Effects of Environmental Clutter,Head-Mounted Display Clutter,and Head-Mounted Display Transparency on Human Efficiency,Safety,and Mental Demand During Navigation

Participants

Priori power analysis in G*Power 3.1 (Faul et al., 2009) indicated that 28 participants would be required to achieve 80% power to detect a medium effect size (f = 0.25) in a repeated-measures ANOVA. Therefore, we recruited a total of 30 volunteers (17 male, 12 female, and 1 nonbinary; mean age: 24.3 years; age range: 18–38 years) from the Colorado State University community. All participants had normal or corrected-to-normal vision, no history of vestibular or neurological disorders, and minimal prior motion-sickness susceptibility. Twenty participants received a $20 Amazon gift card; the remainder participated without compensation. The Colorado State University Institutional Review Board approved all procedures, and written informed consent was acquired following the Declaration of Helsinki.

Virtual environment and task

Participants were required to complete navigation tasks in a virtual environment. The primary objective was to guide the boat from a starting point to an end point (buoy) while navigating around obstacles, including wooden barrels, logs, and islands, in a virtual environment. Participants were instructed to minimize obstacle collisions and complete the task as quickly as possible. Participants operated a minimalist boat that moved only forward at three distinct speeds (500, 650, and 750 units/s), with yaw steering controlled via a joystick. Participants binocularly observed the virtual scene with a 120° field of view. They could freely rotate their heads while maintaining a stable environment and interface during navigation.

The virtual navigational environment covered a 500 × 500 unit grid, represented realistic navigation scenarios, incorporating natural terrain. Islands possess sandy shores, grassy zones, trees, and rocks, creating an organic coastal landscape (see Figure 1(a)). Two island-density setups were used: ten and twenty islands (see Figure 2). Islands were placed pseudo-randomly on a grid, such that their location was unpredictable from one run to the next, keeping a 20 m minimum distance. In the ten-island condition (low environmental clutter), irregular landforms created narrow channels and elongated paths, while the twenty-island condition (high environmental clutter) comprised smaller, circular islands in a maze-like pattern, needing frequent directional choices (see Figure 2). Contact with the buoy prompted a scene switch, indicating task completion (see Figure 1(b)).OPEN IN VIEWER

OPEN IN VIEWER

Figure 2.

Top-down schematic of the virtual maritime environment. (a) Low environmental clutter condition with 10 islands. (b) High environmental clutter condition with 20 islands.

In each scenario, 80 floating obstacles from the 11 obstacle types were dispersed pseudo-randomly throughout the virtual environment. The 11 floating obstacle types, including naval mines, wooden boxes, planks, and logs (see Figure 3). These obstacles were visually distinct on the water surface, allowing them to drift gently.OPEN IN VIEWER

Head-mounted display interface information

The HMD featured fundamental navigational details via a minimalist interface that maintained forward visibility. It displayed eight components: a minimap showing the vessel, endpoint buoy, and nearby terrain (but not floating objects); a speedometer (m/s); a distance-to-target gauge (m); a compass; a static weather icon; a battery level indicator; a voice-call icon; and a naval codename (e.g., “Alpha”) (see Figure 4). In the low display clutter condition, there were four relevant objects on the display: the minimap, speedometer, distance-to-target gauge, and compass were task-relevant. All of those objects could legitimately be used by participants to navigate. In the high display clutter condition, shown in the bottom row of Figure 4, four additional items were included that were not relevant to the participants’ task. The entire interface for low and high display clutter used less than 45% of the display area, where the minimap covered approximately 26%. Therefore, the environmental clutter impacted each of the interface elements, rather than any specific items. Moreover, the minimap was a zoomed-in version that translated and rotated with the boat to show a small section of the map around the user as they traveled.OPEN IN VIEWER

Independent variables

Dependent measures

Navigational performance was assessed using two primary dependent variables: completion time and proximity to hazards. Completion time (in seconds) was measured from trial onset (start-cue scene) to the endpoint buoy contact. We continuously assessed the proximity of the boat to objects or islands to detect both collisions and incidents classified as “unsafe close penetrations” within 8 m beyond the boundaries of these islands and obstacles (proximity to hazards). These values were recorded and saved by the Unity program for each trial that the participants performed.

Design and procedure

This study employed a fully within-subjects factorial design, with participants completing navigation tasks under all combinations of three independent variables (environmental clutter, display clutter, and transparency). Each of the eight conditions was presented twice, for a total of 16 trials per participant. The trial order was block-randomized to avoid consecutive repetition, and start/end locations varied to prevent route memorization. Before data collection, participants completed an informed consent form, a demographic questionnaire, and a baseline NASA-TLX assessment. Headset IPD was adjusted nearly to the measured interpupillary distance (IPD) with a pupilometer, ensuring maximum comfort. Participants were informed they could pause or request assistance at any time. Participants navigated while seated in a controlled lab setting with stable lighting and minimal distractions.

The training phase introduced HMD, controls, and collision/reset mechanics. In an environment with one hazard element (e.g., a house) and a target buoy in sand, participants practiced steering, speed modulation, and collision response. Advancement to reach the experimental trials required three successful completions and one intentional collision. After the training phase, participants performed the experimental trials. Participants were instructed to navigate to the buoy as rapidly and directly as possible, while still avoiding collision with any of the islands or floating hazards. The location of the buoy was randomly placed in such a way that the distance from the start position to the buoy was 450 m in each trial. The buoy could not be seen from the initial position. It could only be seen when the boat was 100 m away from the buoy in the minimap, as well as in the forward FOV when no obstacles were present.

After each trial, participants verbally completed the NASA-TLX survey. One-minute rests followed each trial, with a 10-min break at the midpoint. After all experimental trials were concluded, participants engaged in a brief debriefing session during which they reported any discomfort, nausea, or other adverse effects. Finally, participants received gratitude and appropriate compensation. The entire experiment lasted approximately 1 h and 15 min.

Technical specifications

Participants used an HTC VIVE Focus 3 standalone head-mounted display (HMD), which featured dual 2.88-inch LCD panels (2448 × 2448 pixels per eye, 90 Hz, 120° field of view). Steering was achieved with the right-hand controller joystick, acceleration with the trigger, and deceleration with the thumb grip. Virtual environments were entirely developed in Unity 3D (version 2022.3.35f1 LTS) within the HMD. A custom C# script recorded each participant’s task completion time and proximity data, saving the data to .csv files stored locally on the headset.

Analysis

Repeated-measures ANOVAs analyses were conducted to evaluate the effects of our experimental variables. All analysis was performed in R using ANOVA as well as Bonferroni tests. ANOVA analysis was done using the ez package, while the Bonferroni tests were done using the Desc-Tools package. For effect size, we report the partial generalized eta squared ( η p 2 ), with interpretations as follows: small ( η p 2 = 0.01), medium ( η p 2 = 0.06), and large ( η p 2 = 0.14). Based on the Shapiro–Wilk Test and histogram plots, both the performance metrics and the scores from the NASA TLX violated normality assumptions. Both were log-transformed to mend this violation. Despite these transformations improving the normalized distribution, they did not eliminate the violation entirely; however, ANOVA is robust to these violations when the sample size is moderate to large (Blanca et al., 2017; Schmider et al., 2010), the analysis reported here is a moderate sample size (N = 30). Statistical significance was measured using the standard alpha = 0.05 measure throughout the analysis.

Efficiency (completion time)

Completion time was measured in seconds, representing the time it took the participant to travel from the starting location to the target buoy at the other side of the navigation space. This data was processed by calculating the mean for each participant across the two repetitions; subsequently, the means were log-transformed.

The analysis of mean completion time depicted in Figure 5 showed that the time was significantly shorter with high environmental clutter (F (1, 29) = 17.20, p < .001, η p 2 = .04). Completion time was also significantly shorter when the display was opaque (F (1, 29) = 14.09, p < .001, η p 2 = .04). The significant 2-way interaction between environmental clutter and transparency (F (1, 29) = 42.46, p < .001, η p 2 = .12) indicated that this speeding with opaque displays was only observed under conditions of low environmental clutter (the right half in Figure 5).OPEN IN VIEWER

While display clutter did not affect completion time (F (1, 29) = 0.64, p = 0.43), the significant interaction between display clutter and transparency showed that high display clutter slowed completion time when the display was opaque (right panel), but not when the display was transparent (left panel; F (1, 29) = 39.37, p < .001, η p 2 = .10). The significant 2-way interaction between environmental clutter and display clutter (F (1, 29) = 14.11, p < .001, η p 2 = .04) shows that low display clutter was only observed to be faster under conditions of high environmental clutter. Subsequently, the three-way interaction between the variables showed a nonsignificant trend (F (1, 29) = 3.12, p = 0.08, η p 2 = .01) between the two-clutter condition interaction above, was present when the display was opaque.

Safety (proximity entry)

Proximity entry, violating safety constraints, was calculated whenever the boat crossed an 8-m limit surrounding either islands or floating hazards. These were counted for each trial. There were very few actual collisions with islands, so safety was defined only by the penetration of the safety zone around the islands.

Figure 6 presents the findings for unsafe proximity entry. ANOVA on the log-transformed data revealed that, as with completion time, high environmental clutter led to more proximity entries (F (1, 29) = 208.70, p < .001, η p 2 = .40). Proximity entry also increased when the display was more opaque (F (1, 29) = 7.77, p < .01, η p 2 = .02). The interaction between environmental clutter and transparency was not significant.OPEN IN VIEWER

The significant effect of display clutter showed that safety violations were higher for high clutter (F (1, 29) = 4.95, p < .05, η p 2 = .02). However the statistically significant interaction between display clutter and transparency indicated that higher display clutter increased safety violations only when the display was opaque, and not when it was transparent (F (1, 29) = 33.66, p < .001, η p 2 = 0.08). The 3-way interaction between the variables was not significant (F (1, 29) = 0.45, p = 0.51).

To follow up on this analysis, we examined separately whether the penetrations were related to islands or the floating hazards. Figure 7 depicts the same data as Figure 6, but now broken down separately by penetrations of floating objects (left) or islands (right). It is apparent that obstacle penetrations were minimal and not affected by the independent variables. Thus, the composite effects shown in Figure 6, are exclusively associated with penetrations of the unsafe space surrounding the islands.OPEN IN VIEWER

Mental workload TLX

While participants rated all six subscales of the TLX rating scale, in the current analysis, we focused exclusively on the mental demand scale. The other subscales were excluded because mental demand captured participants’ cognitive effort, decision making, and attention during the experimental task. Figure 8 presents the NASA-TLX mental demand rating as a function of clutter and transparency. An ANOVA on these data revealed that there was no effect of environmental clutter (F (1, 203) = 3.22, p = 0.074) but significantly higher workload when the display was transparent (F (1, 203) = 18.59, p < .001). There was also a statistically significant interaction between environmental clutter and opacity, such that low environmental clutter increased workload when the display was transparent, but not when it was opaque (F (1, 203) = 5.83, p < .05). Furthermore, during our debrief, 6 out of 30 participants reported nausea, blurry images, and dizziness as discomfort after finishing the experiment.OPEN IN VIEWER

Efficiency

We measured efficiency in terms of the environmental transit time. A preliminary analysis revealed that all of the influences of our three variables on transit time were imposed on the mean speed of travel, rather than the distance traveled. A slower speed, we reasoned, was a result of greater caution or less confidence in navigating the water safely, and hence a conscious choice to use lower throttle settings.

For each measure, we evaluated three hypotheses:

H1a proposed that increasing environmental clutter, here the number of islands would lead to slower travel. In fact, we found the reverse: more islands yielded faster transit, particularly and consistently when the display was transparent (display interface elements were hard to interpret), and hence environmental objects were highly visible in the field of view. We reasoned that this strategic choice to travel faster was guided by better visibility of environmental objects: if you can see them better, you travel faster. And if you can see smaller objects (Islands) more clearly, it further influences the decision to travel faster.

H1b proposed that transparency, and hence greater visibility of objects would slow travel because of greater caution exercised by the participants, and this was in fact found; particularly when there was low environmental clutter; that is, fewer, but larger, islands to be seen, a conclusion supported by the large interaction of the two variables, visualized in Figure 5.

H1c proposed slower travel with the more display clutter, an effect observed only with the opaque display (Figure 5) where the obscuration would “hide” more islands.

Safety

In terms of the penetrating safety violations with islands and hazards, our results revealed that nearly all of these, as well as the effects thereon, were witnessed by penetration of, or near collisions with islands, rather than with floating hazards.

H2a proposed more penetrations in a more cluttered, island-rich environment, and this was indeed strongly confirmed in Figure 6. Indeed, this effect of environmental clutter on safety penetrations could be predicted on purely statistical grounds: with twice as many islands (20 versus 10), participants choosing to navigate just within the bounds of safety nevertheless penetrated approximately twice as often (M = 2.7, SE = 0.17 vs M = 1.3, SE = 0.17). Figure 6 shows that this environmental clutter effect is relatively independent of the level of the other two variables, a pattern reflected by the absence of interactions.

H2b proposed that the opaque display, more obscuring of hazards, would increase the number of penetrations with those hazards, and, indeed, it did so. However, the interaction and the visual trend seen in Figure 6 indicated that this increased penetration rate was only observed under conditions of high display clutter, when, of course, there are more of those islands to be somewhat obscured by the opacity of the display.

Finally, H3 proposed that the increase in display clutter would partially obscure more environmental objects and hence lead to more “near collisions” with them, and this was indeed confirmed, although, as noted, only when those objects (islands) were more obscured by the more opaque display. Thus, display clutter and opacity both amplified each other’s degrading effect on safety.

The pattern of effects of the three variables on mental demand (Figure 8) appeared to closely mimic those on completion time (Figure 5), as did the pattern of statistical effects on the two variables. This would suggest that participants are highly sensitive to the factors that caused a delay, and use these as a proxy for their posttrial rating of mental demand.

Given the overall pattern of results, an HMD systems designer may wish to assess which combination of display features provides the most effective HMD design in nautical navigation. Here, the pattern of data provides no clear guidance, although they do suggest that when navigating in a highly cluttered environment, the poorest design for both efficiency and safety is an opaque display with high display clutter. However, such guidance is quite tentative at this point and requires considerably more research in which, in particular, more than a single level of transparency is examined. The current data also support the viability of theory-based metrics of clutter in predicting performance in this operational task. Collectively, the data reveal only partial success in this endeavor. Employing the well-validated object-count metric of clutter (Warden et al., 2024; Yeh & Wickens, 2001), our manipulations doubled the amount of both display and environmental clutter. Doubling environmental clutter (twice as many islands) produced a clear performance decrement of slightly more than doubling the number of unsafe proximity violations (Figure 7). However, it had no effect at all on completion time, suggesting the existence of a speed-accuracy tradeoff. It is quite possible that any negative effect of doubling the number of islands was offset by the benefit of increasing the ability to find a straighter path through them.

Doubling display clutter, and hence doubling the amount of either full (opaque) or partial (transparent) occlusion of the environmental hazards, appeared to have the predicted effect only when the display was in the fully occluded mode (right panels of Figures 6 and 7). Under this condition, the doubling of clutter nearly doubled the number of safety-violating penetrations and significantly increased completion time (no speed-accuracy tradeoff). However, the transparent display, whose goal was to decrease the clutter costs, expectedly reduced the clutter effects; but by effectively eliminating these costs altogether, it thereby rendered the clutter metric useless.

Limitations and future directions

One limitation of the current study was that the design of the environment, and particularly that the floating objects did not appear to be hazards at all, as their safety zones were never penetrated, and hence, floating hazard safety was not influenced by any of the independent variables. Future research will require a combination of increasing numerosity and varying the salience of such objects to create true hazards, akin to the runway incursions used in aviation HUD research (e.g., Fadden et al., 2001). Another limitation is the absence of eye movement data to assess the degree of use of various display elements and hence the extent to which these could be defined as necessary clutter. Another constraint on this research, also to be addressed in future experiments, is that we did not consider three other UI design factors, all deemed to be important in assessing the desired properties of HMDs for transportation systems:

• The similarity in hue between displayed information and the dominant elements in the scene (Warden et al., 2023b). For example, the perception of green imagery rendered by an HMD would be expected to be degraded when viewed against a blue-green seascape beyond.