Twenty-one older adults (9 males and 12 females, age = 74.4 ± 5.3 years) and 19 younger adults (16 males and 3 females, age = 25.1 ± 5.0 years) participated. Each older and younger adult was randomly assigned to one of two training groups: the intervention group (older adults: n = 10; younger adults: n = 10) or the control group (older adults: n = 11; younger adults: n = 9). We checked on a self-reported basis that all participants had normal or corrected-to-normal vision, no current musculoskeletal injuries, and no neurological disorders. Testing was approved by the Ethics Committee of Tokyo Metropolitan University, Japan (H31–98). Written informed consent was obtained from all participants in accordance with the Ethics Committee of Tokyo Metropolitan University and the Declaration of Helsinki.

The tasks, apparatus, and procedures were the same as those used by Kondo et al. (2021), with some modifications. The experiment consisted of four parts (Figure 1): measurement of participants' details, a pre-training test in a real environment, training in a VR environment, and a post-training test in a real environment.

Participants' height and shoulder width were measured in cm, and participants' weight was measured in kg. For older participants, cognitive function was assessed using the Mini–Mental State Examination (MMSE) (Folstein et al., 1975), and the mobility function was assessed using the Timed Up and Go (TUG) test (Podsiadlo and Richardson, 1991).

We conducted tests of pre- and post-training performance along a straight 5.5 m path in a real environment (Figure 2). A custom-made, moving doorway apparatus (Uchidadenshi Co., Japan) was made of two long boards (0.6 m wide × 1.75 m tall) and two short boards (0.6 m wide × 0.6 m tall) and was located 4 m from the starting position. The optical switch, which was used to widen the doorway before participants passed through the aperture to avoid collision, was positioned 0.7 m in front of the aperture. To prevent any feedback to participants regarding their success or failure, the aperture was made wider just before participants passed through it (0.7 m from it) so that actual collisions did not occur. Participants held a horizontal bar while walking. The horizontal bar consisted of a long rod (91 cm wide, 1.6 cm in diameter) and two extra rods (23 cm long, 2.3 cm in diameter) that were attached perpendicularly to the long rod (18 cm apart from each other) with T-shaped connectors (Figure 3A). To maintain the horizontal alignment of the bar and the body in the front dimension while walking, participants gripped each T-shaped connector with one hand and anchored the edges of the rods to the body at the chest. A three-dimensional motion analysis system (OQUS 300, Qualisys, Sweden) with 13 cameras was used to analyze kinematic data relating to the behavior of walking through an aperture. Five reflective markers on the horizontal bar and six reflective markers on the moving door were tracked.

Participants tried to pass through the aperture without collision and were allowed to rotate their shoulders/trunk, but only by as little as possible at the time of crossing (Figure 2). Three different aperture widths were randomly presented: 0.9, 1.0, and 1.1 times the bar width held by the participants. Participants performed a total of nine main trials (three trials for each of three aperture widths). Prior to performing the main trials, participants performed three practice trials (one trial for each of the three aperture widths) to familiarize themselves with the task.

Computer graphics of a walking path and an aperture were created with VR-authoring software (Omega Space, Solidray Co., Ltd., Japan). These images were projected using a 3D stereo projector (Sight 3D, Solidray Co., Ltd., Japan) onto a 3.25 m wide × 2.44 m tall screen. Participants wore stereo shutter glasses (3D vision, NVIDIA Corp., USA) with three reflexive markers (Figure 3B) to view the stereoscopic images. Participants also wore knee pads with three reflexive markers for each to track their stepping behavior (Figure 3C). The same horizontal bar (91 cm in length) was used in the VR environment as in the real environment. The reflexive markers were tracked using a 3D motion analysis system (OptiTrack Flex13, NaturalPoint, USA) with five cameras. The 3D motion analysis software (Motive, OptiTrack Japan, Ltd., Japan) was used to reflect participant's movements obtained from the 3D motion analysis system in a VR environment with a time delay of about 65 ± 16 ms. A custom-made vibration stimulator (Solidray Co., Ltd., Japan; Figure 3D) was placed on either the right or left wrist to provide vibratory stimulation (the frequency was 116 Hz) as tactile feedback about virtual collisions.

The procedures for the VR training of the intervention groups were as follows. In the intervention groups, participants were asked to stand 1.5 m in front of the large screen and start stepping in place as if the VR image on the screen were moving in response to their stepping (Figure 4). The VR image was presented to them with a perception of forward self-motion at 1.00 m/s, which was slower than in Kondo et al. (2021) (1.12 m/sec). During VR training, participants held the long horizontal bar. Participants were asked to pass through the aperture without collision and to minimize the spatial margin created between the bar and the inner edge of the board. The aperture width was 0.9 times the length of the bar held by participants. Participants performed a total of 35 main trials for the aperture width. When a virtual collision between the horizontal bar and either the right or left side of the inner edge of the board occurred, both visual and tactile feedback was presented to the participants. The visual feedback was the phrase “Left HIT” or “Right HIT” presented at the top left or top right of the screen, respectively. The tactile feedback, which was newly introduced in the present study, was a 0.2-s vibration applied to either the right or left wrist. Vibration stimulators were placed around the wrist rather than the arm to mimic the tactile information produced by the contact between the hand-held bar and the edge of the aperture.

The word “Good!!!” was also displayed on the screen as positive visual feedback when the participants successfully met a criterion. The criterion was to walk through an aperture while the spatial margin created between the inner edge of the board and the hand-held bar was within the critical value for each trial. The spatial margin was measured on both sides of the doorway; positive feedback was given to the participants when the spatial margin calculated on both sides met the criterion. The critical value was initially set to 20 cm and then changed gradually according to the following rules: (a) When the participants successfully met the criterion, the critical value was reduced by 2 cm (e.g., when they successfully met the first criterion, the critical value was changed to 18 cm). The critical value was reduced by 2 cm each time participants met the criterion. Theoretically, the criterion value became smaller until reaching zero if participants continuously meet the criterion. (b) When they were unsuccessful, the critical value remained unchanged. (c) When they failed to meet the criterion for two consecutive trials, the critical value was increased by 2 cm, but the value then remained unchanged even if the participants further failed to meet the criterion. The latter rule was necessary to prevent meeting the critical value from being too easy as a result of the participants' trial and error.

The procedures for participants in the control group were basically the same as for those in the intervention group, except that the width of the aperture was always 2.0 times the bar width, so that body rotation was unnecessary to avoid collision, and no feedback was presented. Prior to the main trials, participants performed nine practice trials to familiarize themselves with the task (three trials: the width of the aperture was 2.0 times the bar width; six trials: the width of the aperture was 0.9 times the bar width).

A Mann–Whitney test was used to compare participants' characteristics (age, height, weight, shoulder width, and MMSE and TUG scores), excluding gender. A Pearson's chi-squared test was conducted to compare the gender ratio.

To determine whether collision-avoidance strategies were improved in terms of both safety and efficiency after the VR training, two main dependent variables, the collision rate and the spatial margin, were collected during the pre- and post-training test phase. The collision rate during each phase was defined as virtual collision between the inner edge of the board and the horizontal bar because the aperture opened wider before the participants passed through it, and physical contact with the board did not occur (Kondo et al., 2021). The spatial margin, which was defined as the distance between one inner edge of the board and the edge of the horizontal bar on the same side, was calculated at the time of aperture crossing on one side of the horizontal bar (Higuchi et al., 2012). The spatial margin was obtained only from successful trials (i.e., trials with no collisions).

To investigate more deeply the behavior of walking through an aperture, two additional dependent variables, the mean absolute angle of body rotation and the relative distance between mid-body and mid-doorway at aperture crossing, were also collected only from successful trials during the pre- and post-training test phase. The absolute angle of body rotation at the time of aperture crossing was defined as the absolute angle between the horizontal bar and the door (Kondo et al., 2021). To determine whether the mid-body position became closer to the inner edge of the board at the time of doorway crossing, the relative distance between the midsagittal position of the body and the center of the aperture at the time of doorway crossing was obtained for each pre- and post-training test phase (Higuchi et al., 2006). Positive values indicated that participants passed closer to the inner edge of the board than to the center of the aperture, and negative values indicated that participants passed farther from the center of the aperture. The smaller the value, the farther from the inner edge of the board their passing position.

The main dependent variables (collision rate, spatial margin, absolute angle of body rotation, and relative distance between mid-body and mid-doorway) were statistically tested using a three-way (age × training group × phase) analysis of variance (ANOVA) with repeated measures by phase (pre- and post-training tests). Regarding the collision rate, the distribution was not normal, which negated the assumption of using the ANOVA. To deal with the issue, we adjusted the collision rate data by using the arcsine transformation for the analysis.

The main hypothesis of the study—that VR training could help participants to transfer the behavior learned in the VR environment to real walking—was related to the two-way interaction between the effect of the training group and the phase interaction. For this reason, post-hoc comparisons were performed using the Bonferroni method only when significant interaction was found for the training group × phase interactions (significance level was corrected to 0.0125). The significance threshold was set to 0.05. The software package SPSS (version 27.0) was used to conduct all statistical analyses.

To discuss the quality of the VR training for participants in the intervention group, their performance during the training session was evaluated using two measurements. First, to see how well participants met the critical value throughout the training trials, the mean value of the critical value was sampled. Second, the positive feedback signals (i.e., the trials in which the word “Good!!!” was given as feedback) throughout 35 training trials were counted. To verify whether improvement was observed in the former or latter half of the trials, the positive feedback signals in the first half (trials 1–17) and in the second half (trials 18–35) were also counted. Third, the behavioral characteristics of the participants during unsuccessful trials (i.e., the trials in which collision occurred) were described in terms of the absolute body rotation angles and the relative distance between mid-body and mid-doorway at aperture crossing. No statistical analyses were made regarding these four measurements.

The participants' characteristics are summarized in Table 1. For both older and younger adults, no significant differences between the training groups were found in any of the measurements. The MMSE scores of all older adults were above 24 points, which has been used as a cutoff value for cognitive function (Folstein et al., 1975), and the TUG scores were below 13.5 s, which has been used as a cutoff value for motor function (Shumway-Cook et al., 2000).

The mean collision rate is shown in Figure 5. Because we used an arcsine-transformed data for the statistical analysis, we present an additional figure as Supplementary Figure 1 to show that the distribution of the transformed data was comparable to that of the original one. An ANOVA for the adjusted data using the arcsine transformation showed a significant two-way interactions of the training group × phase [F_(1,36) = 7.37, p < 0.01, ηp2 = 0.17]. To test the study's hypothesis, post-hoc comparisons of the interactions were performed. During the post-training test, the mean collision rate in the control group was significantly higher than that in the intervention group. In the control group, the mean collision rate in the pre-training test was significantly higher than that in the post-training test. Two-way interactions of age × training group [F_(1,36) = 7.97, p < 0.01, ηp2 = 0.18], three-way age × training group × phase interaction [F_(1,36) = 12.24, p < 0.01, ηp2 = 0.25], and a main effect of age [F_(1,36) = 8.95, p < 0.01, ηp2 = 0.19] were also significant.

The mean spatial margin when crossing the aperture is shown in Figure 6. An ANOVA showed a significant two-way training group × phase interaction [F_(1,36) = 15.02, p < 0.01, ηp2 = 0.29]. Post-hoc comparisons showed that, in the intervention group, the mean spatial margin was significantly smaller in the post-training test than in the pre-training test. During the post-training test, the spatial margin was significantly smaller in the intervention group than that in the control group. A significant main effect of phase was also found [F_(1,36) = 19.65, p < 0.01, ηp2 = 0.35].

The mean absolute angle of body rotation is shown in Figure 7. An ANOVA showed a significant two-way training group × phase interaction [F_(1,36) = 10.55, p < 0.01, ηp2 = 0.23]. Post-hoc comparisons showed that, in the intervention group, the mean body-rotation angle was significantly smaller in the post-training test than in the pre-training test. A significant main effect of phase was also found [F_(1,36) = 25.44, p < 0.01, ηp2 = 0.41].

The relative distance between mid-body and mid-doorway is shown in Figure 8. An ANOVA showed a significant two-way training group × phase interaction [F_(1,36) = 6.89, p < 0.05, ηp2 = 0.16]. Post-hoc comparisons showed that, in the intervention group, the relative distance was significantly greater in the post-training test than in the pre-training test. During the post-training test, the relative distance was significantly greater in the intervention group than that in the control group. There were also significant main effects of age [F_(1,36) = 9.35, p < 0.01, ηp2 = 0.21] and phase [F_(1,36) = 5.63, p < 0.05, ηp2 = 0.14].

In the older adults, the mean of critical values for each practice trial in the intervention group is shown in Figure 9. Visual inspection of the figure shows that the critical values gradually became smaller until around trail 20. The number of positive feedback signals in a total of 35 practice trials was 11.1 ± 3.1. Of these, 7.8 ± 1.8 occurred in the first half (trials 1–17), and 3.3 ± 1.9 occurred in the second half (trials 18–35). The total number of unsuccessful trials (i.e., trials in which collisions occurred) observed for older participants was 38. In 36 of 38 trials (94.7%), the collision occurred at the “trailing” side of the body (i.e., the side of the body opposite that with which they approached an aperture with body rotation). The mean angle of body rotation was 33.84 ± 14.94 deg., which was slightly smaller than that for successful trials (38.14 ± 16.10 deg.). The relative distance between mid-body and mid-doorway was −9.21 ± 5.69 cm, which was slightly smaller than that for successful trials (−6.02 ± 4.96 cm).

In the younger adults, the mean of critical values for each practice in the intervention group is shown in Figure 9. The trend was similar to that in older adults. The number of positive feedback signals from a total of 35 practice trials was 11.9 ± 3.1. Of these, 7.6 ± 2.2 occurred in the first half (trials 1–17), and 4.3 ± 1.8 occurred in the second half (trials 18–35). The total number of unsuccessful trials observed for the younger participants was seven. All of these occurred on the trailing side of the body. The mean angle of body rotation was 45.01 ± 5.41 deg., which was slightly greater than that for successful trials (39.64 ± 10.86 deg.). The relative distance between mid-body and mid-doorway was −13.68 ± 2.14 cm, which was slightly smaller than that for successful trials (−8.61 ± 4.84 cm).