Age differences in spatial navigation stem from a preference for familiar routes rather than impaired landmark-dependent strategies

Abstract

A leading hypothesis in the field of aging and navigation is that older adults are selectively impaired on tasks that require allocentric (landmark-based) strategies to navigate, resulting in a shift towards more egocentric (self-based) strategies. However, most evidence in humans comes from studies that restrict body-based sensorimotor cues that are essential to both egocentric and allocentric navigation. In the present study, young and older adults navigated a virtual environment in each of two conditions: a stationary desktop condition that relied on visual input, and an immersive condition that enabled unrestricted ambulation and sensorimotor feedback during navigation. Both age groups performed worse when initially learning locations from novel compared to familiar locations – often considered a hallmark of allocentric navigation. The cost of switching from familiar to novel start locations was equal between age groups, pointing to a null effect of age on allocentric strategies. Older adults also employed distal landmarks to a comparable extent to young adults, suggesting that landmark-dependent strategies did not differ by age. However, older adults were more likely to replicate previously taken paths, potentially indicative of a preference for egocentric strategies. The path replication effect was significantly attenuated in the immersive condition, particularly in the presence of geometric boundary cues that could be used to infer distance. Age differences in spatial navigation may therefore be driven in part by a selective bias for navigating familiar routes, although these differences were lessened in the presence of multimodal visual and sensorimotor cues. The present study highlights that navigation is a complex cognitive construct that draws on multiple parallel systems and strategies that cannot be easily explained by a simple allocentric-egocentric dichotomy.

Past research suggests that successful navigation of new spatial environments declines with healthy aging, although the reasons for this finding remain unresolved (Lester et al., 2017). Spatial navigation is often considered to rely on two complementary strategies. Allocentric or ‘placed-based’ navigation requires learning the relationships between external spatial cues, such as landmarks or environmental boundaries, which in turn are thought to underlie the computation of novel and more efficient routes. Egocentric or ‘route-based’ navigation involves self-reference to landmarks and typically relies more on travelling familiar, well-learned routes and is often considered to be less flexible and efficient than allocentric navigation. One hypothesis in the field of aging is that allocentric-based navigation is impaired in older age, resulting in a preference for more egocentric-based navigation strategies (Colombo et al., 2017; Iaria et al., 2003; Li & King, 2019; Moffat et al., 2006).

The ability to form a cognitive map of the environment, which is central to most models of allocentric navigation, is generally believed to involve the integration of external (e.g., landmarks, boundaries) and body-based (e.g., motor efference copy, proprioceptive, vestibular) sensory cues (Campbell et al., 2018; O’Keefe & Nadel, 1978; Siegel & White, 1975). Prior studies in humans, however, have relied overwhelmingly on experimental paradigms that require participants to navigate a two-dimensional virtual reality (VR) environment rendered on a computer monitor (Daugherty et al., 2015; Iaria et al., 2009; Moffat et al., 2007; Moffat & Resnick, 2002; Zhong et al., 2017). These types of ‘desktop VR’ paradigms rely primarily on visual input while restricting the types of body-based sensory cues that are central to allocentric and egocentric forms of spatial navigation. Experimental studies employing desktop VR can also be confounded by age differences in prior experience with using a joystick or keyboard and mouse to navigate a first-person virtual environment. Finally, many of these studies do not directly test how landmarks are integrated during body and head movements. A critical gap in the literature, therefore, concerns whether age differences in navigation strategies persist when body-based cues are available to support navigation (McAvan et al., 2021).

Age differences in allocentric navigation may also be dependent on the types of visual cues available within the environment. Human studies of spatial navigation often require participants to learn the relationships between discrete landmarks, such as a unique building or street sign. There is growing evidence, however, that older adults may be selectively disadvantaged at using these types of visual landmarks to orient in space (Colmant et al., 2023; West et al., 2023). Instead, older adults may be more adept at using spatial cues that convey the relative lengths, distances, and angles between large-scale surfaces within the environment, such as walls or walkways. Studies employing these types of geometric spatial cues have shown that older adults are just as likely as younger individuals to orient using putative allocentric search strategies (Bécu et al., 2019, 2023). An overarching aim of the present study, therefore, is to examine the extent to which age differences in navigation strategy preferences are affected by the availability of enriched visual and body-based sensorimotor cues.

The Morris watermaze (MWM) task is a commonly used assessment of allocentric navigation and spatial memory (Morris et al., 1982). Originally developed for rodents, the MWM task involves learning the locations of hidden objects within an environment surrounded by distal landmark cues. Successful performance on the task requires participants to learn and subsequently remember the location of hidden objects with reference to the distal cues, which is generally considered to depend on externally referenced allocentric memory processes. Virtual adaptations of the MWM task developed for human participants and administered on a desktop computer have been used extensively to examine putative allocentric navigation abilities in aged humans. Older adults are consistently observed to remember locations with respect to distal landmarks less accurately than do younger individuals on the MWM task (Daugherty et al., 2015; Daugherty & Raz, 2017; Hill et al., 2024; Iaria et al., 2009; McAvan et al., 2021; Moffat et al., 2007; Moffat & Resnick, 2002; Zhong et al., 2017). By contrast, older adults are generally observed to perform just as well as younger individuals during conditions in which visual cues indicate the location of a previously ‘hidden’ target (i.e., visible platform trials).

When empoying the MWM and other spatial tasks, however, it is important to note that allocentric and egocentric navigation strategies are difficult to isolate and common behavioral measures often involve a combination of strategies and spatial reference frames (Ekstrom et al., 2014; Johnsen & Rytter, 2021; Wolbers & Wiener, 2014). In the present study, we considered multiple measures of putative allocentric navigation in order to reduce the potential influence of this issue. One such measure involves manipulating the familiarity of the navigation start positions such that participants are cued to locate hidden targets from familiar (i.e., well-learned) and novel locations (Eichenbaum et al., 1990; Kolarik et al., 2018; McAvan et al., 2021; Wolbers & Wiener, 2014). The assumption is that performance on familiar and novel start trials should be relatively similar in those individuals using an allocentric reference frame to infer the location of a hidden target. By contrast, those individuals that rely more heavily on an egocentric reference frame should perform reasonably well when taking a well-learned route from a familiar start position but show reduced performance when navigating from a novel start position, which places a greater emphasis on the distal landmark cues.

In the present study, we examined the impact of enriched visual and body-based sensory cues on spatial memory as young and older adults navigated a virtual MWM environment (Figure 1). Participants were trained to find the locations of hidden targets in a virtual outdoor environment in each of two virtual reality conditions: a desktop VR condition which required using a keyboard and mouse to navigate, and an immersive and ambulatory VR condition which enabled unrestricted self-motion during navigation (see Hill et al., 2024). We employed a version of the virtual MWM in which four distal landmarks were clearly visible outside of the navigable space in all environments. We tested older and young adults on memory for a hidden target location from both a novel and familiar (repeated) start location during both acquisition and delayed probe trials.

Figure 1. Virtual Morris Water Maze Task Design.

(a) Visual depiction of the immersive and desktop VR conditions. The assignment of the respective environments (snow, desert) to each VR condition (desktop, immersive) was fully counterbalanced across participants, as was the testing order of the respective conditions. (b) First person depiction of the available spatial cues in the no boundary (distal mountains only) and boundary (distal mountains and perimeter wall) conditions. (c) During the spatial acquisition blocks, participants were trained to locate the spatial positions of three hidden target objects (color coded). Immediately following each spatial learning block participants were cued to recall the location of the hidden object without feedback (immediate probe trials). Following the learning blocks, participants completed eight visible target trials to rule out potential age-related motivational and/or sensorimotor confounds. Participants completed 12 delayed probe trials without feedback, during which they were cued to locate the hidden target when starting from a familiar (solid bars) or novel (dashed bars) location. In a subset of the delayed probe trials, a singular mountain cue was rotated 20 degrees clockwise or counterclockwise unbeknownst to the participant. Figure adapted from Hill et al. (2024), Age differences in spatial memory are mitigated during naturalistic navigation. Aging, Neuropsychology, and Cognition, 1–25. Licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/).

On a subset of delayed probe trials, we rotated one of the distal mountain cues to test participant reliance on the stable three mountain cues or the single (moved) mountain. Greater reliance on the single moved mountain cue might suggest a greater reliance on a beaconing strategy. In a subset of participants (N = 48), a circular boundary was located outside of the navigable arena to serve as a geometric spatial cue (Figure 1B). Critically, the distal mountain cues and circular boundary wall were intended to convey unique sources of spatial information during navigation. Whereas the distal mountain cues could be used to infer relative directions within the otherwise featureless environment, the circular boundary was meant aid in the estimation of relative distances within the navigable environment. The experimental design therefore allowed us to test whether age differences in navigation strategies are differentially affected by the inclusion of geometric spatial cues (boundary vs. no boundary) and/or body-based sensory cues (immersive VR vs. desktop VR).

Materials and Methods

Transparency and openness

We report exclusionary criteria and describe all manipulations and measures that were collected. Deidentified data and analysis code are available from the first author upon request and a link to these materials are provided in the Author Note. The study design and analysis procedure were not pre-registered. The data presented here was also presented in a previous paper with a focus on the impact of virtual testing modality on age differences in spatial memory (Hill et al., 2024). The focus of the current paper is instead on testing hypotheses related to spatial strategy and age, an issue not covered in the previous paper. The descriptions of the participant information and virtual reality procedures overlap heavily with the descriptions given in that report and are partially summarized here. The analyses and behavioral findings described below are novel and have not been reported previously.

Participants

We recruited 45 young and 46 older adults from the University of Arizona and surrounding communities. Participants were recruited in two separate waves to be enrolled in the no boundary (20 young, 20 older) and boundary (24 young, 24 older) versions of the MWM task (see ‘Procedures’). One older adult voluntarily withdrew from the study due to a scheduling conflict; one young and one older adult withdrew after experiencing mild motion sickness during the virtual reality task. The final participant sample included 44 young adults (18–31 yrs.; M = 21.50 yrs; SD = 3.23 yrs; 27 females) and 44 older adults (63–80 yrs.; M = 70.48 yrs.; SD = 4.89 yrs.; 21 females). The final sample size (N = 88) was informed by an a priori power analysis conducted in G*Power employing between-repeated measures. The sample was selected to ensure power of 0.85 (at p < .05) to detect medium age group x VR condition interactions (f = .25) in a 2 (age group) x 2 (VR condition) ANOVA. Given the novelty of our study design, we elected to perform a power analysis based on a conservative (medium) effect size estimate to avoid underestimating the sample size needed to detect a meaningful two-way interaction. A chi square analysis confirmed that the distribution of gender did not significantly differ between the young and older adult groups (χ² = 1.14, p = .285). All participants gave informed consent in accordance with the University of Arizona Institutional Review Boards and were compensated at the rate of $18 per hour. All participants had normal or corrected-to normal color vision, normal or corrected-to-normal hearing, and reported no history of cardiovascular problems, neurological conditions, or history of motion sickness.

All older adult participants completed a neuropsychological test battery on a day prior to the experimental VR session. The test battery included multiple tests and scores in each of five broad cognitive domains: memory [California Verbal Learning Test-II (CVLT) Long Delay Free Recall (Delis et al., 2000), Rey-Osterrieth Complex Figure Test Long Delay Free Recall (RCFT-LDFR) (Rey, 1941)], executive function [Trail Making Tests A and B total time (Reitan & Wolfson, 1985)], language [F-A-S fluency (Spreen & Benton, 1977), category fluency (Benton, 1968), Bostron Naming Test (BND; Goodglass et al., 2001)], visuo-spatial abilities [Wechsler Adult Intelligence Scale 4^th edition(WAIS-IV) Block Design Test (Wechsler, 2009), RCFT copy score (Rey, 1941)], and verbal intelligence [WAIS-IV Vocabulary and Similarities (Wechsler, 2009)]. Participants were excluded from entry into the study if they scored greater than 1 SD below age-appropriate norms on any memory test or greater than 1 SD below age norms on any two other tests. These criteria were employed to minimize the likelihood of including older individuals with mild cognitive impairment (Bondi et al., 2008; Jak et al., 2009). Scores on the neuropsychological assessments did not differ between older adults enrolled in the boundary and no boundary conditions (Supplementary Table 1). Older adults in the respective boundary and no boundary conditions also did not differ in years of education obtained (t_(40.98) = −0.45, p = .657; boundary M = 17.42 yrs, SD = 2.02 yrs; no boundary M = 17.65 yrs, SD = 1.42 yrs).

Procedures

Participants performed virtual adaptations of the Morris water maze task. The virtual environments and experimental tasks were built in Unity 3D (Unity Technologies ApS, San Francisco, CA) using the Landmarks virtual reality navigation package (Starrett et al., 2021). All participants performed the task in each of two virtual conditions. In the immersive VR condition, participants were equipped with a wireless head mounted display (HTC Vive Pro Eye) and a handheld HTC Vive controller. This setup was used to simulate an immersive experience of being in a mountainous environment while simultaneously allowing for unrestricted locomotion and body-based sensory cues during navigation. Participants also completed an analogous version of the task on a desktop computer while seated in a behavioral testing room. This desktop VR condition required using a keyboard and mouse to navigate the two-dimensional environment. The order of the respective immersive and desktop VR conditions was fully counterbalanced across all participants. Additional technical specifications for the respective VR conditions are described by Hill et al. (2024).

Each virtual environment was rendered with a unique floor (snow-covered or desert, counterbalanced across the respective desktop and immersive VR conditions) and four distally rendered mountains located on the horizon. These distinctive renderings were meant to reduce interference and potential carryover effects from the preceding virtual session such that participants would treat the two environments independently. The navigable environment in both the desktop and immersive VR conditions was approximately 5 × 5 m in size, with the full visible space spanning approximately 750 × 750 m. A subsample of young (N = 20) and older (N = 20) adult participants performed a version of the task in which the distal mountains were the sole source of spatial information within the environment (i.e., no boundary condition). A separate sample of young (N = 24) and older (N = 24) adult participants completed an identical version of the task apart from a circular wall (20 m diameter) placed around the perimeter of the navigable space in addition to the four distal mountain cues (i.e., boundary condition).

Participants were trained to learn the location of three hidden objects placed at fixed positions within the 5 × 5 m navigable space. Each virtual environment was associated with three unique 3D rendered objects (book, rubiks cube, teapot in the snow-covered environment; alarm clock, mug, rotary phone in the desert environment). The objects were presented on pedestals at approximately chest height. Participants completed 16 consecutive learning trials for each target location. Prior to each trial in the immersive VR condition, the headset displayed a blank screen while an experimenter guided the participant along a random path for 30 s. This disorientation phase was meant to prevent participants from tracking their locations and movements from the previous trial. A similar disorientation phase was included in the desktop VR condition, during which participants viewed a blank screen for 30 s before starting the next trial.

Following the 30 s disorientation, participants were placed at one of eight predetermined starting positions. Each block of 16 study trials was organized into four sub-blocks, each corresponding to a unique start position (e.g., 4 consecutive trials from position 1, followed by 4 consecutive trials from position 5, etc.). Each study trial began with visual instructions indicating the navigation goal (e.g., “Please find the book”). Participants then freely navigated while searching for the hidden object which remained invisible for the first 30 s of the trial.

To encourage spatial learning, participants were instructed to remember the fixed positions of the hidden objects and to navigate to this location and respond via button press before 30 s had elapsed. This button press was treated as a spatial memory response and the location and timestamp of the response were recorded in the data output. Upon making a response, the object became visible to permit spatial updating during the study trials. If a button response was not made, the target object would appear after 30 s and the participants location at the time the object became visible was logged in the data output. At the end of each 16-trial spatial learning block, participants were placed in a novel start position and instructed to navigate to the location of the hidden target (immediate probe trials). Unlike the immediately preceding study trials, the object did not appear after 30 s, nor did the object appear upon making a button press.

Following the three study blocks, participants completed one block of eight visible target trials. During these trials, objects were placed in their original studied locations and remained visible for the entire duration of the trial. This was meant to serve as a control for motivational and sensorimotor differences while performing the task (Moffat & Resnick, 2002; Morris et al., 1982). The results of the visible target trials are summarized in the Supplemental Materials and described in more detail in Hill et al. (2024).

After completing eight visible target trials, participants performed 12 delayed probe trials to assess delayed spatial recall of the hidden target locations. As with the immediate probe trials, the target object did not appear after 30 s, nor did it appear after a button press. Participants were thus required to navigate to the location of the hidden target object based solely on their memory. Upon reaching this location, participants made a button press to record their location and timestamp. In a separate set of three delayed probe trials (one trial for each hidden target object), we rotated the position of a single mountain cue 20° clockwise or counterclockwise to assess potential age differences in the relative use and weighting of distal landmark cues during navigation (see Statistical Analyses). For these trials, we rotated the mountain with an angular orientation closest to that of the hidden target object when standing in the center of the arena. This was done to ensure the mountain was clearly visible when facing the hidden target location.

Analytic Approach

Position and rotation data were recorded throughout the entire experiment at a sampling rate of approximately 10 Hz. All statistical analyses were conducted with R software (R Core Team, 2022). All t-tests were two-tailed and performed using the t.test function in the base R package. Welch’s unequal variance t-tests were performed when assumptions of equal variance were not met. Multiple regression analyses were conducted using the lm function in the base R package. ANOVAs were conducted using the afex package (Singmann, H. et al., 2016) and the Greenhouse-Geisser procedure (Greenhouse & Geisser, 1959) was used to correct degrees of freedom for non-sphericity when necessary. Post-hoc tests on significant effects from the ANOVAs were conducted using the emmeans package (Lenth, R., 2018) and corrected for multiple comparisons using the Holm-Bonferroni procedure where appropriate. Bayesian mixed-effect regression models for circular data were performed using the bpnme package in R (Cremers et al., 2018). Unless otherwise specified, we report the results of three-way mixed-factorial ANOVAs with between-subject factors of age group (young, older) and cue type (boundary, no boundary) and the within-subject factor of VR condition (desktop, immersive). Bayes factor values (BF₁₀) were computed using the BayesFactor package (Morey et al., 2022) and interpreted using the categorical labels proposed by Jeffreys (1961).

Switching between familiar and novel routes.

During the study phase, young and older adult participants were trained to learn target locations by navigating from fixed locations within the respective virtual environments. These locations were organized in a blocked format such that each starting location was repeated over four consecutive trials for a given target before moving to a novel start position and repeating this process for the same target. This blocked format was meant to ensure participants learned a specific start location by having participants repeatedly navigate from this location. Participants were therefore cued to recall hidden target locations from familiar (i.e., previously learned) and novel start positions. Prior studies have employed familiar and novel start positions as an assay of egocentric and allocentric navigation strategies, respectively (Eichenbaum et al., 1990; Kolarik et al., 2018; McAvan et al., 2021; Wolbers & Wiener, 2014).

We tested whether older adults would show greater ‘switching costs’ when taking a novel compared to familiar route to locate the hidden target (Harris et al., 2012; Harris & Wolbers, 2014), a proxy for egocentric navigation strategies. Spatial memory was operationalized as the Euclidean distance (in virtual meters) between the remembered and true location of each target object (i.e., spatial distance error). To maximize route familiarity, performance on familiar trials was computed as the mean distance error on the fourth consecutive trial initiated from a given location within the environment (i.e., the fourth and final trial of each block). Performance on novel trials was similarly computed as the mean spatial distance error on trials immediately following each familiar trial (i.e., the first trial of the next block, including immediate probe trials).

We also examined whether young and older adults differed in their tendency to replicate familiar, previously learned paths when switching to a novel start position. We reasoned that participants preferring more of an egocentric or route-based strategy would tend to replicate familiar paths even when navigating from a novel start position. This should result in greater familiar-to-novel path similarity compared to those preferring a more allocentric strategy (Iggena et al., 2023). We quantified path replication as the spatial distance between navigation paths taken on consecutive familiar (i.e., fourth and final trial of a given block) and novel (i.e., first trial of the next block) trials. Navigation paths were defined as the vector of XY coordinates of the VR headset over the duration of a given trial. Consecutive paths were aligned to obtain a common starting coordinate (0, 0). To account for differences in path length, we used a non-linear dynamic time warping algorithm to find the optimal match between each respective pair of navigation trajectories using the dtw package in R (Giorgino, 2009) function. This analysis thus yielded a measure of dissimilarity between respective paths, with higher scores indicating less similarly (or more allocentric), and lower scores indicating more similar trajectories (or more egocentric).

Landmark-based navigation during delayed probe trials

Successful allocentric navigation requires encoding the spatial relationships between stable landmarks within the environment. One possibility is that older adults are less likely than young adults to orient towards the distal mountain cues when attempting to locate the hidden target. To examine this possibility, we computed the trialwise angular difference between participant heading at the time of retrieval (button press) during the delayed probe trials and the heading error for each of the respective distal mountains cues, yielding four estimates of landmark error. Landmark error was computed separately for each of the respective distal mountain cues (yielding four landmark error estimates). For each trial, we identified the smallest of the four landmark error estimates and used these values as the dependent variable in a mixed-effect regression analysis. This approach was motivated by the assumption that the most salient distal cues used to orient within the environment would yield the smallest angular distance error. We similarly computed the trialwise angular difference between participant heading at retrieval and the correct heading of the hidden target object (target error). Target and landmark error values could thus range from 0° to 360°.

We used Bayesian mixed-effect regression models for circular data (Cremers et al., 2018) to examine whether young and older adults differed in their tendency to orient towards the distal mountain cues and/or hidden target during the delayed probe trials (Cremers & Klugkist, 2018; Du et al., 2023). We constructed separate models for landmark and target error. For each model, we entered heading error (transformed to range from −180° to 180° then converted to radians) as the dependent variable. Age group, VR condition, and cue type were entered into the model as fixed effect predictors, and by-subject intercept terms were entered as random effects (angular error ~ age group + subject). We used the bpnme package in R (Cremers, 2020) to run a Bayesian Markov Chain Monte Carlo (MCMC) sampler with 1000 iterations for the mixed effect model to calculate the posterior estimates of the landmark and target errors for each age group, VR condition, and cue type, the circular means, and 95% higher posterior density (HPD) interval of the respective angular error estimates.

We next examined the subset of probe trials in which a single distal mountain cue was rotated 20° clockwise or counterclockwise. We computed the distance error between a given object’s remembered location and the hypothetical location of that object if it were rotated 20° commensurate with the rotated mountain cue (rotated error). We then divided the rotated error by the sum of the rotated error and the true distance error between the estimated and true (un-rotated) location of that target object [error index = rotation error / (rotation error + true error)]. Values approaching 0 signify greater weighting of the rotated mountain cue (i.e., beacon strategy) and values approaching 1 indicate greater weighting of the three stationary mountain cues (i.e., allocentric strategy). Due to a programming error, the direction (clockwise vs. counter-clockwise) of the cue rotations were not properly recorded in the desktop condition. Analysis of the rotated mountain cues was therefore restricted to the immersive VR condition.

Results

Older adults do not show selectively worse performance compared to young adults when finding a hidden target from a novel start location compared to a familiar one

The ability to switch between familiar and novel routes to locate hidden targets is often used as a proxy for allocentric navigation strategies in the MWM task (Eichenbaum et al., 1990; Kolarik et al., 2018; McAvan et al., 2021; Wolbers & Wiener, 2014). Here, we tested whether older adults would show worse performance compared to young adults when navigating to the hidden target from a novel start position compared to a well learned and familiar start position within the environment. To this end, we submitted the mean distance error during the acquisition trials to a four-way mixed factorial ANOVA with factors of age group (young, older), VR condition (immersive, desktop), cue type (boundary, no boundary), and start position (novel, familiar). This analysis revealed a significant main effect of start position (F_(1,84) = 60.47, p = 1.7⁻¹¹, partial-η² = .419, BF₁₀ = 78.546). Post-hoc tests confirmed that this effect was driven by increased distance error when starting from a novel start position compared to starting from a familiar start position (t₍₈₄₎ = −7.78, p < .001).

Critically, the interaction between age group and start position was not significant and a Bayes value suggested strong evidence in favor of the null hypothesis (F_(1,84) = 0.46, p = .499, partial-η² = .005, BF₁₀ = .167). The respective interactions between start position and VR condition (F_(1,84) = 1.49, p = .226, partial-η² = .017, BF₁₀ = .183) and between start position and cue type (F_(1,84) = 1.24, p = .269, partial-η² = .015, BF₁₀ = .197) were also non-significant, with strong Bayes evidence in favor of the null hypothesis. Indeed, familiar vs. novel effect size estimates were largely similar for young and older adults across the respective VR and cue type conditions (see Table 1). These findings suggest that older adults are not selectively impaired when attempting to locate the hidden target from a from a novel start location compared to young adults.

Table 1.

Mean distance error (with standard error) when navigating from familiar and novel start positions

	Young Adults			Older Adults
	Familiar	Novel	t-value	Familiar	Novel	t-value
Study Trials
Immersive Boundary	.99 (.11)	1.22 (.11)	−3.13**	1.53 (.11)	1.78 (.11)	−3.32**
Immersive No Boundary	1.62 (.12)	2.03 (.12)	−5.15***	2.44 (.12)	2.71 (.12)	−3.34**
Desktop Boundary	.82 (.16)	1.03 (.18)	−2.05*	2.54 (.16)	2.72 (.18)	−1.82
Desktop No Boundary	1.78 (.18)	2.02 (.19)	−2.20*	3.44 (.18)	3.66 (.19)	−2.03*
Delayed Probe Trials
Immersive Boundary	1.59 (.15)	1.25 (.12)	3.12**	2.18 (.15)	1.93 (.12)	2.53*
Immersive No Boundary	2.69 (.16)	2.25 (.13)	3.62***	3.27 (.16)	2.57 (.13)	5.72***
Desktop Boundary	1.35 (.25)	1.17 (.19)	1.42	3.32 (.25)	2.72 (.19)	4.52***
Desktop No Boundary	2.36 (.27)	1.99 (.21)	2.54*	4.04 (.27)	3.16 (.21)	6.14***

Estimated marginal means (with standard error). Values from two-sided paired t-tests comparing distance error on familiar and novel start trials are presented.

^*p < .05

^**p < .01

^***p < .001

During the delayed probe trials, participants were prompted to locate each of the respective hidden targets from memory following a delay while navigating from novel start locations (i.e., starting locations not encountered during the spatial learning phase). On a subset of delayed probe trials (two of 12 total trials), participants were prompted to locate the hidden target when starting from a familiar location encountered during the spatial learning phase. We submitted the mean distance error during the delayed probe trials to a four-way ANOVA with factors of age group (young, older), VR condition (immersive, desktop), cue type (boundary, no boundary) and start position (novel, familiar). The four-way ANOVA revealed a significant main effect of start position (F_(1,84) = 114.91, p = 5.7⁻¹⁶, partial-η² = .578) which, in contrast to the study trials, was driven by greater distance error when starting from a familiar compared to novel start position (t₍₈₄₎ = 10.72, p < .001). The interaction between cue type and start position was also significant (F_(1,84) = 8.26, p = .005, partial-η² = .090). Post-hoc tests confirmed this interaction was driven by significantly larger switching costs in the no boundary (t₍₈₄₎ = 9.20, p < .001) compared to the boundary (t₍₈₄₎ = 5.82, p < .001) conditions. Note that a permutation test accounting for the unbalanced number of familiar and novel trials produced an equivalent pattern of results (see Supplementary Materials).

The three way interaction between age group, VR condition, and start position was also significant (F_(1,84) = 4.25, p = .042, partial-η² = .048). To unpack this interaction, we ran a follow-up age x start position ANOVAs separately for the immersive and desktop VR conditions. The interaction between age and start position was significant in the desktop VR condition (F_(1,86) = 11.01, p = .001, partial-η² = .113) and was driven by a significantly larger effect of start position in older (t₍₈₆₎ = 7.43, p < .001) compared to younger (t₍₈₆₎ = 2.74, p = .007) individuals. The analysis of the immersive VR condition revealed a significant main effect of start position (F_(1,86) = 49.37, p = 4.7⁻¹⁰, partial-η² = .365). Importantly, the interaction between age and start position when navigating in the immersive VR environment was not significant (F_(1,86) = 0.28, p = .596, partial-η² = .003). These results suggest that, in contrast to the study trials, participants generally performed worse (i.e., greater distance error) when attempting to locate the hidden target from a familiar start position compared to a novel start position following a delay. One possibility is that this pattern of effects relates to interference, a possibility we explore in the discussion. Notably, though, age differences in finding the hidden target from a novel compared to familiar position were only evident in desktop VR and not present in either of the two immersive VR conditions tested (boundary and no boundary).

Age differences in path rigidity when switching between familiar and novel start positions during spatial learning are attenuated in immersive VR

We next examined whether young and older adults differed in their tendency to replicate previously learned paths when switching from a familiar to a novel start position during the study phase (see Figure 3). We submitted mean path dissimilarity (see Analytic Approach) to a three-way mixed factorial ANOVA with factors of age group (young, older), VR condition (immersive, desktop), and cue type (boundary, no boundary). This analysis revealed significant main effects of age (F_(1,84) = 57.39, p = 4.3⁻¹¹, partial-η² = .406), VR condition (F_(1,84) = 68.90, p = 1.5⁻¹², partial-η² = .451), and cue type (F_(1,84) = 18.84, p = 3.9⁻⁵, partial-η² = .183). The interaction between age and VR condition was also significant (F_(1,84) = 19.34, p = 3.2⁻⁵, partial-η² = .187). Critically, these effects remained significant when controlling for age- and VR-related differences in angular rotations (see Supplemental Materials).

Figure 3.

Path replication during spatial learning. (A). Example routes taken on consecutive trials initiated from a well-learned and familiar start position (left panels) and a novel start position (right panels). The paths were realigned to create a common starting coordinate. The top row illustrates high path dissimilarity which is indicative of an allocentric strategy. The bottom row illustrated highly similar paths which is indicative of an egocentric strategy. (B) Barplot (top panel) illustrating the effect of age group on mean path distance (i.e., path dissimilarity) separately for VR condition and cue type. Age differences in path similarity were significantly attenuated in the immersive VR condition compared to the desktop VR condition. The spaghetti plots (bottom panel) emphasize that attenuated age effects in the immersive VR condition were driven by increased path similarity among young adults.

Post-hoc pairwise comparisons of age group and VR condition revealed significant differences between all levels of these interacting variables. We note that effect size estimates were maximal when comparing young and older adults in the desktop VR condition (Young > Older contrast; Immersive VR: t₍₈₄₎ = −4.72, p < .001; Desktop VR: t₍₈₄₎ = −7.44, p < .001) and when comparing immersive and desktop performance in young adults (Immersive > Desktop contrast; Young: t₍₈₄₎ = −8.98, p < .001 Older: t₍₈₄₎ = −2.76, p = .007). These results suggest that older adults were generally more likely to stick with a familiar path compared to young adults, but that this age difference was significantly reduced in the immersive VR condition. Critically, reduced age differences in the immersive VR condition were driven by a tendency for young adults to take more familiar (i.e., egocentric) routes, rather than a shift towards more novel (i.e., allocentric) routes in older adults (see Figure 3B).

The three-way interaction between age group, VR condition, and cue type (boundary vs. no boundary) was not significant (F_(1,84) = 2.56, p = .113, partial-η² = .030). Prior work, however, suggests that the presence of environmental boundaries might moderate age differences in allocentric navigation (Bécu et al., 2019; Bécu et al., 2023). Motivated by these prior findings, we submitted mean path dissimilarity to exploratory age x cue type ANOVAs separately for the desktop and immersive VR conditions to examine whether the presence of boundaries might moderate any of the observed age differences in path dissimilarity. The analysis of the immersive VR condition revealed a significant interaction between age group and cue type (F_(1,84) = 4.72, p = .033, partial-η² = .053). Post-hoc comparisons revealed significant age differences in path similarity in the no boundary condition (t₍₈₄₎ = −4.67, p < .001, BF₁₀ = 248.893) along with a null effect of age in the boundary condition (t₍₈₄₎ = −1.89, p = .062, BF₁₀ = 1.421). The interaction between age group and VR condition was not significant in the desktop VR condition (F_(1,84) = 0.61, p = .436, partial-η² = .007) and post-hoc tests confirmed the effect of age was highly significant in both the boundary (t₍₈₄₎ = −4.94, p < .001) and no boundary (t₍₈₄₎ = −5.66, p < .001) desktop VR conditions. Although they should be interpreted cautiously, these results suggest that the presence of the perimeter boundary mitigated the tendency of older adults to replicate a previously learned path when navigating from a novel start location, but only in the immersive VR condition.

Older adults orient toward landmarks as effectively as younger adults but show greater angular error when orientating towards the hidden target

We next examined whether older adults were impaired at using the distal landmarks to navigate. To this end, we examined whether young and older adults differed in their relative tendency to orient in the direction of the distal mountain cues during the delayed probe trials (i.e., landmark error, see Analytic Approach). Landmark error was numerically but not significantly lower in older adults (M = 0.29 radians, 95% HPD 0.25–0.35) compared to young adults (M = 0.36 radians, 95% HPD 0.30–0.42). Landmark error did not significantly differ between the immersive (M = 0.28 radians, 95% HPD 0.24–0.32) and desktop (M = 0.38 radians, 95% HPD 0.32–0.45) VR conditions, or between the boundary (M = 0.35 radians, 95% HPD 0.30–0.45) and no boundary (M = 0.30 radians, 95% HPD 0.25–0.35) conditions. These findings suggest that older adults were able to orient to each of the four landmarks comparably to young adults.

We next considered how young and older adults might orient differently towards the hidden targets. Consistent with past reports of greater spatial distance error in older adults when locating hidden targets (Hill et al., 2024; McAvan et al., 2021), angular target error was reliably lower in young adults (M = 0.55 radians, 95% HPD 0.45–0.66) compared to older adults (M = 1.02 radians, 95% HPD 0.79–1.32). Angular target error was also lower in the boundary condition (M = 0.44 radians, 95% HPD 0.37–0.52) compared to the no boundary condition (M = 1.32 radians, 95% HPD 1.00–1.63). Target errors in the immersive (M = 0.66 radians, 95% HPD 0.54–0.79) and desktop (M = 0.82 radians, 95% HPD 0.61–1.01) VR conditions did not reliably differ, as indicated by overlapping 95% HPD intervals. These findings suggest that older adults performed worse at orienting toward the target, although this did not appear to relate to an inability to orient toward the landmarks.

Older adults do not show a selective tendency to rely on a single landmark compared to young adults

External landmarks can often act as reference points or ‘beacons’ when located near a goal location. Navigating towards these fixed landmarks moves the navigator closer to the goal location without placing any explicit demands on egocentric stimulus-motor associations or more abstracted allocentric representations (Waller & Lippa, 2007; Weiner et al., 2013). Although young and older adults did not differ in their tendency to orient towards the distal landmarks, the above analyses do not rule out the possibility that older adults were over reliant on using a single landmark to navigate (beacon strategy), rather than triangulating spatial positions based on multiple landmarks (allocentric strategy) (see Weiner et al., 2013). To address this possibility, we rotated the single distal mountain cue with the angular orientation closest to that of the hidden target on a subset of the delayed probe trials. We reasoned that among those participants relying on a single mountain cue to locate the hidden target, distance error would predictably rotate along with the moved mountain (Figure 5A). Alternatively, participants that triangulate their position using multiple distal mountain cues should be less affected by the single rotated mountain (see McAvan et al., 2021).

Figure 5. Rotated Mountain Trials.

(A) A singular distal mountain cue was rotated 20 degrees clockwise or counterclockwise on a subset of the delayed probe trials. The blue and red dots indicate the actual and hypothetical rotated location of the hidden target, respectively. (B) Rotation indices did not significantly differ between age groups or cue types. (C) Rotation indices negatively covaried with distance errors estimated during the delayed probe trials. This negative association was moderated by cue type, with a stronger negative association observed in the no boundary condition compared to the boundary condition.

The rotation indices for each age group and cue type are illustrated in Figure 5B. A two-way ANOVA with factors of age group (young, older) and cue type (boundary, no boundary) showed no significant main effect of age group (F_(1,81) = 0.48, p = .493, partial-η² = .006, BF₁₀ = .272) as well as a non significant age group x cue type interaction (F_(1,81) = 0.53, p = .468, partial-η² = .007, BF₁₀ = .362). These findings suggest that older adults did not rely on a single mountain cue to a greater extent than young adults. The main effect of cue type was also not significant (F_(1,81) = 0.87, p = .353, partial-η² = .011, BF₁₀ = 1.25), though the Bayes value suggested weak or ‘anecdotal’ evidence for the alternative hypothesis.

To further explore the effects of the rotated mountain cue, we performed one-sample t-tests (two-tailed) to compare rotation indices against a hypothetical mean of 0.50 separately for each age group and cue type. Values at or below .5 would suggest a tendency to use a single landmark to a greater extent than the collection of the three landmarks. Error indices were significantly greater than 0.50 among older adults in the boundary condition (M = 0.52; SD = .04; t₍₂₂₎ = 2.38; p = .026), suggesting a potential greater reliance on an allocentric strategy. Rotation indices in young adults did not significantly differ from 0.5 in either of the cue type conditions (boundary: M = 0.52; SD = .07; t₍₂₂₎ = 1.50; p = .148; no boundary: M = 0.52; SD = .05; t₍₁₈₎ = 1.68; p = .110), nor did they differ from 0.5 among older adults in the no boundary condition (M = 0.50; SD = .03; t₍₁₉₎ = 0.52; p = .607). These results suggest that, although young and older adults show a relatively equal weighting of the rotated and unmoved mountain cues, older adults might show a slight preference for using multiple stationary landmarks when additional boundary cues are also present.

Individual differences in reliance on multiple landmarks relates to the precision of remembering the hidden target independent of age

In a final set of analyses, we examined whether individual differences in rotation indices were related to the precision of spatial memories for target locations during the non-rotated delayed probe trials. We reasoned that those that rely more heavily on the collection of unmoved landmarks to orient within the environment (allocentric strategy) would be able to remember target locations more precisely than those that relied more heavily on a single rotated cue (beacon strategy). To test this idea, we performed a multiple regression analysis with mean distance error on the delayed probe trials as the dependent variable and rotation indices, age group, cue type, and the respective two- and three-way interaction terms as independent variables. This analysis revealed a significant negative linear relationship between distance error and rotations indices (β = −8.63; F_(1,77) = 28.23, p = 1.0⁻⁶). Critically, the interaction between age group and rotation indices was not significant (F_(1,77) = 0.79, p = .378). This pattern of results suggests that, regardless of age, participants who were more likely to use multiple unmoved mountains to locate the hidden target on rotated probe trials generally performed better on the non-rotated delayed probe trials.

We also found a significant interaction between rotation indices and cue type (F_(1,81) = 4.67, p = .034). To unpack these results further, we computed partial correlations between distance error and rotation indices separately for the boundary and no boundary conditions while controlling for age group (Figure 5C). This revealed a robust and significant negative correlation between distance error and rotation indices in the no boundary condition (r_partial = −.61, p = 3.3⁻⁵). A non-significant negative correlation was also evident in the boundary condition that was comparatively weaker than that observed in the no boundary condition (r_partial = −.29, p = .054). In other words, those who relied on multiple mountains to a greater extent when the boundary was present did not show the same advantage when finding the target compared to trials on which the boundary was not present.

Discussion

To test the hypothesis that older adults have a selective deficit in allocentric navigation, we examined spatial strategies in a sample of young and older adults as they navigated a virtual Morris watermaze (MWM) environment. This paradigm has a long history of being used to examine spatial strategies and age-related differences in spatial memory in rodents (Barnes et al., 1980; Gallagher et al., 2015). Studies using virtual adaptations of the MWM in humans suggest that older adults often experience navigation difficulties. This is often suggested to be linked to a selective deficit in hippocampal-mediated allocentric wayfinding strategies, which is frequently accompanied by a shift toward a preference for egocentric-based navigation strategies in older age (Colombo et al., 2017; Li & King, 2019). It bears special mention, however, that egocentric and response-based strategies have been conflated in some past studies of the MWM and debate remains regarding the extent to which allocentric and egocentric measures can be cleanly dissociated (Ekstrom et al., 2014; Johnsen & Rytter, 2021). In the present study, we considered multiple measures of putative allocentric and egocentric navigation in order to reduce the potential influence of this confound and isolate the specific factors that might best account for age differences in navigation performance. Moreover, the design of our experimental conditions allowed us to test whether age differences in navigation performance are differentially affected by the inclusion of geometric spatial cues (boundary vs. no boundary) and/or idiothetic sensorimotor cues (immersive VR vs. desktop VR).

Evidence for age differences in putative allocentric vs. egocentric navigation preferences was mixed. Young and older adults did not differ when navigating from a novel start location compared to familiar start location on the spatial acquisition trials. The two age groups also did not differ when using distal landmarks to guide navigation during the delayed probe trials. However, older adults were more likely than younger individuals to replicate previously learned paths when navigating from a novel start position, consistent with a preference for familiar routes and egocentric navigation. The size of this age effect, however, was significantly attenuated in the immersive version of the task, particularly when a circular boundary cue was included within the environment. Taken together, our results suggest that the effects of age on spatial navigation are complex and cannot be fully explained in terms of a simple ‘deficit’ in allocentric navigation.

The clearest evidence in support of an age-related preference for egocentric strategy use was observed with our path replication analyses. Older adults were more likely than young adults to replicate a well learned path when switching between familiar and novel start positions. As described in the foregoing paragraph, however, these age differences were significantly attenuated when tested in the immersive VR condition, particularly when a boundary cue was present. Interestingly, this age effect was driven by younger individuals. Whereas older adults did not reliably differ between the two VR conditions, young adults were more likely to replicate a previously learned path in the immersive version of the task, when the full range of ambulatory cues were available (see Figure 3, bottom panel). Body-based cues have been suggested to exert a disproportionate influence on egocentric tasks that require monitoring one’s own position in space (Huffman & Ekstrom, 2021; Tuena et al., 2022). The enriched ambulatory cues in the immersive VR condition may have biased young adults towards repeating more familiar paths when navigating to the hidden target locations. An alternative (though not mutually exclusive) possibility is that young adults were more likely to adopt landmark-dependent strategies when navigating in the absence of body-based cues in the desktop VR environment, resulting in a greater tendency to navigate novel paths when starting from a new start position. Though speculative, this interpretation is consistent with our observation that landmark error was reliably reduced in the primarily visual desktop VR condition compared to the ambulatory immersive VR condition (see Fig 4A).

Figure 4.

Angular distance between participant heading and (A) distal mountain landmarks and (B) hidden target location at the time a memory response was made during the delayed probe trials. The polar histograms in the left panels illustrate the distribution of landmark and target error (in degrees) collapsed across the VR conditions and cue types. The bar plots in the right panels illustrate the circular means of landmark and target error (in radians) separately for each age group, VR condition, and cue type. The error bars represent the 95% HPD derived from the circular mixed effect regression models.

It is not readily apparent why path replication in older adults was not similarly affected by VR modality. One possibility is that the influence of body-based cues on egocentric strategy preferences are weakened in older age, potentially due to reduced sensorimotor acuity or changes in gait patterns and functional mobility (Hill & Ekstrom, 2025). Alternatively, older adults may have been less likely than young individuals to navigate novel paths in the desktop VR condition due to a comparative lack of computer fluency and gaming experience (Charness & Boot, 2022; Hill et al., 2024). Yet another possibility is that age differences in path replication reflect a general failure in older adults to accurately update their spatial position orientation towards fixed landmarks when navigating from a novel location (Merhav & Wolbers, 2019; Rosenzweig et al., 2003). Nevertheless, the path replication analysis suggests that older adults were more likely than their younger counterparts to stick with a familir route in both VR conditions, even when doing so was less efficient and more likely to result in spatial errors (i.e., when navigating from a novel start position). We caution, however, that this pattern of results should not be taken to infer a selective deficit in allocentric navigation, per se. Instead, these results may simply reflect a deficit in spatial updating or a preference for taking the more familiar, albeit less efficient, path.

Allocentric wayfinding strategies are typically evaluated based on the ability to use novel paths and shortcuts to reach a spatial goal. Critically, these novel paths are almost always designed to offer a more efficient alternative to taking a familiar but less direct route to reach the same goal. As a consequence, most experimental paradigms conflate path familiarity with path inefficiency. This represents a critical and often overlooked confound in studies assessing age-related differences in allocentric/egocentric navigation. Older adults are generally more risk averse than young adults, and are more likely to engage in familiarity-seeking behaviors to avoid more uncertain outcomes (Albert & Duffy, 2012; Frank & Seaman, 2023; Spreng & Turner, 2021). Thus, behaviors that resemble allocentric deficits in older age may simply reflect a preference for more familiar and predictable routes. Developing novel experimental methods to dissociate between path familiarity and efficiency are necessary to adjudicate between these respective sources of navigation errors in older age.

During the delayed probe trials, both young and older adults showed an advantage when recalling the hidden target from a novel compared to familiar start location. This novelty advantage was significantly greater in older adults compared to younger individuals when navigating in the desktop VR environment. The two age groups did not significantly differ, however, when navigating in the immersive VR environment. We reported a similar novelty advantage in an independent sample of young and older adults navigating an immersive VR MWM similar to that reported here (McAvan et al., 2021). We note, however, that this prior study did not involve a direct comparison with desktop VR and, unlike the current paradigm, provided continuous visual feedback during the delayed probe trials which may have enabled spatial updating during retrieval. One possibility is that navigating from novel locations introduced a degree of proactive interference, resulting in less precise spatial memory retrieval when navigating from familiar locations that were encountered during the spatial acquisition period. In other words, participants may have updated their spatial representations when tested from novel locations and this may have interfered with spatial memories that were encoded during the spatial acquisition period. At present, the reasons for this effect are unclear and additional work is necessary to replicate and further parse the source of this novelty advantage.

Impaired spatial navigation in older age has also been proposed to stem from inefficiencies in using discrete landmarks to learn and subsequently navigate spatial environments (Colmant et al., 2024; West et al., 2024). This form of ‘landmark-dependent’ navigation is often suggested to rely on hippocampal mediated allocentric computations (Lester et al., 2017; Li & King, 2019). To examine whether older adults were less efficient than young adults at using the distal landmarks, we computed the angular distance between participant heading at the time a memory response was made on the delayed probe trials, and the heading of the closest distal mountain cue. Critically, young and older adults did not significantly differ in their tendency to orient towards one of the distal landmarks, with older adults even showing a slight numerical advantage. Moreover, both age groups were more likely to orient towards the distal mountain cue than the actual location of the hidden target.

It is worth noting, however, that the statistically equivalent orienting to landmarks in young and older adults does not rule out the possibility that older adults were simply relying on a single distal cue as a beacon to guide spatial memory. To address this possibility, we rotated the distal mountain cue closest to the hidden target location on a subset of the delayed probe trials. We reasoned that those using a beacon strategy would rotate their memory response commensurate to the rotation of the distal landmark. Spatial performance on the rotated mountain trials did not differ between young and older adults, suggesting that neither age group was more likely to rely single landmark during navigation. Moreover, performance on the rotated mountain trials predicted spatial distance error on the delayed probe trials, and this association was independent of age. Taken together, our results suggest that observed age differences in spatial memory could not be explained soley in terms of a deficit in landmark-dependent navigation (c.f., Colmant et al., 2023; West et al., 2023).

It has been proposed that healthy older adults may preferentially rely on geometric spatial cues over discrete landmarks to navigate. Indeed, recent studies suggest that age differences in allocentric navigation are eliminated when geometric cues, such as environmental boundaries, are available to guide navigation (Bécu et al., 2019; Bécu et al., 2023). We tested this idea by having a subset of young and older adult participants navigate a version of the MWM environment that included a circuilar boundary wall, which served as a geometric cue. Young and older adults were less likely to replicate familiar paths when navigating from a novel start location in the boundary condition compared to the no boundary condition. That is, both age groups were less likely to replicate a previously learned path when geometric spatial cues were included within the environment. We also note that, within the immersive VR condition, age differences in path replication were were eliminated in the boundary condition. We caution that this analysis was exploratory and should thus be interpreted cautiously. Nevertheless, the pattern is largely consistent with prior studies that suggest age differences in allocentric-based navigation strategies are attenuated when geometric cues are available to guide navigation (Bécu et al., 2019, 2023; see also Hill, 2023).

Inclusion of the boundary wall did not affect the tendency to orient towards the discrete landmarks, regardless of age or VR condition (Figure 4A, right panel). This result contrasts with findings from a recent eye tracking study that reported older adults spent less time fixating on landmarks when geometric cues were present (Bécu et al., 2023). We note that the circular boundary cue in our study - which was selected to mimic the circular enclosures often present in rodent (Eichenbaum et al., 1990; Gallagher et al., 2015; Morris et al., 1982) and human (Daugherty et al., 2014; Gazova et al., 2013; Iggena et al., 2023; Moffat et al., 2006) adaptations of the Morris watermaze - differed from the types of asymmetric geometry cues used in prior studies, such as the Y-maze (Bécu et al., 2023) and rectangular enclosures (Bécu et al., 2023). Asymmetries in the angles and lengths of rectangular boundaries convey both directional and distance information, which may have been sufficient to guide navigation even in the absence of discrete landmark cues. By contrast, the circular wall and distal mountain cues used in our study provided distinct yet complementary information about the environment. The distal mountain cues could be used to infer relative directions within the otherwise sparse environment but were too far (~750 meters) to allow for distance estimations. The circular boundary wall, while directionally uninformative, could be used to estimate relative distances within the navigable arena.

One possibility is that young and older adults use a combination of landmark and geometric cues to orient, but only when the respective cues provide complementary information about the environment, as was the case in our boundary condition. This interpretation may explain the performance advantage observed in both age groups when navigating in the boundary condition, as indicated by reduced angular target error (Figure 4B, right panel) and spatial distance error (Hill et al., 2024) compared to the no boundary condition. Consequently, reliance on any given spatial cue to guide behavior is likely lessened when alternative cues are also present to optimize spatial performance. This may account for our observation that the correlation between distance error (a proxy for optimal navigation performance) and landmark-guided orientation (as assessed on the rotated mountain trials) was significantly weakened by the presence of a boundary wall in both young and older adults (Figure 5C).

A principal aim of the present study was to identify whether age differences in navigation strategies are affected by the modality of the virtual testing environment. Prior studies of aging and navigation in humans have relied overwhelmingly on experimental paradigms that assess wayfinding on desktop computers or mobile platforms (Coutrot et al., 2018; Diersch & Wolbers, 2019). This type of ‘desktop VR’ approach offers several advantages, including relative ease of implementation, minimal space requirements, reduced need for specialized equipment, and compatibility with functional neuroimaging environments, such as fMRI. Desktop VR is also effective when examining many of the purely cognitive phenomena that support successful navigation, such as spatial memory or landmark perception. A major limitation of desktop VR is its restriction of natural body-based movements and sensorimotor feedback (e.g., vestibular, proprioceptive, motor efference copy), both are which are critical for real-world navigation (Hill & Ekstrom, 2025; Plácido et al., 2022). Furthermore, cross-sectional studies of navigation may be confounded by group differences in visuo-motor dexterity and/or prior experience with computerized gaming equipment (Charness & Boot, 2022; Hill et al., 2024). The ability to freely walk and rotate one’s body in an immersive VR environment can provide a more ecologically valid platform for capturing dynamic cognitive and sensorimotor interactions during more naturalistic navigation behavior (Hill & Ekstrom, 2025) in a carefully controlled laboratory environment.

Nevertheless, we cannot rule out the possibility that age differences in prior exposure to VR headsets and other forms of immersive media may have influenced the observed age differences in navigation strategies, particularly path replication. Factors such as the restricted field of view (~110 ° horizontal FoV in the VR headset vs. ~180 ° in real-world binocular vision) or mild discomfort from dizziness or eye strain may have disproportionately affected navigation performance in the older adult group. Examining spatial behavior in real-world contexts could help address some of the limitations of immersive VR. We note though that while this approach would offer the highest ecologicaly validity, it would also sacrifice the experimental control and manipulation achievable when using immersive VR techniques in a controlled laboratory setting.We acknowledge that many of the foregoing interpretations are based on non-significant effects of age on allocentric navigation and that null results do not necessarily reflect the true absence of an effect. Follow-up studies are therefore necessary to indepenently replicate and validate our findings. We are encouraged, however, that the present findings are largely consistent with those reported by a previous study from our laboratory that failed to identify age differences in putative allocentric strategy preferences in a separate sample of young and older adults navigating in immersive VR (McAvan et al., 2021). We also note that null effects of age on spatial strategy were typically supported by strong evidence for the null based on Bayesian statistics.

To summarize, spatial navigation is a complex and multifaceted behavior that draws on multiple parallel cognitive, sensory, and motor systems (Ekstrom & Hill, 2023). Past work has suggested that impaired spatial navigation in older age may arise primarily from an allocentric deficit. As we have described above, however, allocentric and egocentric navigation strategies are difficult to isolate and there is also growing evidence that allocentric and egocentric sensory signals elicited during navigation recruit partially overlapping brain circuits (for review, see Wang et al., 2020), further complicating attempts to dichotomize these variables. Here, we show that older adults were more likely than young individuals to replicate a familiar path, even when this type of route-based egocentric strategy was less efficient and more likely to produce spatial errors. This age difference was significantly attenuated, however, when geometric and body-based motor and vestibular cues were available during navigation. Critically, we did not identify any evidence to suggest that spatial errors in older age stem from a failure to attend to stable landmarks within the environment, which is a key component of putative allocentric navigation. These results underscore the limitations of using a simple allocentric-egocentric dichotomy to explain the complexity of spatial behaviors observed in older age.

Figure 2.

Distance error ‘switching costs’ when starting from familiar start position compared to novel start position. (A) On study trials, distance error was greater when starting from a novel position. The size of this effect was not moderated by age group, VR condition, or cue type. (B) On delayed probe trials, distance error was greater when starting from a familiar position. This effect was significantly larger in the absence of the boundary cue. Likewise, age differences in the magnitude of these novel-to-familiar switching costs were significant in the desktop VR condition, but not when navigating in the immersive VR environment.

Public Significance Statement:

Spatial navigation, which is essential for well-being and independence, declines during healthy aging. A prevailing hypothesis is that impaired navigation in older age stems from selective deficits in allocentric-based wayfinding strategies. In the present study, we used advanced virtual reality techniques to reveal that age-related differences in spatial navigation are more nuanced and likely arise from a combination of factors, rather than being solely attributable to impairments in allocentric navigation.