“X-Ray Vision” as a Compensatory Augmentation for Slowing Cognitive Map Decay in Older Adults

2.1 Spatial Memory, Aging, and Cognitive Maps

Previous literature has revealed that older adults have significant difficulties in the process of forming cognitive maps due to a combination of age-related change/degradation of spatial abilities [19, 21, 26, 33]). The neural basis of age-related losses have also been identified in general memory structures (i.e., in the hippocampus, parahippocampal gyrus, posterior cingulate gyrus, parietal lobes, and pre-frontal cortex) that are thought to further impact the storage and maintenance of spatial information [1, 30, 35]. The storage of cognitive maps specifically may also be reduced in older adults due to general spatial memory limitations linked to changes in the density of the hippocampal formation [3, 36]. Given that effective human navigation relies on multiple sources of spatial knowledge and a complex neural architecture supporting the requisite spatial processes, it is not surprising that access to inaccurate cognitive maps could lead to error-prone navigation behaviors. Although the memory structures impacting spatial navigation are well-identified, the decay functions (i.e., the rate of decay over time) characterizing cognitive maps is not well understood. However, there is an abundance of research investigating basic memory retention based on college-aged populations. A comprehensive review by Rubin and Wenzel compiled, transformed, and analyzed retention functions for word recognition and free recall from 210 published data sets representing a large body of literature over many years [39]. In their work, they fit each data set to over 100 different mathematical functions. Logarithmic and linear trends were consistently well fit for most time intervals (especially for time delays on the order of days and weeks) and were the preferred trend of the authors. Of these two trends, the logarithmic function showed the highest significance for time intervals over days and weeks, which might suggest comparable decay trends for cognitive maps over similar time intervals. In the only known research studying cognitive map decay over time (i.e., tested longitudinally), participants (ranging from 20 to 80 years of age) freely explored and learned an outdoor space to criterion, after which the developed cognitive maps were tested immediately after learning, one day later, and one week later [6]. Though the existing research provides a glimpse of cognitive map decay, the present work using an extended test interval and a larger sample size than the previous work, allowing for a more accurate fit of mathematical functions to cognitive map decay data. Furthermore, the error magnitude in performance is also analyzed here between age groups and temporal effects are assessed through decay functions that have been well studied in previous memory research [39], a significant novelty of the work.

Addressing issues around changes in spatial abilities and cognitive aging is where technological solutions may play an important role. Although technology supporting spatial learning generally focuses on younger participants, behavioral results on indoor route completion tasks comparing younger and older adults with visual impairment demonstrated that performance for those over 60 years was nearly identical to a younger cohort, leading to almost 50% less errors when using a real-time, speech-based navigation system describing the route/environment compared to an information-matched, memory-based control [17]. Further supporting the efficacy of technological solutions for older adults, the 60+ age group in this previous study showed reliably greater improvements in route learning/completion accuracy and more confidence compared to the under 60 age group when using the system. These results suggest that technology providing context-sensitive environmental information during spatial learning and navigation, as is used in the current study, can reduce cognitive load on working memory and improve navigation performance, which may yield magnified benefits for older adult navigators.

2.2 Augmented Reality and Spatial Knowledge

Compensatory augmentations, particularly those that involve augmented reality (AR), have demonstrated positive user results among older adults in the driving context [25], during daily tasks (e.g., cooking) [43], and for fall prevention [7]. Notably, military work with AR has shown the technology is capable of improving spatial awareness [20]. In that work, the AR system provided indicators and text for important hardware that mechanics had to access to complete repairs, with results showing improved accuracy and speed of repairs. Additional research has found that using AR can improve spatial knowledge for visually obscured/complex information and serve as an enhanced learning tool [14, 23, 44]. Furthermore, AR has been found to provide improved spatial knowledge in various training/visualization [2], rehabilitation [9], and therapeutic fields [4]. Emerging work has also explored AR solutions that provide access to enhanced visual information by removing obstructions in the field of view (i.e., by providing an “X-Ray Vision” like experience). For instance, research by Li et al. explored the benefits of a similar X-ray augmentation for enhancing navigation performance of multi-level buildings in younger adults [27]. Results from Li et al. showed positive benefits of the augmentations during spatial learning, which supported more accurate spatial knowledge acquisition than was observed from environment-only (unaided) conditions. More recently, Matviienko et al. proposed BikeAR, an X-ray AR solution that improved identification of gaps between cars while cycling, while also reducing cognitive load [29]. It is important to note that these X-ray AR studies often rely on virtual environments to safely test the solutions and to provide insight while physical devices (like the previously mentioned X-AR solution [8]) are being developed. Similarly, instead of directly testing AR in this work, we use readily available VR environments to investigate the practical role AR-based augmentations could one day play. While the available literature provides strong support for AR in improving spatial awareness, including among older adults, no research to our knowledge has explored the important impact that AR augmentations could have on cognitive map development and retention among older adults, as is explored here using our simulated approach.

3.1 Participants

Forty-five participants completed this study, evenly split between three groups. The first group consisted of 15 younger adults (7 female, 8 male), ages 18 - 36 (M = 23.8, SD = 4.4), who served as a control and completed the study without the experimental augmentation. The second group included 15 older adults (9 female, 6 male), ages 60 - 78 (M = 67.9, SD = 5.7), who also completed the research without the augmentation and functioned as a direct (age-matched) control. The third group was comprised of 15 older adults (8 female, 7 male), ages 60 - 80 (M = 68.4, SD = 5.9), and served as the experimental group with the augmentation. Data for the older adult participants can be found in Table A1 of Appendix A. Recruitment for all participants was done through on-campus and local outreach. The research was approved by a university's local ethics committee and written informed consent was obtained from all participants. Participants were compensated for their participation by being entered into a raffle for a $50 gift card. Older participants were recruited between 60-80 years of age. To be eligible, all participants needed to self-report as independently navigating outside of their home at least once per week. Prior to starting the experiment, both older adult groups completed the Montreal Cognitive Assessment [35]. All participants scored equal to or greater than 26, indicating no abnormal cognitive impairment.

3.2 Apparatus

A traditional desktop VR system was used in the research, employing a Samsung 32” 720p LCD TV and a Mac Book Pro 17” laptop with an i7 processor and 8 GB of ram. Modeling for the virtual environment was done in-lab using the Maya 3D modeling software by Autodesk (www.autodesk.com). The experiment was programmed in Unity 3D and movement within the environment was controlled through keyboard input using the arrow keys.

The primary experimental layout for this study was informed by prior work involving older adults and cognitive maps using a two-box grid system in the form of a figure eight [6]. This approach enables a relatively simple layout, but also multiple routes to potentially navigate. Landmark locations, marked as specific rooms, were placed throughout the environment. Participants learned these four salient landmarks within the map (office, classroom, restroom, and stairwell). An additional landmark (starting location) was placed in the environment, and each participant began their learning phase at this start location, making five total landmarks. The rationale for using five landmarks was informed by early pilot testing, which indicated that five landmarks was adequately complex without exceeding working memory demands. See Figure 1 below for a top-down view of the hallways and landmark placement.

Figure 1.

View Figure

The aim of the compensatory spatial augmentation used here was to improve understanding of landmark relations through enhancing environmental perception of the user. This augmentation was accomplished by allowing X-ray-like visualization within the building structure that could be triggered using the “space bar” on the keyboard. Figure 2 below demonstrates the augmentation when toggled on and off. As can be seen, while traditional visual access would be blocked by walls, thereby occluding landmarks while navigating through hallways of a building, the augmentation eliminates this traditional navigational limitation of visual access, thus promoting stronger inter-landmark learning and better perception of global spatial relations. It is important to note that our implementation leaves floor and ceiling boundary markers in place (i.e., the X-ray effect only applies to wall boundaries), as the goal was solely to enhance inter-landmark relations within our single floor scenario.

Figure 2.

View Figure

3.3 Experimental Design

The experiment incorporated an initial in-lab learning and testing phase in the virtual environment, as well as a delayed testing phase that involved take-home trials performed at one day, one week, and two week intervals after the initial in-lab session. The rationale for these extended time intervals was that they built on the only known work exploring older adult cognitive map decay over time, which included one day and one week intervals [6], by doubling the max time period to two weeks. As is described in the following section, cognitive map development and retention was measured in both phases using an allocentric map recreation drawing task and an egocentric pointing task. Map recreation is considered a strong measure of evaluating cognitive map accuracy, as it requires the participant to explicitly recreate all global spatial relations of the learned environment, meaning that producing an accurate physical map reflects accessing an accurate cognitive map [18, 34]. Likewise, egocentric pointing, where participants draw lines or point in a specific direction from a reference location, has been shown to reflect the accuracy of inter-landmark knowledge contained in one's cognitive map. This additional test was included because previous research has demonstrated differences in egocentric pointing performance between younger and older adults [6, 16]. The map sketching and egocentric pointing data collected in-lab served as a baseline for comparison to the time delayed testing data that was collected one day, one week, and two weeks later.

3.4 Procedure

Participants began the experiment with a 5-minute practice phase. During this period, they practiced in a simplified building layout where they could adapt to the keyboard input, desktop VR experience, and experimental protocol. Any questions were answered and corrective feedback was provided during this training period as needed. Similar practice procedures have been used effectively in related work involving older adults learning outdoor virtual environments of similar complexity [5, 6]. After completing the practice session, participants proceeded to the experimental building layout.

During the learning period, participants were allowed unlimited exploration of the environment. They were notified once they had explored the entire layout at least once, but were encouraged to continue their exploration until they self-assessed as having a comfortable level of spatial knowledge of the environment as a whole, including the position of all landmarks. After completing free exploration, participants engaged in a learning criterion test, which involved drawing a top-down map of the environment, including the hallway network and placement of the four landmarks in the proper locations on the map. The start location (i.e., the starting point during the learning phase) was provided as a reference. If participants were unable to correctly recreate the learned environment, they were told their map did not meet the learning criterion (without any explicit feedback given) and were prompted to return to the VR building for an additional learning phase. The map recreation criterion was determined by comparing the drawn maps to the correct physical map of the space being learned (landmark and hallway layout placement). Specifications for this criterion were based on placement within a 10% threshold of the overall map-sketching space. Thus, meeting criterion required accurate landmark placement and hallway layouts to be within a 0.6 x 1.0-inch rectangular region centered on the correct landmark location or hallway intersection/turning point for the 6 x 10-inch drawing space of the sheet. Participants repeated this cycle of learning followed by sketch map completion testing until they had created an accurate map of the learning environment. The number of learning repetitions needed to meet criterion was also collected for each participant. Of note, ensuring that participants had developed an accurate baseline cognitive map is critical to the paradigm outlined in this work. This process avoids any potential cognitive map formation issues (age-related or otherwise) and guarantees that any subsequent testing is initiated based on a cognitive map that is known to be accurate and complete. As a result, any subsequent performance changes observed in the delayed testing phases are known to arise solely from issues with the maintenance process and not from original formation.

Once a correct map was created, participants performed additional egocentric pointing trials to ensure the veracity of the developed representation. These were done on paper, using a blank version of the sketch map sheet where only the boundaries of the learned space and starting landmark were provided (i.e., no internal connectivity or landmark locations were given). Participants were asked to imagine that they were at the home location, oriented to their starting direction, and to then draw a straight line representing the direction between the start point and each landmark location. Egocentric pointing was done using separate sheets for each landmark, the rationale being to discourage further inter-landmark learning.

The protocol for the compensatory augmentation condition was identical to the non-aided younger and older adult control groups, except for the addition of the augmentation. Participants were told that the augmentation could be triggered at any time using the space bar. Participants were also told that augmentations automatically triggered at every hallway intersection. This was done to ensure that the augmentation was guaranteed to be available a minimum amount of time at key decision points during the learning phase, but also allowed the participants additional opportunities to make use of the visualization.

Upon completion of the in-lab learning and testing phases, each participant was given three packets to take home with them. Each packet included the same mostly blank pages (i.e., without grid marks provided and just an indication for the starting location) used during in-lab testing for sketching their map and performing the egocentric pointing trials. Participants were then instructed to wait until approximately the same time on the following day to complete the first packet, one week from then at approximately the same time to complete the second packet, and two weeks from the in-lab session for completing the third packet. At those designated times, they were requested to open the packet and to complete all of the tasks it contained, thereby following the same procedure as was done in the lab. Each packet also included a stamped and self-addressed envelope in which all materials were mailed back to the lab upon completion. The testing packets were labeled with the participant numbers and designated dates for completion. Participants were instructed to complete the packets on the specific dates, but asked to report if there were any deviations from this pre-determined schedule. All packets were completed on schedule as self-reported by the participants.

Once received, sketch maps and pointing trials were analyzed for each time interval. The sketch maps were analyzed by comparing the participant's creation of the hallway layout and landmark placement (representing specific x-y coordinate points on the map) against the correct map. The analysis was performed by using sketch map analyzing software, Gardony Map Drawing Analyzer [15] based on a bi-dimensional regression analysis [13, 41]. This process uses the sketch map landmarks as points to form polygons that are then compared for translation, rotation, and scaling differences through a least squares method. Use of bi-dimensional regression has proven effective in previous work analyzing cognitive maps formed from multi-target arrays [16] and those developed by older adults after learning similar virtual environments as are employed here [6, 45].

To score the egocentric pointing data, responses for each trial were physically measured on the pages using a protractor. These values were then compared to the correct values for each trial in order to calculate signed error and absolute error. Signed error represents directional bias of responses and absolute error indicates the overall magnitude of the response errors. Learning times and learning counts, the number of learning phases required for a participant to form an accurate baseline sketch map, were determined from the data logs collected during the experiment. Bi-dimensional regression variables, all pointing trials, and all learning data were analyzed using a mixed model ANOVA with age group as a between-subject factor and time interval as a within-subject factor.

4.1 Method of Analysis

Participants sketched a map of the experimental environment immediately following learning in the lab and completed egocentric pointing trials at each of the four testing time intervals (in-lab, 1-day, 1-week, and 2-weeks). Group comparison for the sketch map bi-dimensional regression distortion index values and egocentric pointing errors (both measures described above) were analyzed using univariate ANOVAs with condition (younger adult (control), older adult (control), and older adult (augmented) as between-subjects variables. To further assess effects of the temporal delay, best-fit curve estimation was used to assess the pattern of change for both sketch map distortion values and egocentric pointing trials. Furthermore, Bonferroni corrected t-test comparisons between the experimental augmented condition and the control older adult group were conducted for each time interval to specifically evaluate the effects of the augmentations over time on the older adults. Learning times and the number of learning phases were recorded during the experiment and are reported at the end of the results section.

4.2 Group Differences: Sketch Map Distortion

To determine if there was an observable difference between the older adult and younger adult groups and if our augmentation improved cognitive map performance (Research Questions 1 and 2), we began by analyzing the participants’ sketch maps for distortion. The bi-dimensional regression we utilized uses the landmark coordinate points to create a polygon that connects them in order to calculate the overall distortion index (DI). Distortion index value means for all three groups can be found in Figure 3 below. Higher distortion values represent more error, with possible DI values ranging between 0 and 100.

Figure 3.

View Figure

Distortion index values represent a good measure of overall errors in the sketched maps (and by extension, of the underlying cognitive map). ANOVA results for DI values revealed significant group differences [F(2,42) = 7.98, p < .001] with a relatively medium effect size (η² = .09). Post-hoc comparisons revealed this effect was driven by the non-aided older adult control group performing significantly worse in comparison to the other groups (aggregated across all time points): non-aided younger adult control group [t(28) = 3.05, p = .002] and older adult augmented group [t(28) = 3.17, p = .001]. These results have two important implications: (1) they demonstrate that the magnitude of cognitive map decay for older adults (control) was significantly greater than younger adults (control) and (2) they show that the older adult control group experienced significantly greater cognitive map decay than age-matched older participants who had access to our compensatory X-ray augmentation. Of note, no significant differences were found between the younger adult group and the augmentation-aided older adult group (p > .05). This null result is encouraging, as it possibly suggests that exposure to the experimental augmentation was effective in mitigating the age-related differences in spatial performance observed between the younger and older groups not exposed to the augmentation, (acknowledging that lack of a reliable difference could also be due to random chance). Taken together, these outcomes suggest that although older adults may demonstrate faster rates of cognitive decay, this decay can be reduced with X-ray augmentations, perhaps to equivalence with the younger adult baseline.

Further group comparisons were conducted at each time interval to investigate whether the augmentation succeeded at reducing decay in performance over time. Test comparisons between the older adult augmentation group and the older adult control group at each time interval can be found in Table 1, with significant effects also represented in Figure 3.

As represented in Table 1, the distortion index values were significantly reduced with a large effect size for only the 1-day time interval among those with the augmentation compared to those in the control group. This means that the augmentation was successful at eliminating the large initial drop-off in performance, essentially slowing the sharp cognitive map decay that would be observed with a logarithmic function (discussed further below), but then having less of an impact on subsequent decay. As the augmentation was designed to support strong cognitive map development to combat the most susceptible period of immediate decay, we interpret this mitigation of the initial reduction in performance on spatial tasks compared to the control group as an encouraging result. In terms of the impact of these findings, this outcome speaks to the extent to which compensatory augmentations leveraging X-ray-like effects can improve spatial knowledge in the immediacy after learning.

Given these positive results, we were also interested to determine if the distortion index data indicated that the augmentation reduced variability of errors across individual participants. Levene's test for equality of variances provides a measure of differences in variability between samples (in this case the older adult augmentation group and older adult control condition). Differences between individual variabilities would indicate that performance from subject to subject was more tightly clustered around the central mean. Older adult populations tend to reflect higher variability, therefore reducing this variance and heterogeneity would represent a consistent effect of the augmentation for each participant. Findings here reflected significant differences in variability between the augmentation and non-aided older adult groups [F(1,28) = 9.13, p < .001]. These results indicate that the augmentation condition was not only successful at slowing cognitive map decay, but was also able to reduce the variability of errors across individual participants. This reduction in the variability of participant performance is surprising given that in general, older adult populations tend to exhibit greater inter-subject variability [26, 30]. The ability for the augmentation to reduce error and variability between participants is interpreted as strongly supporting two major outcomes related to the research questions: (1) the compensatory X-ray augmentation successfully supported stronger and more enduring cognitive maps over time and (2) the augmentation was beneficial for all of the older participants who used it. The second finding, demonstrated by the decrease in data variability in these older participants, is important for generalization of compensatory augmentations to application for a highly heterogeneous population.

4.3 Group Differences: Egocentric Pointing

To further evaluate the viability of the augmentation for older adults and the differences in cognitive map decay between older adults and younger adults (Research Questions 1 and 2), we analyzed error on the egocentric pointing trials by calculating the difference between response angles and correct angles. In-lab egocentric pointing errors were around 9.5 degrees for older adults and about 6.3 degrees for younger adults. Averaging error across the three delayed time intervals demonstrates that older adult means were approximately 16.0 degrees and younger adults were about 8.1 degrees. By contrast, the difference between age groups during the in-lab session was only about 3.2 degrees while the time delayed group difference was 7.9 degrees. The larger difference found between age groups at the time delayed testing likely reflects greater loss in egocentric spatial memory, but could also be influenced by declining motor memory necessary for accurate pointing behavior. This result reveals that the decay process creates strong vulnerabilities within stored spatial knowledge, impacting related spatial behaviors for older adults.

Before statistically evaluating the angular differences, an initial ANOVA was performed to ensure there was no directional bias within the egocentric pointing data. No significance was found for any of the measures, and no left/right bias (signed error) was observed for the pointing judgments. Average absolute egocentric pointing errors for all groups across the four time intervals can be found in Figure 4 below. Similar to the sketch map distortion data, ANOVA results for egocentric pointing errors revealed a significant effect of group [F(2,42) = 27.28, p < .001], with a relatively medium effect size (η² = .07). Post-hoc comparisons revealed that this effect was influenced primarily by the non-aided older adult control group performing significantly worse than the other groups (aggregated across all time points): non-aided younger adult control group [t(28) = 5.96, p = < .001] and older adult augmented group [t(28) = 5.58, p < .001]. No significant differences were found between the younger adult group and the older adult augmentation group. As with the distortion data, this null result is encouraging as it suggests that exposure to our compensatory augmentation could possibly improve older adult performance to a similar level as the younger adult group. These results as a whole indicate that although egocentric pointing for older adults was consistently worse than younger adults, as with the map recreation data, access to the compensatory augmentation led to reliably improved performance for older adults compared to an age-matched cohort without this access.

Figure 4.

View Figure

Group comparisons were conducted at each time interval to examine the effect of the augmentations over time. All t-test comparisons between the augmentation group and the older adult control group at each time interval can be found in Table 2, with significant effects also highlighted in Figure 4. T-test comparisons of the older adult control group compared to the older adult (augmentation) group revealed significant differences (p < .05) at the 1-day, 1-week, and 2-week time intervals, each around a medium effect size (with d around .5). This means that the initial and continued decay, characterized by the logarithmic function for the control condition (described below), was alleviated entirely across the time intervals tested here when the compensatory X-ray augmentations were available.

4.4 Characterizing Cognitive Map Decay Over Time

When investigating how to best characterize the decay function of cognitive maps (Research Question 3), we further evaluated both the map recreation and egocentric pointing data using a best fit curve estimation. For the map recreation distortion values, results indicated a significant logarithmic trend for the older adults across the time intervals [F(3,14) = 5.99, p = .018]. The predicted logarithmic function for the older adult distortion values was found to be y = 1.369*ln(x) + 24.626. This equation can be used to predict distortion values given an input of time (in weeks) for the value of x. For example, if input for the x variable is given as a non-tested time interval in weeks, the equation will predict how much angular error could be expected at that future point in time. Continuing with this example, entering 8 as the value of x (representing an 8-week temporal testing delay) results in a y value of about 27.5. Given a standard error of about 4.5, as found for the time delayed testing periods here, the expected range of response distortion index values could be estimated to be between 23 and 32 for an 8-week delayed testing point. Given the logarithmic trend found in the delayed testing data, this result would be in line with results found in the lab. The performance should continue to decline past the timeline tested in the present study, but at a slowing pace. Figure 5 below shows the individual distortion values across the four time intervals for the non-aided older adult group as well as the significant logarithmic trend line.

Figure 5.

View Figure

On the other hand, younger adult distortion index values were neither predicted by a linear or logarithmic trend (both p’s > .100). This outcome indicates that the numeric decay observed across the time intervals for younger adults was both non-significant and not readily predicted by a clear decay function as was found to be the case for older adults. Similarly, no significantly predictive trend was found for the augmentation condition, logarithmic [F(3,14) = .147, p = .232] and linear [F(3,14) = 2.18, p = .148]. The lack of a statistically-reliable trend potentially indicates that the augmentation reduced the pattern of decay revealed for the testing time intervals found in the non-aided older adult group, but could also be represented by another equation not tested here, due to random chance, or sample size (as discussed in our later Limitations section).

As with the distortion index data, the best fit function for older adult (control) egocentric pointing errors was a logarithmic trend [F(3,14) = 7.59, p = .006], while younger adults pointing errors showed no reliable fit for either linear or logarithmic trends (all p’s > .100). The logarithmic function fit to the egocentric pointing errors of older adults was y = 0.863*ln(x) = 16.279. As with the distortion index value analysis, this function can be used to predict an approximate value of egocentric pointing error at different time intervals. The logarithmic trend indicates that the most susceptible areas of decay are the short time intervals that this research studied. Including further time intervals would only continue to demonstrate the slowing rate of decay found from the logarithmic trend. The logarithmic function is based on input of time in the form of the number of days, as is illustrated in Figure 6 below along with the individual egocentric pointing errors for older adults.

Figure 6.

View Figure

Results of a best-fit trend analysis revealed no significant logarithmic trend for the augmentation group [F(3,14) = .236, p = .627] as well as no linear trend [F(3,14) = 1.01, p = .316]. As with the distortion index data, we speculate that the lack of a significant trend for this condition could possibly suggest that the augmentation technique yielded a reduction in egocentric pointing decay, but could also be represented by another equation not tested here, the result of the relatively small samples, or due to random chance.

4.5 Learning Data

Data collected from the learning phases included learning times and a count of how many learning phases were required for participants to meet criteria by forming an accurate baseline sketch map. The older adult control group learning times averaged about 6.3 minutes (M = 375.9 seconds, SD = 223.9 seconds) and the younger adult control group took about 2.3 minutes for meeting the learning criterion (M = 137.4 seconds, SD = 50.6 seconds). Learning times for the aided older adult group averaged 8.2 minutes (M = 506.3 seconds, SD = 225.1 seconds). Comparisons between the non-aided younger and older adult groups learning times revealed significant differences t(28) = 4.157, p < .001, with a relatively large effect size (d = 1.47). This outcome indicates that older adults consistently took longer to accurately learn the environments to criterion; approximately two and a half times as was needed by younger adults. Numerically, learning times were longer for the aided condition (8.2) than the non-aided condition (6.3 minutes), but this difference was not statistically significant. Although not reliably different, the learning times were highest when the augmentation was present during learning. Based on informal experimental observation, participants in this condition seemed to stop more often at decision points to survey the area and look back and forth between the various landmarks. Perhaps given the chance, participants wanted to take advantage of the improved visual access granted by the augmentation to form stronger inter-landmark relations. It should be noted that no significant correlation was found between learning time and map recreation accuracy for any group (younger control, older control, and older augmented) or time interval (in-lab, 1-day, 1-week, and 2-weeks).

At most, it took participants two learning phases to form accurate baseline sketch maps. The older adult control group averaged 1.3 learning phases, while the younger adult control group averaged only 1.0 learning phases to reach criterion. The augmented older adult group took 1.08 learning phases to reach criterion. Note that while there were numerical differences between the groups, no statistical significance was found. However, the older adult augmented group did have fewer learning phases when compared to the control older adult group, potentially indicating stronger learning and reduced need for additional exposures.

Overall, these results provide evidence supporting characteristics of age-related cognitive map decay and corroborate the a priori prediction that logarithmic trends best predict performance for older adults. The empirical findings also matched the expected outcomes in that the compensatory X-ray augmentation condition significantly reduced spatial memory degradation. The results of this study contribute to fundamental spatial cognitive aging theory and directly address the goal of identifying practical solutions to aid successful navigation through enhancing formation and maintenance of spatial memory of older adult navigators.

5.1 Cognitive Map Retention, Independence, and Quality of Life

A critical contribution of this work is the significant reduction in cognitive map decay over time among older adults who used the compensatory augmentation compared to the control group who did not. The reduction in group variability (differences between individual responses), as demonstrated by tighter clustering of data around the condition means, indicates consistent success of the augmentation among this demographic. Furthermore, we speculate that older adults who used the augmentation may have been able to retain cognitive maps to the same extent as the younger adult control group, as no significant difference was found between these groups and should be explored in future work. We argue that these findings are important to consider in light of navigation and independence challenges experienced by older adults globally, as older adults travel significantly less than younger adults [22]. Spatial navigation is well connected to independence, travel, and overall quality of life [12]. Consider that older adults often experience barriers in new places (e.g., confusion and sensory overload) that can be addressed, at least in part, by learning landmarks in the environment [38]. Our solution, which improves access to landmarks and the spatial relations between them, may therefore help address the differential travel patterns of older adults and younger adults/millennials. While our results indicate improved cognitive maps as a result of our experimental augmentation, more research is needed to explore how cognitive maps relate to confidence and actual travel behavior. Indeed, future work should focus on the qualitative outcomes and user experience of X-ray augmentations, which was not the focus of this study. Future work should also explore how improved cognitive map access varies across implementation scenarios (e.g., in multi-level buildings and outside scenarios, as discussed by [27]) and how X-ray augmentations would work in multifloored scenarios. Our VR-based approach purposefully kept the floor and ceiling as visible boundaries to reduce potential disorientation, similar to the recommendations in [10, 24], but should be explored in future real world applications. Accordingly, future work should also be conducted “in the wild” with emerging AR diplays opposed to a controlled virtual environment as was studied here. For instance, such displays could be used to rehearse navigation scenarios prior to travel to new locations (e.g., a doctor's office or medical complex) to increase confidence and further reduce the cognitive map decay curve. By so doing, a more complete picture of the real-world benefits of X-ray AR and related solutions can be realized.

5.2 Cognitive Map Decay Functions Over Time

Results based on our analysis of participants’ recreated sketch maps illuminated differences in the underlying cognitive maps of older and younger adults. Older adult cognitive maps were shown to be significantly more susceptible to distortion as compared to younger adults (control). The older adult cognitive maps also demonstrated increased decay over time, which was found to follow a logarithmic trend, as compared to their younger counterparts. These results add to the body of evidence suggesting that older adult cognitive maps decay in logarithmic fashion [6], while also contributing to the corpus of research utilizing bidimensional regression and map recreation tasks for analyzing spatial knowledge [6, 16, 37, 45]. Beyond the current data corroborating previous findings on spatial aging and extending established methodological approaches to the study of cognitive map decay, the true value of our findings are their application to the design of new gerontechnologies aimed at altering, slowing, and mitigating known deficits in spatial memory by older adults. These basic findings provide a clear roadmap for future technologies that may be designed to better visualize the environment and provide improved perceptual access to key spatial information that is known to be difficult for people (especially older adults) to apprehend, learn, and represent. The current findings demonstrate that when such compensatory tools are made available, the X-ray augmentation studied here being just one example, the natural process of memory and spatial ability decline can be slowed and possibly even reversed, particularly at the large initial drop-off at one day intervals (where both the distortion index and egocentric pointing data demonstrated significant effects) and potentially beyond (where the egocentric pointing data alone demonstrated a significant effect). This work opens the door to the design of other such augmentations that could be developed to address preservation of other spatial skills known to degrade with age [26], or to optimize the current approach to have even longer/lasting effects, thereby leading to further amelioration of the decay function over time. An important innovation of this work was the longitudinal nature of the testing procedure, which can be expanded in future work and compared with other spatial scenarios. Many of the same underlying mental structures for spatial memory are used for storage and access of cognitive maps, meaning that the trend identified here with cognitive map decay in older adults may well be generalizable to other forms of spatial behavior (e.g., targeting/localizing, driving, or pattern recognition). Understanding the ways in which X-ray AR supports a multitude of spatial behaviors is crucial for designing new solutions across the spectrum of ability and user demographics.

5.3 The Future of X-Ray AR with Older Adults

The results from this research contribute to emerging work exploring the benefits of devices that enable users to see objects that would be otherwise occluded in various scenarios, such as while cycling [29] and through walls in poor lighting [28]. As the first known work to explore X-ray-like effects as a compensatory augmentation for older adults, we contend this line of research opens the door to future related studies, applications, and potential benefits. For instance, X-ray AR could hold the potential to improve outcomes for the millions of older adults who experience age-related visual impairment, a rapidly increasing demographic worldwide [11]. By combining the augmentations with multisensory cues provided in real time, navigation and cognitive load may well be improved, as demonstrated in [17]. Indeed, promising work with AR and people with visual impairment has shown the powerful real world utility that wearable smart glasses imbued with navigational guidance can have for this population [46]. As AR systems improve, the next generation of wearable and ubiquitous computing holds the potential to provide life-changing augmentations for older adults and people with sensory disabilities — our evidence here suggests that X-ray AR should be part of that equation.