Comparison of response behavior in karate kumite between real world and virtual reality

Virtual reality is increasingly applied to support physical training and improve athletes’ performance in sports. Nevertheless, there is a research deficit in that, especially in martial arts, it has not yet been shown to what extent the response behavior of athletes in virtual reality is the same as in the real world. If this can be confirmed, a transfer of VR-adapted skills to RW can be expected and sports training in a virtual environment can be applied to improve sports performance. Since the response behavior is essential for many sports, this study compares it in karate kumite to the competition-important attack (Kizami-Zuki) of a real and a virtual opponent. Experienced karate athletes wore a head-mounted display and were asked to respond quickly and efficiently to 22 karate attacks, of which eight were Kizami Zuki’s attacks. Using a video-based movement assessment, karate experts quantified the response behavior with the parameters ‘time for response’, ‘response quality’ and ‘kind of response’. Results show no significant differences in ‘time for response’ and ‘kind of response’ between both conditions (virtual reality vs. real world). Only the ‘response quality’ was rated better in real world than in virtual reality. It is concluded that the ‘time of response’ and ‘kind of response’ for karate kumite athletes in virtual reality are similar to that in the real world.


Introduction
In karate kumite as in other martial sports, quick and vigorous attacks require quick responses.Therefore, response behavior, cognitive skills, perception, and decision-making are important factors limiting performance in combat sports [1,2].An essential parameter for characterizing the response behavior in karate kumite and other combat sports is the response time, respectively, reaction time.In this context, reaction time is often defined as the period between the first appearance of the stimulus and the first reaction toward it [3].According to [4], a short reaction time and quick anticipation are essential for a good perception of the athlete to produce an appropriate response because it provides the athlete more time to prepare the motor response [5].
To analyze the reaction time in different martial sports, the athletes were often asked to respond to an attack by a dummy or to external lights [4,5].In this research area, reaction time, reaction velocity, and decision quality (accuracy) are the primary parameters used for analyzing the sport-specific response behavior [6].However, Schorer [7] considers that a short reaction time does not always represent a good performance.Athletes who must compensate motor deficits have to react faster than high-skilled athletes, indicated by a shorter motor response time.After the first recognition, they are waiting to decide how to respond until they receive all information about the opponent's movement (just-in-time-principle).From this, it can be concluded that not only the response time but also the movement technique is essential for an optimal response behavior.
In karate kumite training, as in other combat sports, response time and optimal reaction, e.g., a direct counterattack, are practiced.Another training option would be to use a training tool in virtual reality (VR).In sports science, VR has been used to investigate several skill-related capabilities such as attention, anticipation, and decision-making.In their systematic review [8], the authors find that most research into VR has been conducted on endurance sports.However, in the meantime, other studies exist which apply VR to improve motor behavior and athletes' performances [9,10], e.g., in baseball [11] and handball [12].Further studies also show that sport-specific response times could be shortened when using VR tools: wrestling [13], tennis [14], hockey [15], and karate [16].
VR technology has developed rapidly in the last few years.Today it is possible to provide the user with a first perspective in which the virtual environment changes according to their viewpoint [14], as well as visualizations of their bodies [17].These opportunities allow a realistic presentation of the virtual environment quite similar to natural conditions.Furthermore, VR can be used to investigate and improve sports performance under standardized and controllable conditions enabling training that is inpracticable in real world (RW) [18][19][20].In present days, VR also provides the possibility of standalone training without predefined time, location, or the need for training partners or coaches [21,22].There are also advantages of VR over RW due to the lower risk of injury, as there is no physical contact with other athletes (opponents or teammates) [20,22,23].
Despite all the reported benefits of VR training tools, there is still a deficit in that there is not yet sufficient evidence of transfer to RW [24].An essential requirement for this is that athletes behave similarly in VR as they would in RW conditions.Specifically for martial arts, it needs to be clarified to what extent the reaction behavior of the athletes in VR is comparable to that in RW.
The current study aims to compare the response behavior to the competition-relevant attack Kizami-Zuki (KZ) of high-skilled karate kumite athletes between RW and VR.The following parameters are used for analyzing response behavior:'time for response', 'response quality', and 'kind of response'.The analyzed parameters are compared in both conditions to answer the research question, whether the athletes' response behavior is comparable between VR and RW.In line with studies on the behavior of goalkeepers [11,12], dart throwers [25], karate kumite athletes only in a VR setting [26], it is assumed that there are no significant differences in all parameters of the response behavior of karate kumite athletes between RW and VR.

Participants
Twenty-seven healthy athletes (17 male, 10 female, age: 17.4 ± 3.52 years, 5th Kyu-1st Dan, shotokan style with noncontact, experience in karate kumite: 10 ± 3.5 years) participated in the study.The participants were selected by the coaches, had to have at least 2 years of national competition experience and could not have any acute injuries or illnesses that would make normal karate training unacceptable.They are members of the DJKB (German Japan Karate Association) or of the DKV (German Karate Association).Participants reported normal or corrected-to-normal vision and had no problems with the stereoscopic vision, which was tested with the original stereo fly stereotest to identify vision problems (Titmus Stereoacuity Test by EYESFIRST ® ).Each participant (or their legal guardian) provided written informed consent to the conduction of the study.The study was part of a DFG-project (WI 1456/22-1), and therefore, obtained ethical approval (132/16) from the author's university.

Experimental procedure
At the beginning of the study, the participants performed two sub-tests of sport-unspecific reaction time from the Vienna test system (Schuhfried, Vienna, Austria [27]): S1 (simple reaction test to one visual stimulus) and S4 (recognition reaction test to one visual stimulus with visual and auditory distractors).The reaction time (RT) related to stimulus presentation and motor response time (MT) were measured.After the VR experience in each training session and test the participant completed the simulator sickness questionnaire (SSQ) [28] to analyze cybersickness symptoms.Schorer [7] stated for simulator sickness negligible symptoms for scores lower than 5, minimal (5 to 10), significant (10 to 15), concerning (15 to 20) and bad simulator (higher than 20).In our virtual environment, no participant had to discontinue the study.
The study focused on the response behavior to the karate attack Kizami-Zuki (KZ).KZ is an arm attack in which the front arm moves towards the opponent's head (see Fig. 1).It is one of the most often and successfully used attacks in international karate kumite competitions.
For analyzing karate-specific response behavior, the participants were asked to respond to 22 various randomized karate attacks each in RW and VR (Kizami-Zuki, Gyaku-Zuki jodan, Gyaku-Zuki chudan, and Mawashi-Geri chudan).The presentation of the attacks was randomized in four blocks, each with five to six attacks.The eight attacks considered for the analysis were randomized over the predefined blocks and mixed with the other attacks to avoid habituation.The blocks were interchanged between both conditions.The subjects were instructed to respond as they would in competition.

Real opponents and development of virtual opponents
The research design builds on the studies by Petri et al. [16,26].In RW, five real opponents (age: 19-30 years, 2nd Kyu-1st Dan) performed the attacks (Fig. 2a).To ensure realistic conditions, sex, age, and performance level of the opponents were assigned for each subject.Figure 2b, c show the third (3PP) and first-person perspective (1PP) for a better illustration of the VR setting.
For the attacks in VR, the karate attacks from five athletes with both international competition experience and coaching experience (age: 24-56 years, 1st-4th Dan) were recorded using motion capturing with 12 cameras (MX-13, Vicon, UK) and real time tracking (ART, Germany) [21].Athletes performed all attacks twice from a fixed position towards a fixed small cardboard target (small goal to create a virtual character with realistic movements.The attacks were saved as animations and transferred to humanoid avatars with different skin meshes (female, male) and a karate suit.Like in Petri et al. [16], a sports hall with a fight area represented the virtual environment.
The assignment of the virtual to the real opponent for each athlete was based on movement criteria evaluated by two karate experts (karate trainers with several years of experience in training as well as in competition): • Phase of reducing distance: lowering the body's center of gravity, preloading of the rear leg, pulling the rear leg, pushing the hip forward, front heel goes up, flying phase, landing after distance reduction, • Phase of arm pushing: pulling off the rear leg, pushing front knee forward, extension of the pushing arm, landing of front foot.
Furthermore, the temporal duration between tA1 (opponent starts to reduce the distance with a flight phase towards the athlete) and tA3 (end of the arm pushing forwards) of the attacks in VR was compared with those in RW.Using a t-test, these durations were compared in both conditions and no significant differences were found.

Parameters of the response behavior
The opponent's and athlete's movements were recorded synchronously by two high-speed cameras (Contemplas, Kempten, Germany, frequency rate 100 Hz) to analyze the attack and the response behavior.The following movement parameters were studied for each karate response: 'time for response', 'response quality', and 'kind of response'.'Time for response' was analyzed using specific time points during the opponent's and athlete's movement.Determining the parameter 'time for response' is problematic because it is unknown to which signal the athlete responds.Moreover, it can be assumed that each attack executed by the real opponent has different time durations and that these movement times also differ from the attacks of the virtual attacker.For temporal comparability, the following key points of KZ were determined: tA1: opponent starts to reduce the distance with a flight phase towards the athlete, tA2: end of the distance reduction and pulling of the back leg, tA3: end of the arm pushing forwards.The time point when the athlete starts his response was named by tRB.To achieve a normalization over time, the absolute value of 'time for response' (tRB-tA1) was set in relation to the 'duration of opponents's attack' (time from tA1 to tA3) by the quotient (tRB-tA1/ tA3-tA1) named relative value of 'time for response'.
The parameter 'response quality' was examined with a scoring system used before [16].This parameter indicates to what extent the athlete was able to prevent the attack.If so, it was also regarded what kind of technique was used by the athlete.The athlete got no points when the attack could not be prevented, a point when the attack could be prevented with a successful block or an evasive movement followed by a counterattack, or two points, when the attack was invalidated with a direct and successful attack of the athlete.
The parameter 'kind of response' was defined exclusively to find out whether the athlete responded with a blocking technique and possibly with a subsequent attack (1 point) or with a direct counter (2 points).The importance of this parameter lies in that it shows to what extent the athletes were able to start a direct counterattack without it being successful.

Data analysis
For both conditions (RW and VR), the response movements to the 8 attacks with KZ were rated by two experts using Kinovea (version 0.8.15), regarding the parameters of the response behavior (27 athletes * 8 attacks * 2 conditions = 432 videos).Interrater reliability using Cohen's kappa (κ) coefficient for all three parameters of the response behavior was satisfying (κ > 0.71) and can be rated as substantial [29].Statistic were conducted with SPSS (IBM, Germany, version 25).
Paired t-tests were used to determine the differences between VR and RW in 'time for response' and sign tests with regard to 'response quality' and 'kind of response'.The level of significance was set to α = 0.05.Effect sizes were calculated using Cohen's d and defined as d = 0.2 small effect, d = 0.5 moderate effect, and d = 0.8 large effect [30].

Results
The sport-specific response behavior in both conditions (RW vs. VR) is presented in Table 1.The t-tests show no significant differences for the parameter 'time for response', for both the absolute (p = 0.667) and relative (p = 0.308) values.The 'response quality' was significantly better in RW (p < 0.001) than in VR with a moderate effect.For 'kind of response' no significant differences (p = 0.23) were found.
The sport-unspecific reaction time was analyzed using the S1 and S4 (Vienna test system).The simple reaction to one visual stimulus (S1) lasted RT = 235.0± 35.02 ms and the motor response time MT = 122.68± 27.71 ms.The recognition reaction choice to several visual stimuli with an additional auditive distractor (S4) was for reaction time RT = 351.12± 59.09 ms and motor response time MT = 131.69± 35.77 ms.The reaction times (RT) of S1 and S4 correlate significantly with the 'time for response' in VR (S1: r = 0.524, p = 0.005; S4: r = 0.687, p = 0.000) and in RW (S1: r = 0.395, p = 0.041).Between RT in S4 and 'time for response' no significant correlations could be found in RW (r = 0.292, p = 0.140).

Discussion
The study generally proves that the response behavior of an athlete in karate kumite towards virtual karate attacks is similar to that in reality.For this, 27 experienced karate athletes were asked to react to the competitive attack KZ as quickly and as efficiently as possible to prevent it or, in a better case, to execute a direct counterattack.In RW, KZs were Through randomized trials including other attacks, athletes could not predict KZ.There were no differences between VR and RW in 'time for response'.To fix the problem of different temporal duration of the attack movement (tA3-tA1), the quotient of 'time for response' and the attack's temporal duration was calculated.No significant differences were found for either the attack movement or the relative 'time of response'.Accordingly, the time duration between starting the attack and the athlete's response was similar in both conditions.This result supports findings for handball goalkeepers [12] and karate athletes [16,26].In contrast to [16,26], the starting point of the attack was not related to the movement (extension) of the punching arm, but to the beginning of the whole-body movement from the karate-specific starting position kamae.
The current study design did not determine which signals the athletes were responding to.Similar statements were also made by [31] regarding the behavior of the goalkeeper in the penalty shootout.Future eye tracking studies in combination with the analysis of response behaviour could provide information about the signals the athletes were responding to.
Consistent with our findings of the sports-unspecific reaction time and motor response time (S1 and S4 of the Vienna test system), Petri et al. [26] and Gierzuk et al. [32] also reported comparable reaction times for combat sports athletes.The reaction time from the Vienna test system correlated positively with the time for response' in VR for simple reaction (S1) and recognition reaction choice (S4) and in RW only for S1.However, this correlation should not be overestimated since not only reaction time, but also decision quality and movement time play an important role in a sport-specific reaction [33].Furthermore, it is known that the experience and the performance level of the athletes influence these correlations, as previously shown with soccer and volleyball players [21].
Regarding the 'kind of response', no significant differences were found between RW and VR (p > 0.05).Nevertheless, the data show a tendency that in VR (M = 1.26), a direct counterattack was executed slightly more often than in RW (M = 1.18).This could be explained by the athletes not fearing direct contact with the opponent's fist.Regarding the 'kind of response' in relation to 'time for response', it could be assumed that the athletes responded faster when choosing the direct attack than choosing the blocking technique with the following counter attack because a direct attack needs to be executed faster than a blocking technique to reach the opponent first and to get the point.
Regarding the parameter 'response quality', significant differences with a moderate effect were found between both conditions (p < 0.05).That means that the athletes were not able to respond with comparable quality in both conditions.They could prevent less virtual attacks with a successful block or an evasive movement followed by a counter attack.One possible explanation is that the athletes had no previous experiences acting within a virtual environment.
Another explanation for better performance in RW than in VR could be the familiarity with real opponents [2,26].It can be expected that training in VR requires some habituation, as tactile and auditory perception could not be simulated in VR.Nevertheless, it can be assumed that VR training tools create a corresponding novelty effect among athletes, which could have a positive impact on training in VR.This positive impact on learning has already been shown with VR learning tools or VR serious games in students [34].
The results of SSQ should be interpreted carefully.Stanney et al. [35] noted that cybersickness differs from simulator sickness, which is evaluated by the SSQ.Based on a systematic review [36], the usage of the SSQ is less advisable since only one category of symptoms can be measured.Since in the current study, the athletes did not verbally report discomfort and did not have to stop testing, it was assumed that the actual cybersickness occurred only to a minor extent under VR conditions through an HMD display and marginally affected the athletes in their response tasks.
To sum up, there is only a significant difference between RW and VR for the 'response quality', but for the 'time for response' and 'kind of response', the response behavior is comparable.This result is in line with work from [28][29][30], as noted before.Although VR technology will be improved in future applications, it is essential to check to what extent the physical behavior of humans differs in future applications between both conditions.
There are limitations in the current study.First, the athletes, who were recorded to build the virtual avatars, did not execute the real attacks for the current study.Nevertheless, the attacks seem to be comparable in RW and VR, which was indicated by means of qualitative movement analysis and measurements of attack duration, small differences in movement cannot be ruled out.Future studies should use the same attackers in both conditions to avoid individual movement variations.Second, in contrast to the virtual opponent, the real one started the attack individually depending on the athlete's distance, which may have influenced their response behavior.

Conclusion
It is concluded that the 'time of response' and 'kind of response' for karate athletes in virtual reality are similar to that in the real world.However, differences have been shown in response quality.Participants could prevent less virtual attacks with a successful block or an evasive movement followed by a counterattack than in relation to real attacks.One of the reasons for this is that the athletes were not familiar with wearing an HMD.The study stands for future-realizable VR training tools to enhance response behavior, at least for karate kumite.In future, further analysis should include motion capturing to analyze the movement execution in detail, regarding ankle joints and weight distribution.Additionally, methods such as eye-tracking could be helpful to identify relevant signals of the opponents' movement initiation for the athlete's response behavior.

Fig. 1 Fig. 2
Fig. 1 Series of pictures demonstrating the karate attack Kizami-Zuki (KZ).a Starting position (kamae), b distance reduction towards the opponent, c pulling up the back leg, d pushing of the front hand to the head of the opponent

Table 1
The comparison of response behavior (M mean value, S standard deviation) between RW and VR (n = 27) and the indicated effect sizes (Cohen's d) In VR the same attacks were presented, standard deviation for the duration of the opponent's attack was zero Parameter of the response behavior RW VR z p d Time for response (M ± S) Time for response (tRB-tA1) (ms) (absolute) 363Response quality (points) 1.03 ± 0.26 0.65 ± 0.62 − 3.307 < 0.001 0.79 Kind of response (M ± S) Kind of response (points) 1.18 ± 0.24 1.26 ± 0.33 − 1.199 0.23 0.29 performed by a real opponent and in VR by a virtual one.