed to be customizable for either competition or cooperation in order to appeal to different personality types. In order to do that, the advanced trainer implements two behaviour profiles: a competitive profile and a cooperative one. While the competitive trainer profile is programmed to challenge and race against the player, the cooperative trainer profile attempts to help the player achieve a higher score.

Similar to the previous trainer, the advanced trainer always chooses a path that is close to optimal for scoring points and attempts to avoid obstacles. The trainer looks ahead to avoid obstacles in the distance. For example, if there is an obstacle in the centre of the track, and beyond that is one on the left side, the trainer will choose to go right when avoiding the central obstacle. As such, a user can follow the trainer and potentially achieve a higher score. However, the trainer only looks 40 m ahead when planning its path. Beyond this point the flat nature of the track (visible in Figs. 1 and 2) and resolution of the Oculus Rift make it difficult for a human to clearly make out obstacles. Limiting the trainer’s perception to this distance helps to keep its navigational ability on par with that of a human.

A moderate number of participants in Study 1 (6 of 19) mentioned the unrealistic agility of the first trainer system as something that they did not like about it. For this reason, the lateral movement speed of the advanced trainer was capped at the maximum achievable via the motion controls used by the human player.

Competitive trainer

The advanced trainer modifies its behaviour based on the heart rate of the user, considering the same low, average, and high heart rate zones as the simple trainer (see Fig. 3). For the competitive trainer profile, when the player’s heart rate is too low, the trainer’s speed will increase up to 1.3 times that of the player. When in the average heart rate zone, the trainer’s speed will approximately match that of the player. And when the player is in the high heart rate zone, the trainer’s speed will drop down to 0.7 times that of the player.

The speed of the trainer also takes into consideration the distance from the player. If the player spends an extended time in the low heart rate zone, it is important that the trainer does not get too far ahead, otherwise it may be demotivating, or at least no longer motivating if the trainer has moved out of the player’s view. Similarly, it is important that the trainer does not fall too far behind if the player is spending an extended period of time above the target heart rate zone, otherwise the benefit of the trainer will be lost even if the player’s heart rate falls back into the target zone. Because the game is presented via HMD, clamping the trainer’s distance means that the player is always able to look over their shoulder and see the trainer following them.

While in the target zone, the speed variation means that the trainer behaves as a human player of similar abilities, in that it occasionally pulls slightly ahead and occasionally falls slightly behind. This means that the user is always being made aware of the presence of the trainer and is encouraged to compete and stay ahead. As the user tends towards the upper end of the target heart rate zone, the trainer spends more time behind the user, while at the lower end of the zone it spends more time ahead of the user.

Cooperative trainer

The cooperative trainer is designed to cooperate with the player, providing assistance to the player in achieving the goals of maximising their score and maintaining an ideal target heart-rate. The cooperative trainer always gives the player a target to focus on, much like a lead cyclist in real-world group cycling activities. Following the cooperative trainer helps a player to follow a near optimal path along the track, avoiding all obstacles. Within this game, cooperation is one-directional: the trainer cooperates with the player, but the player is not providing meaningful assistance to the trainer.

The cooperative trainer uses a similar heart rate based mechanism to the competitive trainer, but it always sits in front of the player, regardless of speed. If the user is in the low heart rate zone, the trainer maintains a position well ahead of but clearly visible to the player. In the high heart rate zone the trainer stays only barely ahead of the player. And in the average heart rate zone, the trainer varies its position in a similar fashion to the competitive trainer, but within the bounds given by its positions when the player is in the high or low heart rate zones (see Fig. 4).

Procedure

Participants completed a pre-test questionnaire to provide general demographic data: their age, gender, and baseline self-report measures of the typical number of hours spent exercising and playing video games each week. As part of the pre-experiment questionnaire, participants also filled out the Sport Orientation Questionnaire (SOQ) (Gill & Deeter, 1988) and the Task and Ego Orientation in Sport Questionnaire (TEOSQ) (Duda, 1989). These are validated and commonly used questionnaires that provide five personality metrics related to competitiveness in sporting activities: competitiveness, goal orientation, and winning orientation from the SOQ, and task and ego orientation from the TEOSQ.

Following the questionnaires, participants were then given a written outline of the test procedure and conditions, and written instructions on how to play the exergame. The investigator assisted participants with adjusting the exercycle and calibrated the motion-tracking equipment.

Participants completed the three conditions in a counterbalanced order determined by the method of Latin Squares. Prior to each condition, the participants were given a verbal explanation of that condition, including an explanation of the trainer’s behaviour. The three conditions were separated by five-minute breaks. During the break following each condition, participants were given a post-condition questionnaire in which they were asked to rate how enjoyable and motivating they found the condition, and were invited to give general feedback about the trainer (where applicable) and the exergame in general.

Measures

Data analyses

The normality and sphericity assumptions of RM-ANOVA were tested with the Shapiro-Wilk test and Mauchly’s sphericity test, respectively. With a p-value threshold of 0.05, the normality assumption holds for the distance and calories measures, but does not hold for the enjoyment and motivation measures. The sphericity assumption holds for all measures except motivation.

RM-ANOVA with post-hoc Bonferroni tests were conducted to examine the effects of the Control, Competitive, and Cooperative conditions on distances travelled and calories expended.

Due to the non-normally distributed data, the effects of the three conditions on enjoyment and motivation were examined with Friedman tests.

Pearson correlation analyses were used to examine the association between the participant information gathered pre-test, and the measures listed above.

Results and discussion

Table 2 shows the means and standard deviations of the various measures across the three conditions.

The results of a RM-ANOVA showed that there was a significant main effect Condition on distance travelled (F = 7.4, p = 0.002). The results of the post-hoc Bonferroni-corrected tests showed that distance travelled in the Competitive condition was significantly higher than in the Control condition (p = 0.047), and the Cooperative condition (p = 0.006). There was no significant difference between the Control and Cooperative conditions.

There was a significant main effect Condition on calories expended (F = 6.63, p = 0.003). The results of the post-hoc Bonferroni tests showed that calorific expenditure in the Competitive condition was significantly higher than in the Control condition (p = 0.022), and the Cooperative condition (p = 0.011). There was no significant difference between the Control and Cooperative conditions.

The results of a Friedman test did not show a significant difference in enjoyment across the three conditions (p = 0.756).

The results of a Friedman test showed a significant difference in motivation across the three conditions (p = 0.027). The Competitive condition was significantly more motivating than the Control and Cooperative conditions.

Enjoyment of a condition, and level of motivation in that condition showed no significant correlation with distance travelled in the condition.

Task orientation as measured by the TEOSQ does not show any significant correlation with any other measurement.

Ego orientation as measured by the TEOSQ shows a moderate positive correlation with distance travelled in the baseline condition (r = 0.41, p < 0.05), but no significant correlation with distance in the other conditions, or with calories burned in any of the conditions. It shows a moderate negative correlation with motivation in the cooperative condition (r = −0.41, p < 0.05), but no significant correlation with motivation in the other two conditions. There was no significant correlation between ego orientation and enjoyment of any of the conditions.

None of the SOQ measurements showed any significant correlation with distance travelled or calories burned in any of the conditions. They also failed to show any significant correlation with reported enjoyment or motivation in any of the conditions.

Time spent on regular exercise and time spent playing video games do not appear to be correlated with the personality metrics measured by the SOQ and TEOSQ. No significant correlations are shown between these lifestyle factors and the personality traits. Surprisingly, for the participants in our study, there was also no significant correlation between time spent exercising and exercise performance in the exergame (distance travelled and calories burned measures). However, time spent on regular exercise did show a moderate negative correlation with rated enjoyment of the baseline condition (r = −0.41, p < 0.05).

Unsurprisingly, enjoyment and motivation for each of the conditions were closely related, with a moderate positive correlation between enjoyment and motivation of the default condition (r = 0.49, p <0.05), and the competitive condition (r = 0.46, p < 0.05). This was most noticeable in the cooperative condition, with a strong positive correlation between enjoyment and motivation (r = 0.77, p < 0.001).

The competitive trainer provided a fairly interactive experience for the participants. Regardless of whether they were ahead of the trainer or behind it, the variations in the trainer’s speed caused the race to always appear to be in a situation where the lead could be taken by either player. The cooperative trainer however, seemed less interactive. While its behaviour was more strictly defined as cooperative than that of the trainer in Study 1, simply receiving assistance the trainer is an experience of limited interactivity. This is reflected in our results with regard to distance travelled, calories burned, and the motivation rating.

It is interesting to note that the cooperative trainer did not prove to be particularly more effective for non-competitive individuals. This may be because overall, it provided only limited cooperative gameplay elements. Cooperation with the trainer is one directional. While the trainer can help the player by showing them an ideal path and giving them a target to focus on, the player cannot affect the trainer beyond their heart rate changing its speed. Thus players can get the feeling of being helped, but not of helping. The moderate negative correlation between ego orientation and motivation when playing with the cooperative trainer is interesting. In this case, it may be because the non-ego oriented players found receiving assistance with their gameplay to be motivating.

Our results show an interesting contrast with the findings of Song et al. (2010). Like their results, our results do not show non-competitive individuals performing worse in the competitive condition. However, their results showed reduced enjoyment and motivation for non-competitive individuals in a competitive experience. Our results, however, do not show that. This may be because in our case, the competition is against a virtual trainer, rather than another human. Further, in our study the player’s opponent is designed to match their fitness level, thus the competition is always fair. These factors likely reduce the negative aspects of a competitive experience for non-competitive individuals.

Verbal and written feedback about the trainers’ behaviour was consistent with that given in Study 1: when the trainers avoided obstacles in a fashion that the participants perceived as inhuman, the participants reacted negatively. Despite the fact that the agility of the trainer was reduced to a human-like level in Study 2, a Chi-Square test showed no significant difference in the number of participants mentioning “unrealistic” or “cheating” movements in their open feedback (Study 1 N: 6, Study 2 N: 9, p = 0.76). Player perceptions of the behaviour of virtual trainers and players would be an interesting future research area.