This research involved human participants. Approval of all ethical and experimental procedures and protocols was granted by the Ethics Committee of Beihang University (BM20230063). Strict privacy protection measures were applied, and all data were anonymized with no personally identifiable information included. Prior to participation, all individuals were fully informed about the study procedures, potential risks, and their right to withdraw at any time without penalty. All participants provided written informed consent and were compensated at a rate of 100 CNY (approximately US $14) per hour.

Considering that older adults constitute a potentially vulnerable population, additional safeguards were implemented for the older adult group. These included the presence of at least one qualified health professional to provide immediate medical support, repeated verbal reports to monitor fatigue, dizziness, and overall comfort, and the option for participants to terminate the experiment immediately upon experiencing any discomfort.

The study was conducted with 21 healthy older adults (14 males and 7 females) aged 65 to 80 years (mean age 71.79 , SD 4.57 years) and 18 healthy younger adults (10 males and 8 females) aged 21 to 33 years (mean age 26.50, SD 2.99 years). All participants had no history of neurological or psychiatric disorders. Older participants were recruited from the community and were at least 65 years old and able to complete daily activities independently. One older participant with a history of cerebral hemorrhage was excluded prior to the study, and another withdrew due to intolerance to VR-induced dizziness. Younger participants were recruited through public advertisements posted on the university campus. Individuals who expressed interest were screened using the same inclusion criteria applied to all participants, including normal or corrected-to-normal vision, no history of neurological or psychiatric disorders, and no prior diagnosis of cognitive impairment. Additionally, individual information, encompassing age, gender, education, and prior VR experience, was systematically gathered from each participant.

The virtual navigation game used in this study was developed with Unity 2019.4 (Unity Technologies) and deployed via the HTC VIVE Pro head-mounted display. The visual scene was rendered at 90 Hz and presented at a single-eye resolution of 1440×1600 pixels (2880×1600 binocular) with a 110° field of view, providing an immersive experience. For the experimental setup, the head-mounted display was connected to a high-performance desktop PC via a wired connection using the standard VIVE link box to ensure low latency and stable rendering. The system was equipped with an Intel Core i7-9700K CPU, an NVIDIA GeForce RTX 2070 GPU, 32 GB of RAM, and the Windows 10 operating system, enabling smooth rendering and reliable behavioral data acquisition.

The VE was designed as a simplified yet ecologically valid urban block (200 vm×150 vm, virtual meters), featuring residential buildings, street networks, greenery, and small parks. The intentional integration of these elements provided sufficient spatial complexity to elicit meaningful wayfinding behavior while minimizing potential discomfort or anxiety associated with navigation in a VE [46].

Given that older participants may experience limitations in sustained physical walking, we adopted a hybrid navigation control scheme combining the handheld controller and natural facing direction [47]. The VIVE controller features an ergonomic design with a circular touchpad. To minimize operational complexity, participants were instructed to press the touchpad to move forward in the direction they were facing, and releasing it stopped movement immediately. Horizontal body rotation was mapped one-to-one to virtual orientation, allowing participants to turn naturally without additional button input. Throughout the experiment, all participants were instructed to avoid movement in the physical space and to navigate using the controller and body rotation only. Most participants completed the game while standing, whereas three older participants, due to advanced age, performed the experiment while seated in a swivel chair that allowed 360° physical rotation. This simplified control scheme reduced cognitive and motor demands while ensuring consistent interaction across participants.

The experiment consisted of three phases: (1) neuropsychological assessment, (2) pretraining, and (3) formal experiment. The neuropsychological assessment phase aimed to evaluate the cognitive abilities of older participants using standardized measures of general cognitive skills and empirically validated tests. During the pretraining phase, participants were instructed to navigate freely within the VE for 15 minutes. Navigation proficiency was considered achieved when participants demonstrated the ability to independently and fluently operate the controller and navigate to designated target locations, qualifying them to proceed to the formal experimental phase. In the formal experimental phase, participants were required to sequentially navigate to multiple target locations, with spatiotemporal data recorded at fixed intervals throughout the navigation process. Older participants completed all 3 phases, while younger participants, who were healthy university students with presumed normal cognitive function, only participated in phases 2 and 3.

Before the formal navigation tasks, general cognitive functioning was quantified using the Mini-Mental State Examination (MMSE) [48,49] and the Montreal Cognitive Assessment (MoCA) [50]. Specifically, the MMSE administered in this study was the standardized Chinese adaptation based on the international MMSE framework, which has been validated through prior pilot testing and is widely used in studies involving Chinese populations [48,49,51]. The MoCA Full Chinese (Beijing) Version 7.1 was the official Beijing version published on the MoCA website [52]. MoCA scores were adjusted according to participants’ education level, with one additional point added for individuals with 12 years of formal education or fewer. Nonverbal memory was further assessed using the delayed recall condition of the Rey-Osterrieth Complex Figure Test, while visuospatial functioning was evaluated through the Copy condition of the Rey-Osterrieth Complex Figure (ROCF-C) test [53]. To examine executive function and working memory, Part A of the Trail Making Test (TMTA) was administered [54]. Additionally, spatial rotation ability was assessed using a standardized Mental Rotation Test (MRT) [55]. These assessments provided a comprehensive cognitive profile for each participant prior to engagement in the virtual navigation experiment.

During the pre-experimental phase, identically shaped but differently colored target objects were placed at various locations within the VE (Figure 1). Participants were instructed to memorize the spatial locations corresponding to each colored object. The specific verbal instructions can be found in Multimedia Appendix 1. During the formal experimental phase, 5 target locations previously encountered during the training phase were used (Figure 2C). Target locations were presented in a fixed sequential order, allowing for consistent route comparisons across participants. Each trial began with a fixation cross (1000 ms), followed by the target cue (2000 ms), another fixation cross (1000 ms), and then participants were required to navigate to the designated target location (Figure 2A). Upon reaching the target, the current trial was concluded, and the next trial was automatically initiated. This sequence continued until all tasks were completed. Participants were instructed to navigate as quickly and accurately as possible (Figure 2B), and a congratulatory animation was displayed after each trial to prevent boredom.

Data from the experiment were processed using custom scripts developed in MATLAB, and all statistical analyses were performed using SPSS (IBM Corp). The normality of each variable was assessed through visual inspection of histograms and the Shapiro–Wilk test. Spearman’s rank correlation analyses were conducted to examine the associations between navigation-related behavioral measures and cognitive test scores. To evaluate age-related differences, independent-samples t tests were applied to variables that met normality assumptions, whereas the Mann-Whitney U test was used for variables that were not normally distributed. In addition, correlations between navigation efficiency and the various experienced spatial metrics were examined using Pearson correlations for normally distributed variables and Spearman rank correlations for nonnormally distributed variables.

All 19 older participants and 18 young participants successfully completed the experiment without experiencing severe motion sickness. Table 1 summarizes the cognitive characteristics of the older group, including the means and SDs for various cognitive assessment scales. Notably, none of the participants had prior experience with VR technology, which reduced potential biases associated with equipment familiarity.

After clustering the data, points labeled as wayfinding state exhibited clear spatial concentration and age-related differences (Figure 5). This spatial density map visualizes wayfinding patterns, offering an intuitive overview of participants’ spatial behavior based on spatiotemporal features. The red-outlined areas indicate the top 1% of regions with the highest density values, highlighting zones of intensified wayfinding activity within the VE.

Spearman rank correlation analysis was conducted to examine the relationships between navigation behavior metrics and cognitive assessment scores (Table 2). No significant correlations were observed between either distance or error and any of the cognitive scales. However, temporal-based navigation parameters performed well. Navigation time was significantly associated with both TMTA (r=0.503, P=.047) and the MRT (r=−0.526, P=.03). Wayfinding time showed strong correlations with several cognitive measures, including TMTA (r=0.716, P=.002), MoCA (r=−0.549, P=.02), ROCF-C (r=−0.588, P=.01), MMSE (r=−0.458, P=.046) and MRT (r=−0.751, P<.001). Additionally, navigation efficiency demonstrated significant associations with MoCA (r=0.495, P=.04) and ROCF-C (r=0.658, P=.003) scores, and particularly strong correlations with TMTA (r=−0.761, P=.001) and MRT (r=0.848, P<.001). These findings confirmed that cognitive abilities play a crucial role in influencing navigation efficiency, including navigation time and wayfinding time.

The Mann-Whitney U test was used to assess differences in navigation performance between age groups (Figure 6). Compared to young participants, older adults, who exhibited significantly lower cognitive ability, demonstrated significantly longer navigation time (z=−4.771, P<.001) and wayfinding times (z=−4.315, P<.001), as well as reduced navigation efficiency (z=−4.285, P<.001).

The above metrics reflect participants’ overall wayfinding performance. However, navigation behavior is not solely influenced by internal factors such as cognitive ability. It also varies depending on the location of participants within the VE. Figure 7A illustrates differences in the navigation process across age groups during the formal experiment. To investigate this, participants’ spatial positions are mapped onto the shortest navigation routes (excluding trials involving route errors), and the time spent per unit distance is calculated for each individual. Based on these data, we plot curves of navigation time and navigation efficiency for each age group. Furthermore, a permutation test is conducted on navigation efficiency within each unit distance between older and young adults. The shaded regions on Figure 7B correspond to areas where significant differences are observed (P<.05) in Figure 7A, revealing spatial zones where age-related differences in navigation behavior are most pronounced. These results highlight age-related differences in spatial behavior and reinforce the utility of these behavioral metrics as sensitive indicators of cognitive function.

The influence of environmental characteristics on navigation performance is analyzed from the perspective of spatial configuration using both AMA and VGA. Between-group comparisons revealed significant age-related differences in several experienced spatial measures. Older adults showed higher EAI (z=–2.43, P=.01) and EVI (t=2.48, P=.02), whereas young adults exhibited higher EMAD (z=–2.13, P=.03) and EMVD (t=–2.49, P=.02). No significant group differences were found for experienced axial connectivity and experienced visual connectivity.

Correlation analyses further revealed distinct age-specific associations between navigation efficiency and experienced metrics. Among older adults, navigation efficiency showed a significant negative correlation with EVI (r=–0.48, P=.04) and a positive correlation with EMVD (r=0.49, P=.03). In contrast, young adults demonstrated strong associations with axial properties, with navigation efficiency negatively correlated with EAI (r=–0.64, P=.005) and positively correlated with EMAD (r=.72, P<0.001). Young adults also exhibited a significant positive relationship between navigation efficiency and EMVD (r=0.57, P=.01).

This study investigated how cognitive abilities and environmental structure jointly shape navigational performance in a virtual navigation game across different age groups. Consistent with prior research, our findings demonstrate that performance in navigation games, particularly navigation efficiency, is strongly influenced by individual cognitive capacity, as well as the spatial configuration of the environment. These results underscore the relevance of navigation behavior as an indicator of cognitive status and highlight the potential of VR-based navigation tasks as effective tools for cognitive assessment. Moreover, the insights gained from this work also provide valuable guidance for future urban design and navigation-game development.

The behavioral metrics used in this study, including traversal distance, number of errors, and navigation time, are widely used in human navigation research [34,60]. Spatiotemporal representation has long been essential for analyzing navigation behavior, and the present study extends conventional approaches by classifying participants’ navigation states using an individualized method. Navigation tasks are generally considered to consist of 2 main components [61]: locomotion and wayfinding. Locomotion refers to the physical execution of movement in the space, whereas wayfinding is involved in determining and planning a route to a desired destination [62]. Traditional approaches typically distinguish these components using fixed absolute speed thresholds [47,63]. To account for individual differences in navigation ability, we applied k-means clustering to each participant’s instantaneous speed distribution to derive personalized relative speed thresholds. k-Means is a classical clustering algorithm that iteratively assigns data points to the nearest cluster center [64]. To assess the reliability of this classification, clustering solutions with different numbers of clusters (k) were evaluated across all younger and older participants using the sum of squared errors (SSE) [65] and the silhouette coefficient [66] (Figure S3 in Multimedia Appendix 2). The SSE quantifies within-cluster variance by measuring the squared distance between each data point and its assigned cluster center, whereas the silhouette coefficient provides a complementary assessment of clustering quality by comparing consistency within clusters with separation between other clusters [67]. When k=3, the SSE curve showed a clear inflection point, indicating a markedly diminished rate of decrease in within-cluster variance. Meanwhile, the silhouette coefficient remained relatively high and exhibited more stable variance, suggesting an optimal balance between cluster cohesion and separation. In addition to these statistical indicators, the three clusters aligned well with meaningful navigation states observed in participants’ movement patterns (Figure S2 in Multimedia Appendix 3): (1) low-speed hesitation or searching, (2) medium-speed transitional adjustments, and (3) high-speed goal-directed movement. Taken together, both the quantitative metrics and the behavioral interpretability support the individualized clustering approach, which enabled the effective categorization of each sampled time point into low-, medium-, or high-speed states, thereby identifying regions within the environment where participants associated with wayfinding, transition, or continuous movement.

Effective spatial navigation relies on the integration of multiple abilities, including visual perception, spatial orientation, learning, and memory [68]. Wayfinding encompasses all of the ways in which people orient themselves in physical space and navigate from place to place. As anticipated, wayfinding ability, as a cognitive element of the navigation process [69], is effectively captured by parameters such as wayfinding time and navigation efficiency, both of which are shown to be sensitive indicators of cognitive capacity in the current study. As summarized in Table 2, cognitive scales including the MMSE, MoCA, ROCF-C, TMTA, and MRT exhibited robust correlations with these navigation metrics. Among these, the MoCA and MMSE are widely used for comprehensive cognitive assessment; the TMTA primarily reflects executive functioning; ROCF-C performance is sensitive to visuoconstructive and geometric processing abilities; and the MRT assesses mental rotation capacity. These cognitive functions collectively support navigation by contributing to processes such as initial route planning, continuous movement control, reorientation, and decision-making. SG-based VR navigation tasks elicit continuous, ecologically relevant behavior under dynamic task constraints, enabling the simultaneous assessment of multiple cognitive functions within an integrated behavioral context. Quantifying wayfinding performance provides an indirect yet meaningful reflection of an individual’s cognitive functioning [34]. Longer wayfinding times and lower navigation efficiency were associated with poorer cognitive performance in our older participants. Taken together, these findings highlight the potential of well-designed navigation games to serve as effective tools for cognitive assessment [70].

Human navigation in outdoor environments involves a series of distinctive behaviors, such as retracing steps, hesitation, and reorientation [71]. These behaviors become particularly evident in complex spatial settings where individuals must process environmental cues and make directional decisions. Wayfinding time serves as a key indicator for revealing navigation uncertainty, effectively pinpointing locations where external environmental factors impact individuals’ sense of directional uncertainty (Figure 4). We observe that prolonged wayfinding times were predominantly concentrated in the turning areas of the path and the starting or ending locations of each trial. These highlighted areas in the figure represent regions where participants are more likely to encounter decision-making bottlenecks, experiencing uncertainty in selecting the correct path [72]. Furthermore, integrating the analysis in Figure 6, we find significant differences in navigation performance between older and younger participants in certain regions.

Synthesizing findings from previous research, we propose that these results may be explained by age-related differences in navigation strategy preferences. Human navigation relies on 2 primary strategies: allocentric (survey-based) and egocentric (route-based). Younger adults tend to rely more on the allocentric strategy [18] and show greater flexibility in switching between allocentric and egocentric frames of reference [73]. However, aging is associated with a decline in allocentric navigation abilities, accompanied by a preferential shift toward egocentric strategies [18,21,22], which may contribute to less efficient route planning and increased reliance on familiar or visually accessible paths. These strategy differences align closely with the behavioral patterns observed in our study. Young participants adopted a more proactive and goal-directed approach: they typically scanned the environment at the beginning of each trial, extracted relevant spatial cues, and formed a planned route before moving. In contrast, older participants showed a stimulus-response navigation pattern [74,75], making decisions only upon arriving at intersections rather than planning ahead. This led to hesitation at nearly every decision point (Figure 6B), reflecting greater reliance on immediate cues and reduced anticipation of upcoming turns. Consequently, participants spent more time searching for landmarks or other environmental cues, leading to decreased navigation performance [76,77]. Such behaviors are consistent with age-related declines in spatial working memory and executive function [78], which limit their ability to efficiently integrate spatial cues.

To explore how environmental structure influences navigation across age groups, we used line-based (AMA) and grid-based (VGA) analytic approaches. Both methods quantify spatial configuration and offer complementary perspectives on how spatial structure shapes navigation behavior. Within this framework, integration serves as a core space syntax metric capturing the degree to which a path is connected to its surrounding environment [13,29]. In this study, experienced integration reflects the actual spatial integration values of the segments that participants traversed, thereby characterizing the dynamic interplay between environmental structure and navigational behavior. Older adults exhibited significantly higher EAI and EVI than young adults, suggesting a tendency to navigate toward more integrated areas rather than taking shorter or more direct routes. Prior studies have shown that spending more time in highly integrated areas facilitates the formation of accurate cognitive maps [13]. With aging, older adults require more time to form cognitive maps and encode spatial information [79], which may partly explain their tendency to remain within highly integrated axial regions during navigation. From another perspective, although integration is the normalization form of mean depth [25], the two measures capture different aspects of spatial structure. Mean depth quantifies the average topological steps from a location to all others, with higher values indicating deeper and harder-to-reach positions. Older adults showed lower EMAD and EMVD values, suggesting a tendency to remain within relatively shallower, more accessible areas.

In our navigation game, all participants were instructed to reach the target location as quickly as possible. By comparing each group’s experienced metrics with those of the shortest path (Figure 8A), we found that the experienced metrics of younger participants closely matched the optimal path. This pattern is consistent with behavioral observations showing that young participants, driven by the goal of rapid completion, tended to select more direct and efficient routes. Correlation analyses further support these environment-mediated navigation patterns. Among young adults, EAI was negatively correlated with navigation efficiency, indicating that more movement through globally integrated axial lines was associated with increased wayfinding behaviors (eg, hesitation, pausing, reorientation), thereby reducing overall efficiency. For older adults, EVI showed a significant negative correlation with navigation efficiency, suggesting that more extensive traversal within visually integrated regions corresponded to heightened hesitation or searching behaviors. The results for EMAD and EMVD exhibited similar trends: achieving higher navigation efficiency and shorter routes requires participants to strategically decide whether to traverse spatially advantageous or disadvantageous areas. Younger adults tended to accept navigating through deeper or less accessible regions to optimize route efficiency [56], whereas older adults were more likely to remain within visually accessible areas, even when these were not part of the shortest path.

This subjective tendency in navigational behavior indicates that participants do not passively respond only to environmental structure; rather, their navigational choices also reflect intentional, strategy-driven preferences [13]. Such findings point to a bidirectional relationship between spatial configuration and navigation behavior, where environmental structure shapes movement patterns, and individuals’ strategic preferences, in turn, influence the parts of the environment they traverse. Understanding whether environmental properties or individual behavior play the dominant role in guiding navigation will require further empirical investigation.

Additionally, our findings extend this literature by demonstrating age-specific patterns in the relationship between experienced metrics and navigation efficiency. Navigation efficiency in younger adults was strongly associated with line-based (AMA) metrics such as EAI and EMAD, whereas older adults showed stronger associations with grid-based (VGA) metrics, such as EVI and EMVD. These differences can be meaningfully interpreted through the lens of allocentric and egocentric navigation strategies. Younger adults tend to rely more on allocentric representations, enabling them to rapidly extract structural information from the environment, integrate axial connectivity, and flexibly switch between global structure and local visual information. This cognitive flexibility allows them to traverse deeper or less accessible areas. In contrast, aging is associated with an increased reliance on egocentric navigation. Older adults become more dependent on the visual field and immediate cues, which aligns with their stronger correlation between navigation efficiency and visual-based experienced metrics (EVI and EMVD). Their tendency to remain within visually accessible, highly integrated regions suggests that they prioritize environments that offer clear visual guidance and reduced cognitive demands [14].

Together, space syntax highlights the fundamental role of spatial configuration in shaping navigation behavior, and our findings further show that its influence varies with age, ultimately affecting navigation performance. These insights have important implications not only for developing more effective cognitive assessment tools but also for designing age-friendly urban environments that support safe, efficient, and accessible navigation. Strengthening the coordination and integration of street space is essential for urban street-design guidelines and road-network optimization [80]. Through axial and grid-based analyses, space syntax provides a powerful framework for identifying high integration. Such information can guide the placement of landmarks, directional signage, and other environmental cues that facilitate cognitive map formation and enhance navigability for diverse user groups. Moreover, because spatial integration is closely tied to navigation difficulty [29], purposeful modulation of integration levels can help calibrate the cognitive demands of navigation tasks. From an SG perspective, the majority of participants verbally reported a high level of immersion and task engagement, suggesting that the SG-based framework effectively supported sustained interaction with the spatial environment. By integrating principles of spatial configuration into an immersive VR navigation game, this approach may provide promising opportunities for the development of gamified tools for cognitive assessment, training, and rehabilitation, particularly for individuals with spatial or cognitive impairments.

This research highlights the critical role of cognitive abilities and spatial configuration in shaping navigation behavior. However, several limitations should be acknowledged. First, the participant sample may not fully represent the broader population. The young participants were primarily between 20 and 30 years of age, whereas the older group ranged from 65 to 85 years, leaving a substantial gap between ages 30 and 65 years, which was not included. This omission limits the generalizability of the findings, as individuals in midlife may exhibit distinct navigation patterns and cognitive characteristics that were not captured in the current analysis [72]. Second, the individualized k-means clustering approach adopted in this study is particularly suited to navigation games characterized by relatively stable movement speeds. It may be less appropriate for tasks involving pronounced, irregular, or continuous speed transitions, in which clear and interpretable cluster boundaries are difficult to establish. Moreover, although a 3-cluster solution showed good behavioral interpretability within the current navigation game, the generalizability of this clustering scheme to other navigation games or more complex real-world environments remains to be validated in future studies. Third, discrepancies between virtual and real-world navigation are unavoidable [76]. The absence of environmental features such as directional signs, combined with the lack of natural physical movement, may influence how spatial information is encoded, updated, and retrieved [81]. Finally, the study did not explicitly account for the influence of demographic factors such as gender, educational background, or geographic and cultural experience. Ignoring these variables could obscure important individual differences and interactions relevant to wayfinding performance.

Future research should aim to address these limitations by recruiting a more demographically diverse participant sample, including middle-aged individuals, and by designing VEs with greater ecological validity. Incorporating navigation methods that more closely approximate natural physical movement may yield richer behavioral data and uncover additional nuances in navigational performance. Additionally, systematically examining the role of experienced metrics could provide deeper insights into urban design, spatial planning, and navigation-aid development in both virtual and real-world contexts. From a rehabilitation perspective, the game demonstrates promising potential for virtual navigation tasks as a controlled and adaptable platform for personalized cognitive assessment and training. Further research could investigate the incorporation of adaptive task difficulty and longitudinal monitoring into VR-based rehabilitation, thereby supporting the maintenance or recovery of spatial and cognitive functions.

This study provides quantitative evidence for how environmental configuration and cognitive abilities jointly shape navigational behavior in VEs. By applying an individualized k-means clustering approach, we extracted key behavioral indicators, such as wayfinding time and navigation efficiency, that proved sensitive to cognitive decline. This method offers a more personalized and potentially more effective framework for assessing cognitive function through navigation tasks. Through spatial analysis, we further identified age-specific regions of wayfinding difficulty and quantified the extent to which environmental structure influences navigation behavior. These findings reveal distinct navigation patterns between younger and older adults and underscore the value of integrating behavioral analytics with spatial-syntax metrics. Overall, the results lay an important foundation for future applications in clinical cognitive assessment and rehabilitation, as well as in the development of age-friendly urban environments.