The construction of virtual reality task training system to enhancing emergency handling ability for pilots

1 Introduction

With new concepts and models of warfare represented by efficient cognition, decision-making, and collaboration, future air combat will become a competition based on the cognitive efficiency of pilots, which will determine the direction of the development of warfare (Heikki et al., 2023; Heikki et al., 2021). The psychological state, cognitive abilities, and, especially, ability of pilots to handle emergencies are crucial factors that influence the outcome of warfare (Bernd et al., 2017). Flight emergency handling capability refers to an individual’s ability to perform correct, timely, and accurate actions when unexpected or beyond-expected events are encountered during flight. Military flight emergency handling capability refers to the ability of military pilots to use effective measures to restore normal flight status when they face emergencies or abnormal flight conditions to ensure the safety of the aircraft and personnel (Rui et al., 2023). In emergency situations, pilots must endure significant psychological pressure while maintaining their ability to react quickly to avoid accidents, reduce damage to the fighter jet, and ensure their chances of survival (Grid et al., 2020; Ting et al., 2022). Therefore, the study of military pilots’ ability to handle emergencies is highly important for the development of the field of aviation psychology and is crucial for ensuring training safety and enhancing combat capabilities (Causse et al., 2025; Li et al., 2025; Zhang et al., 2024). Pilots’ training in emergency handling is a highly specialized and technology-driven field that involves the integrated application of various technologies and methods. In recent years, research based on physiological signals, such as electrocardiograms (ECGs), electroencephalograms (EEGs), and functional near-infrared spectroscopy (fNIRs), has provided effective tools for assessing pilots’ physiological states under stressful conditions (Alessandro and Mackie, 2024; Tao and Yonggang, 2023). By analyzing indicators such as heart rate variability (HRV), heart rate (HR), and galvanic skin response (GSR), the psychological stress levels of pilots can be accurately assessed. The use of multisource physiological signals to identify changes in pilots’ psychological states under complex flight conditions provides a new tool for the objective assessment of pilots’ psychological states and contributes to an accurate understanding of pilots’ psychological states during mission execution (Luzzani et al., 2024; Shao et al., 2019).

Several training platforms have been developed for aviation applications in recent years. Koskelo (2024) recorded ECG during simulated military flights and demonstrated that HRV can differentiate flight phases with varying cognitive workload levels, suggesting HRV is useful for assessing individual responses to cognitive load in simulator training. A growing body of work has examined the use of virtual reality (VR) flight simulation with physiological monitoring. A study compared VR versus desktop flight simulations and found that VR was associated with improved flight performance, while also observing EEG-based workload correlates (Van Weelden et al., 2024). Ji et al. (2023) constructed a VR experimental framework to analyze the relationship between EEG data and NASA-TLX subjective workload, providing reference information for pilot training system design. A recent systematic review and meta-analysis reported an overall positive effect of extended reality applications on pilot performance in flight simulator training, confirming the effectiveness of VR-based training (Somerville et al., 2025). In parallel, virtual reality exposure therapy (VRET) has been used for treating fear of flying (FoF). Ribé-Viñes et al. (2024) examined VR exposure with vibrotactile cues, finding that multisensory stimulation enhanced presence, though its effect on anxiety reduction was inconclusive. Rus-Calafell et al. (2013) reported that VR exposure for FoF resulted in less anxiety during real flights after treatment compared to imaginal exposure, supporting the desensitization potential of VR in aviation-related contexts.

Based on the above theoretical basis and research objectives, three hypotheses were proposed. Hypothesis 1: After VR emergency desensitization training, pilots’ emergency task success rate will be significantly increased, and their physiological stress responses (HR, GSR, SCL, SCR) will be significantly reduced. Hypothesis 2: Pilots’ mental health factors (emotional stability, self-discipline, ambition) are positively correlated with emergency task performance, while anxiety and sensitivity are negatively correlated with task performance. Hypothesis 3: Age is negatively correlated with VR emergency task success rate, and younger pilots exhibit better emergency handling performance.

Despite these advances, existing VR aviation training systems have limitations. Most studies integrate only one or two physiological signals (e.g., HRV or EEG alone) and rarely use a pretest/posttest design to directly quantify desensitization effects. Moreover, VR exposure therapy research has focused on treating flight phobia rather than enhancing emergency handling ability in operational pilots, and has relied primarily on subjective reports rather than objective physiological data. In contrast, our system offers three distinct innovations: (1) simultaneous recording of task performance (success/failure, handling time) alongside a comprehensive set of physiological indicators [HR, HRV, GSR, pulse rate variability (PRV), motion, skin conductance response (SCR), skin conductance level (SCL)] within a single experimental framework; (2) three high-fidelity emergency scenarios (bird strike, in-flight fire, engine failure) designed to induce valid stress levels; and (3) a pretest/posttest design to directly assess desensitization effects. These features make our platform uniquely suited for evaluating and improving pilots’ emergency handling ability in a controlled, repeatable, and safe environment.

This study explores the development and application of an emergency handling training system for pilots based on VR technology. Different VR scenarios are presented to evaluate the task execution capabilities of training pilots in special and emergency conditions. The potential and limitations of VR training systems in pilot training are discussed to provide direction for future improvements in flight training.

2 Participants and methods

2.1 Participants

To increase the accuracy and effectiveness of the research results in the context of new types of warfare characterized by efficient cognition, decision-making, and cooperation and to ensure that the research reflects the role of VR technology training in improving pilots’ emergency handling capabilities, a convenience sampling method was adopted. In September 2023, 23 pilots were selected as research subjects. To ensure that the subjects met the relevant requirements and could fully participate in and cooperate with the experimental research to achieve a comprehensive screening of the subjects and minimize the impact of interfering factors on the research, we established the following four inclusion criteria: (1) aged between 20 and 55 years; (2) right-handed, with normal or corrected-to-normal vision; and (3) no prior participation in similar experiments and no VR contraindications. All subjects were male. The success rate of VR tests for different age groups was determined by grouping and proportional calculation methods. The success rate for each age group was calculated by dividing the number of successes by the total number of attempts. Their ages ranged from 23 to 51 years, with an average age of 33.00 ± 7.04 years. The subjects’ number of flight hours ranged from 350 to 8,600 h. The subjects fully understood the content of the experiment and agreed to participate in the study.

2.2 Experimental design

A mixed experimental 2 (pretest × posttest) * 2 (success × failure) * 3 (mid-air bird strike × mid-air fire × mid-air engine failure) design was adopted. All participants were required to complete the basic scale test and baseline value test before the VR task to collect physiological data. Then, they wore VR equipment to complete the VR special situation handling task. Physiological indicator signals were collected throughout, and special situation handling time was recorded. After the VR task, the participants rested and adjusted, and then they completed the special situation handling task again.

2.3 Pilot psychological health scale

The Mental Health Inventory of Chinese Pilot (MHI) used to collect information on the personality traits of the participants (Liu et al., 2002). The scale consisted of 196 items, including 12 personality factors: confidence (Factor Z), vigorousness (Factor C), character tendencies (Factor X), ambition (Factor J), boldness (Factor G), self-discipline (Factor LJ), excessive self-concern (Factor D), emotional stability (Factor Q), anxiety (Factor A), optimism (Factor B), sensitivity (Factor M), and tension (Factor N). Additionally, four dimensions were included: psychological health conditions, manifest behavior, self-affirmation, and performance achievement. Among these, confidence, character tendency, optimism, and psychological health were scored in reverse.

2.4 VR platform development and technical specifications

The “Pilot Emergency Desensitization Training Software System Based on Virtual Reality Technology” was developed by SHINBAN Company (Xi’an, China) using UnrealEngine5 as the core graphics engine, with C++ scripting for logic implementation. The system runs on a dedicated workstation equipped with an Intel Core i7-1260 P processor, 16 GB RAM, and an NVIDIA GeForce RTX 3050Ti graphics card. The head-mounted display (HMD) is an HTC VIVE Pro Eye (resolution 1440 × 1600 per eye, refresh rate 90 Hz), accompanied by controllers (Thrustmaster HOTAS WARTHOG A-10C) for cockpit interaction and the relevant operations in flight control.

The software adopts an event-driven architecture. Three independent emergency event modules (bird strike, in-flight fire, engine failure) are programmed with different triggers using keyboard. Once triggered, the system simultaneously activates visual effects (flames, smoke, bird collision animation) and auditory alarms, while logging the exact timestamp. The function, information security, reliability, and usability of the platform has been independently tested by the China Electronic Product Reliability and Environment Test Research Institute (Guangzhou, China), also known as the Fifth Research Institute of the Ministry of Industry and Information Technology or Ceprei Laboratory.

For the user interface and interaction, the virtual cockpit replicates the layout of a typical fighter jet, including a throttle lever, control stick, and digital multifunction displays. Pilots use the HMD’s head tracking to look around the cockpit and the controllers to push buttons, flip switches, and manipulate the throttle. For technical validation, the frame rate range across all three emergency scenarios were about 45 fps. The latency between software trigger and actual presentation of emergency cues was most no delay. The image quality was adequate (2 K resolution). These metrics confirm that the platform meets the technical requirements for pilot training.

2.5 VR task

The self-developed “Pilot Emergency Desensitization Training Software System Based on Virtual Reality Technology” was used for evaluation and training. The system had two main modes.

• Free flight (i.e., free flight in the scene without emergency situations, mainly for familiarizing the user with operations)
• Emergency mode (this mode can be added during flight, and emergencies can be triggered manually; this is used mainly for data collection). There are three types of emergency modes: ① Bird strike: This type simulates a fighter jet encountering a bird strike in the air, which causes an emergency for the aircraft. The aircraft may experience a power imbalance or loss of power, requiring an emergency return to the airport or landing. ② In-flight fire: This type of emergency simulates a fighter jet experiencing an engine fire in the air accompanied by alarms, thick smoke, and flames, causing the pilot to experience stress. The pilot is required to remain calm, control the burning aircraft, and return to the airport. ③ Engine failure mode: This mode simulates a fighter jet experiencing engine failure in the air accompanied by alarms and loss of power. The fighter jet becomes difficult to control and continuously loses altitude. The pilot must maintain altitude under pressure and return to the airport before crashing.

Four standardized performance metrics were used to evaluate emergency handling performance: ①Emergency Response Latency: Time interval from emergency trigger (visual + auditory cues) to the first correct operational action (unit: second). ②Procedure Compliance Rate: Percentage of mandatory standard emergency operating procedures completed correctly in the required sequence (unit: %). ③Total Task Completion Time: Total time from emergency trigger to full completion of stable aircraft state (unit: second). ④Task Success Rate: Proportion of trials with fully compliant emergency disposal within the time limit (unit: %).

Scoring Criteria: Task Success; All of the following must be satisfied: (1) Correctly identify the emergency type within 10 s; (2) Complete of mandatory steps in the correct order; (3) Maintain stable aircraft attitude and speed; (4) Finish all operations within 180 s; (5) Achieve safe aircraft status (emergency return, stable glide, or controlled descent). Task Failure; Any one of the following: (1) Failed to identify the emergency type within 10 s; (2) Missed ≥2 mandatory steps or incorrect operation sequence; (3) Critical operational error leading to abnormal flight; (4) Aircraft stall, spin, or crash.

2.6 Extraction of physiological indicators

Extraction: HRV, PRV, motion, GSR, SCL, and SCR indicators.

Heart rate variability (HRV) refers to the small time differences between consecutive normal (sinus) heart cycles. HRV reflects the balance and tension between the parasympathetic and sympathetic nerves (SNs) under the influence of the autonomic nervous system (ANS) and the condition of the cardiovascular system. HRV is obtained by analyzing the sequence of RR intervals (RRIs) in electrocardiogram (ECG) signals. The analysis can be divided into time domain analysis, frequency domain analysis, and nonlinear analysis. The pulse wave is similar to the ECG signal waveform, and there are small differences between consecutive pulse intervals, which can be used to calculate HRV by replacing the RR interval (RR interval) in the ECG signal with the peak-to-peak interval (PPI) in the pulse wave signal. HRV calculated using pulse waves is called pulse rate variability (PRV), which, under certain conditions, can replace HRV to reflect autonomic nerve activity. Motion represents the original motion acceleration signal, i.e., the signal recorded each time the arm wearing the watch moves. Motion can be used to infer the motion posture and calculate the corresponding emotional response. GSR is the fluctuation of skin resistance caused by sweat gland activity or changes in the sympathetic nervous system. GSR has the advantages of stability, simple collection, and high sensitivity. The skin conductance level (SCL) is the baseline level of skin conductance without any stimulation. The rise and fall of the SCL changes continuously with an individual’s reaction, skin dryness, or autonomic regulation ability. Basic scale test refers to the standardized assessment of the Mental Health Inventory of Chinese Pilot (MHI) to collect the personality and mental health characteristics of pilots; Baseline value test refers to the collection of resting-state physiological signals (HR, HRV, GSR, SCL, SCR, etc.) in a quiet environment before the VR task training, so as to eliminate the interference of transient physiological fluctuations and ensure the accuracy of subsequent indicator comparison. The skin conductance response (SCR), which is commonly adopted in studies of emotional response, anxiety, stress, and cognitive load, is used to evaluate changes in skin conductance. To address the inflated risk of Type I error resulting from repeated statistical testing across multiple variables, conditions, and pairwise comparisons, Bonferroni correction was applied as the primary method for controlling multiple comparisons in this study. Physiological data processing and indicator extraction were performed using Python 3.9. All statistical analyses including ANOVA, correlation analysis and regression analysis were conducted using IBM SPSS Statistics 29.0.

3 Results

3.1 Comparison of the success rates of VR tasks by age and special situation

There were differences in VR test success rates across age groups. The success rate was highest in the 20s group, at approximately 80%; the 30s group showed the highest success rate among the four groups; the 40s group had a relatively lower success rate, around 30%; and the 50s group had a success rate of zero. In summary, younger age groups performed better in the VR test, with performance peaking between 30 and 39 years old, while success rates decreased significantly for groups aged 40 and above.

3.2 Analysis of mental health and manifest behavioral characteristics

To analyze the relationships between psychological health characteristics and explicit behavior characteristics, descriptive statistics and correlation analyses were conducted on the factors of psychological health and explicit behavior characteristics. Additionally, a scatter plot (Figure 1) and a heatmap (Figure 2) were created.

Each cell’s number represents the correlation coefficient between two variables, ranging from −1 to 1. The closer the value is to 1 or -1, the stronger the correlation between the two variables. The color indicates the strength of the correlation: red indicates a strong positive correlation, blue indicates a strong negative correlation, and white indicates no or very weak correlation.

Pearson correlation analysis showed a significant negative correlation between psychological health characteristics and manifest behavior (r = −0.42, medium effect size). This result indicates that higher psychological health characteristics were associated with lower manifest behavior characteristics.

3.3 The relationships between performance achievement and various factors

Correlation analysis was used to detect the strength and direction of the relationship between performance achievements and other factors. The closer the absolute value of the correlation coefficient is to 1, the stronger the relationship.

Using achievement characteristics as the dependent variable, self-confidence (Factor Z), energy (Factor C), personality tendencies (Factor X), enterprise (Factor J), daring (Factor G), self-discipline (Factor LJ), excessive self-concern (Factor D), emotional stability (Factor Q), anxiety (Factor A), optimism (Factor B), sensitivity (Factor M), and tension (Factor N) were analyzed as independent variables.

Linear regression was applied to construct a relationship model between performance achievement and other factors, and the impact and direction (positive/negative) of each factor on performance achievement were estimated.

3.4 Relationships between special situations of VR task results and various factors

The correlation distribution of psychological factors, regression coefficients and VR task outcome variables are visualized in Figures 3–5, which intuitively present the correlation strength and directional relationships between all study variables. Using age, factors Z to N, and four dimensions, namely, psychological health characteristics, overt behavior characteristics, self-affirmation characteristics, and performance achievement characteristics, as independent variables with VR test results (success or failure) as the target variable, the correlations were calculated and a logistic regression model was built to evaluate the impact of each factor on the VR test results.

The VR test results for “success” were converted to 1 and “failure” was converted to 0. The Pearson correlation coefficient was calculated, and the heatmap of the correlation matrix was plotted. Factors J, LJ, and Q were highly positively correlated, whereas factors G and D had relatively low correlation coefficients.

The distributions of the factors that were highly correlated with the VR test results and their relationships with the success or failure of VR tests were further explored through box plots. Based on the correlation matrix heatmap results between the VR task results and various factors, factor J, factor LJ, and factor Q were selected for analysis to show the quantiles and outliers of the data and whether significant differences existed.

3.5 Special task times and basic information table of physiological indicators

Physiological data, including time, heart_rate, motion, GSR, battery_info, and PRV, for each special task were processed (Table 1).

TABLE 1. Statistical analysis of specific task times and physiological indicators.

3.6 Heart rate

SPSS 29.0 was used to further compare the differences in heart rate means across groups via analysis of variance to verify the research hypothesis. A three-way analysis of variance was conducted with the variables of time (pretest × posttest), task completion status (success × failure), and special circumstances (bird strike × in-flight fire × in-flight parking).

From the descriptive data presented in Table 2, numerical changes in heart rate were observed across different emergency scenarios. For the bird strike scenario, the failure group showed a slightly higher posttest heart rate relative to the pretest level, while the success group exhibited a marginally lower posttest heart rate. A similar numerical trend was also found in other emergency tasks. For the in-flight engine shutdown scenario, the failure group had a higher numerical value of posttest heart rate compared with the pretest, whereas the success group presented a lower posttest heart rate in numerical terms.

TABLE 2. Descriptive statistics of heart rate.

Three-way ANOVA results (effect size: Partial η ^*2* reported in Table 3) showed that only the main effect of Condition was significant; all first-order, second-order, and third-order interactions were not significant. Post hoc tests revealed that heart rate during bird strikes was significantly lower than heart rate during in-flight fires (P = 0.010) and in-flight holds (P = 0.006).

TABLE 3. Summary table of heart rate variability.

3.7 Motion

Based on the hypothesis of equivalence and homogeneity of variance, SPSS 29.0 was used to analyze the mean and standard deviation of motion. Further analysis of variance was used to compare the differences in average values of motion among groups to further verify the research hypothesis using the variables of time (pretest x posttest) * task completion status (success x failure) * special situation (bird strike x in-flight fire x in-flight shutdown) for three-factor analysis of variance.

When birds were hit in the air, the posttest motion of the failing group was lower than the pretest motion, whereas the posttest motion of the successful group was lower than the pretest motion. When there was a fire in the air, the posttest motion of the failing group was lower than the pretest motion, and the posttest motion of the successful group was lower than the pre-test motion. With regard to in-flight shutdown, the post-test motion of the failing group was lower than the pretest motion, and the posttest motion of the successful group was lower than the pretest motion.

Three-way ANOVA results (effect size: Partial η ^*2* reported in Table 5) showed no significant main effects or interaction effects for Condition, Time, or Task Completion. All first-order, second-order, and third-order interactions were not statistically significant Descriptive statistics and variance analysis results of motion indicators are presented in Tables 4, 5.

TABLE 4. Descriptive statistics of motion.

TABLE 5. Summary of variance for motion.

3.8 GSR

To test the equivalence and compare the latent means, SPSS 29.0 was used to analyze the average and standard deviation of the GSR. Furthermore, analysis of variance was used to compare the differences in the average values of the GSR among groups to further verify the research hypothesis. The variables of time (pretest × posttest) * task completion status (success × failure) * special situation (bird strike × engine fire × emergency landing) were used to conduct a three-factor analysis of variance.

In the case of bird strikes in the air, the posttest GSR of the failure group was lower than the pretest GSR, and the posttest GSR of the success group was lower than the pretest GSR. In the case of in-flight fires, the posttest GSR of the failure group was lower than the pretest GSR, and the posttest GSR of the success group was lower than the pretest GSR. In the case of in-flight parking, the posttest GSR of the failure group was lower than the pretest GSR, and the posttest GSR of the success group was lower than the pretest GSR.

Three-way ANOVA results (effect size: Partial η ^*2* reported in Table 7) showed significant main effects of Time and Task Completion, as well as significant Condition × Time and Condition × Time × Task Completion interactions. Simple effects analysis indicated that in the failure group, pretest GSR was significantly higher than posttest GSR across all three scenarios (P < 0.001); no significant pretest-posttest difference was found in the success group (P > 0.05). In the failure group, pretest GSR of bird strikes was lower than that of in-flight fires (P = 0.028), and posttest GSR of in-flight fires was lower than that of in-flight shutdowns (P = 0.016). The failure group had significantly higher pretest and posttest GSR than the success group in all scenarios (P < 0.05). Tables 6, 7 report the descriptive data and three-way ANOVA outcomes of GSR indicators.

TABLE 6. Descriptive statistics of GSR.

TABLE 7. Summary of variance for GSR.

3.9 PRV

Based on the hypothesis of testing equivalence and comparing latent means, SPSS 29.0 was used to analyze the mean and standard deviation of PRV. An analysis of variance was conducted to compare the differences in the mean values of PRV among groups to further validate the research hypothesis via a three-factor analysis of variance with the following variables: time (pretest × posttest) * task completion (success × failure) * special situation (bird strike × in-flight fire × in-flight shutdown).

In the case of bird strikes in the air, the posttest PRV of the failing group was higher than the pretest PRV, and the posttest PRV of the successful group was also higher than the pretest PRV. In the case of in-flight fires, the posttest PRV of the failing group was lower than the pretest PRV, whereas the posttest PRV of the successful group was higher than the pretest PRV. In the case of in-flight parking, the posttest PRV of the failing group was higher than the pretest PRV, and the posttest PRV of the successful group was higher than the pretest PRV.

Three-way ANOVA results (effect size: Partial η ^*2* reported in Table 9) showed no significant main effects or interaction effects for Condition, Time, or Task Completion. All interaction effects were not statistically significant. The descriptive statistics and variance summary table for PRV are shown in Tables 8, 9.

TABLE 8. Descriptive statistics of PRV.

TABLE 9. Summary of variance for PRV.

3.10 Hypothesis testing of the mean effects for VR special situation tasks

Python 3.9 was used to capture the various indicators of the VR special task completion of the subjects and process the physiological data of the occurrence of special emotions such as heart_rate, GSR, SCR, and SCL. The aggregated mean test data of all physiological indicators are listed in Table 10.

TABLE 10. Basic information table for mean effects test.

3.11 Heart rate

On the basis of the hypothesized testing equivalence and comparison of latent means, further analysis was conducted using ANOVA to compare the differences in the mean heart rate among the groups to further validate the research hypothesis with a two-factor ANOVA using the factors of time (pretest × posttest) and task completion status (success × failure).

In the pretest, the mean heart rate of the failure group was lower than that of the success group. In the posttest, the mean heart rate of the failure group was also lower than that of the success group. Repeated-measures ANOVA was conducted on the data of the two groups, and the results are shown in Figure 1.

Two-way ANOVA results (effect size: Partial η ^*2* reported in the text) showed that the main effect of Time was not significant, F _(1,21) = 5.899, P = 0.562, partial η ^*2* = 0.016. The interaction effect of time and task completion was not significant, F _(1,21) = 0.070, P = 0.794, partial η ^*2* = 0.003. The main effect of task completion was significant, F _(1,21) = 3.893, P = 0.062, partial η ^*2* = 0.156. Grouped pretest-posttest heart rate descriptive results are summarized in Table 11.

TABLE 11. Description of heart rate.

3.12 GSR

Based on the hypothesis supporting the equivalence and latent mean comparison, further analysis of variance was conducted to compare the mean differences in GSR among groups to verify the research hypothesis using a two-factor analysis of variance with time (pretest × posttest) and task completion status (success × failure).

In the pretest, the mean GSR of the failure group was greater than the mean GSR of the success group. In the posttest, the mean GSR of the failure group was greater than the mean GSR of the success group. A repeated-measures analysis of variance was conducted on the two groups of data.

Two-way ANOVA results (effect size: Partial η ^*2* reported in the text) showed a significant main effect of Time, F _(1,21) = 17.109, P < 0.001, partial η ^*2* = 0.449, with a pretest GSR that was higher than the posttest GSR. The interaction between time and task completion was not significant, F _(1,21) = 3.184, P = 0.089, partial η ^*2* = 0.132. The main effect of task completion was significant, F _(1,21) = 8.572, P = 0.008, partial η ^*2* = 0.290, with a lower GSR in the failure group than in the success group. Pretest and posttest GSR data of success and failure groups are displayed in Table 12.

TABLE 12. Descriptive statistics of GSR.

3.13 SCR

To further verify the research hypothesis, a two-way analysis of variance was conducted to compare the differences in the mean SCR values of each group on the basis of the hypothesis of testing equivalence and comparison of latent mean ratios with the factors of time (pretest × posttest) and task completion status (success × failure).

In the pretest, the mean SCR of the failure group was greater than that of the success group. In the posttest, the mean SCR of the failure group was also greater than that of the success group. Repeated-measures analysis of variance was conducted on the data of the two groups.

Two-way ANOVA results (effect size: Partial η ^*2* reported in the text) showed a significant main effect of Time, F _(1,21) = 17.629, P < 0.001, partial η ^*2* = 0.456, with higher pre-SCR than post-SCR. The interaction between time and task completion was not significant, F _(1,21) = 0.005, P = 0.944, partial η ^*2* = 0.000. The main effect of task completion was significant, F _(1,21) = 7.066, P = 0.015, partial η ^*2* = 0.252, with higher SCR in the failure group than in the success group. Group descriptive statistics of SCR indicators are provided in Table 13.

TABLE 13. Description statistics of SCR.

3.14 SCL

Based on the hypothesis of testing equivalence and comparison of the latent mean, further analysis of variance was conducted to compare the differences in the mean SCL values among the groups to further validate the research hypothesis using a two-way analysis of variance with time (pretest × posttest) and task completion status (success × failure) as factors.

In the pretest, the mean SCL value of the failure group was greater than the mean SCR value of the success group. Similarly, in the posttest, the mean SCL value of the failure group was greater than the mean SCL value of the success group. Repeated-measures analysis of variance was conducted on the data from both groups.

Two-way ANOVA results (effect size: Partial η ^*2* reported in the text) showed a significant main effect of Time, F _(1,21) = 17.839, P < 0.001, partial η ^*2* = 0.459, with a higher pretest than posttest SCL. The interaction between time and task completion was not significant, F _(1,21) = 2.226, P = 0.151, partial η ^*2* = 0.096. Task completion had a significant main effect, F _(1,21) = 7.547, P = 0.012, partial η ^*2* = 0.264, with higher SCL in the failure group than in the success group. The pretest-posttest SCL data of two task groups are shown in Table 14.

TABLE 14. Descriptive statistics of SCL.

4 Discussion

This study aimed to use virtual reality technology to evaluate and train pilots’ handling abilities when they face special situations. The results indicate that after VR desensitization training, pilots exhibited shorter emergency handling times and lower physiological stress responses, suggesting an association between VR training and better emergency coping responses (Bustamante-Sánchez et al., 2021). Virtual reality provides a highly realistic flight environment for repeated emergency procedure practice and was associated with more stable performance in simulated tasks (Rogier et al., 2024). In this study, changes in pilots’ reaction time and physiological indicators were associated with VR training exposure in the simulated environment. This study identifies significant associations between VR training exposure and improved emergency handling performance in pilots (Muhlberger et al., 2001). Changes in physiological indicators are important references for evaluating pilots’ stress levels and emotional states (Soares et al., 2024). In VR training for special situations, pilots’ heart rate, blood pressure, and other indicators increase, reflecting their stress response in simulated emergency scenarios. Across VR training trials, physiological stress indicators showed corresponding reductions, indicating increased adaptability in the VR environment, reduced physiological reactions to special situations, and a greater ability to remain calm and make correct decisions (Harris et al., 2023; Ji et al., 2023). By comparing the data before and after training, we found that pilots’ ability to deal with special situations improved significantly after VR training. This finding indicates that virtual reality technology can serve as an effective training tool to help pilots develop their decision-making abilities and operational skills. Although this study demonstrates the potential of VR technology for training pilots to handle special situations, several limitations exist. First, the small sample size may have influenced the generalizability of the research findings (Das et al., 2020; O'Neill, 2022). Second, all participants were male, which limits the generalizability of the results in terms of gender (Fu et al., 2024; Mohan et al., 2024). In addition to the lack of female pilots, the sample was also limited to a relatively narrow age range (23–51 years) and did not include student pilots or reserve personnel. To improve generalizability, future studies should recruit a larger and more diverse cohort that includes pilots of different genders, ages, flight experience levels, and aircraft types (Geoffrey, 2014; Muschalla, 2024). Additionally, virtual reality technology is not yet able to replicate all the sensations and experiences of real flight, which may affect a pilot’s performance in the handling of real special situations (Kim et al., 2023; van Weelden et al., 2024). Moreover, combining biofeedback technologies such as EEG monitoring could provide a more comprehensive assessment of a pilot’s psychological state (Cross et al., 2023; Eva-Maria et al., 2009; Ji et al., 2023). Finally, different types and levels of special situation scenarios could be used to evaluate pilots’ ability to cope in a wider range of contexts (Kinney and Hare, 2020).

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Sevinç Gülseçen, Istanbul University, Türkiye

Reviewed by: Fiona Duruaku, University of Central Florida, United States

Liu Zhenlei, Shenyang University of Aeronautics and Astronautics, Macao SAR, China