Impact of Starting Position on Decision-Making in Virtual Reality

Valentina Gorobets

, Mathieu Lutfallah

, Khashayar Ilbegi Teymouri and Andreas Kunz

Institute for Machine Tools and Manufacturing, Swiss Federal Institute of Technology Zurich,

Clausiusstrasse 33, Zurich, Switzerland

{gorobets, lutfallah, kunz}@iwf.mavt.ethz.ch

Keywords:

Virtual Reality, Path Prediction, Natural Walking, Decision Making.

Abstract:

Few studies have explored how a user’s initial starting position or physical obstacles in reality affect decision-

making in virtual reality (VR), particularly when natural walking is used for locomotion. In this paper, we

examine how the starting position in the real world inﬂuences walking path decisions in VR. 24 participants

were positioned next to a physical wall before putting on a VR headset and then asked to walk through a

narrow virtual corridor, making a left or right turn at a decision point. To ensure safety, we employed redirected

walking techniques to subtly steer participants away from the real wall. Our results indicate that users remain

aware of their physical starting position, inﬂuencing their directional choices in VR.

1 INTRODUCTION

In recent years, VR has become more accessible to

the end-user and industry, being utilized not only for

entertainment purposes, but also for training. (Xie

et al., 2021) lists some of the possible VR training do-

mains: ﬁrst responder training, medical training, mil-

itary training, transportation, workforce training, and

interpersonal skills training. To effectively train peo-

ple in VR and later apply the results of VR training to

real-life scenarios, it is crucial to study the decision-

making process in VR and identify potential factors

that may inﬂuence it. One such factor can be a user’s

awareness of the physical boundaries of the real envi-

ronment they are in.

To explore virtual environments (VEs), various

locomotion techniques can be used, such as natural

walking, walking-in-place, treadmills, or teleporta-

tion (as shown in (Usoh et al., 1999)). Natural walk-

ing is the most intuitive of these; however, it also re-

quires enough physical space to allow users to freely

walk and navigate in the VE. Currently, VR research

does not concentrate on the effect of the starting po-

sition of the user in the real world on human behav-

ior in VR. However, previous research by (Interrante

et al., 2007) has demonstrated that the real-world

space where the virtual experience begins, impacts

the user’s experience in VR. Their ﬁndings indicate

https://orcid.org/0000-0002-8615-5972

https://orcid.org/0000-0001-7863-8889

https://orcid.org/0000-0002-6495-4327

that spatial perception accuracy in immersive virtual

environments can be inﬂuenced by starting in a real-

world space that is later replicated virtually. However,

to our knowledge, no work has explored the effects of

the paths users take in VR based on their initial phys-

ical position. While the effect of the starting position

in reality might be negligible when the implemented

locomotion technique does not require natural walk-

ing, we want to focus our research on natural walking

and investigate how a starting position may affect the

walking paths.

1.1 Spontaneous Alternation Behaviour

To better understand path decisions, it is important to

introduce the concept of Spontaneous Alternation Be-

haviour (SAB). SAB describes an animal’s tendency

to avoid repeating choices when exploring a maze.

(Nguyen et al., 2017) investigated the potential pres-

ence of this behaviour in humans using VR. To deter-

mine its existence, a study was conducted in which a

virtual maze of equal corridors, consisting of an initial

◦

forced turn, followed by three T-junctions, was

designed. A forced turn describes a turn in any di-

rection that is imposed on the user, meaning no other

path is available. The overall alternation rate in VR

was found to be 72%. Their studies showed that also

in humans, the decision-making in path selection is

inﬂuenced by previous walking experiences, such as

forced turns, given that no other visual effects inﬂu-

ence the the decision-making process.

604

Gorobets, V., Lutfallah, M., Teymouri, K. I. and Kunz, A.

Impact of Starting Position on Decision-Making in Virtual Reality.

DOI: 10.5220/0013180800003912

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2025) - Volume 1: GRAPP, HUCAPP

and IVAPP, pages 604-610

ISBN: 978-989-758-728-3; ISSN: 2184-4321

1.2 Path Decisions Inﬂuenced by Visual

Differences

(Abu-Saﬁeh, 2011) examined the effect of visual dif-

ferences, such as brightness and color, on decision-

making in a maze. Two different mazes were de-

signed for this purpose. One focused on brightness

differences between corridors, and the other on color

differences. The results indicated that there was a

high preference for brighter corridors, i.e., corridors

with windows, and cold-colored corridors (e.g., blue,

green, or white). (Vilar et al., 2013) explore the

impacts that width and brightness have on path de-

cisions. Four different experiments were designed.

All comprised T-junctions and F-junctions with var-

ious manipulations applied to them. Similar to (Abu-

Saﬁeh, 2011), they utilized brightness modiﬁcations

as well. However, contrary to (Abu-Saﬁeh, 2011), in-

stead of experimenting with colors, they researched

the inﬂuence of the width of a corridor. The re-

sults yielded a preference for wider corridors, with

ca. 70% of the choices being the wider corridor. Also,

the results showed a preference for brighter corridors,

with approximately 84% of the outputs for that ex-

periment preferring the brighter corridor. However,

when it came to choosing between brighter or wider,

brightness ended up having a stronger inﬂuence on

the users’ choices. The narrower but brighter corridor

was preferred over the wide but darker one. Addi-

tionally, when combining the properties, even higher

percentages of 87% were reached.

1.3 Redirected Walking

In our experiment, the user’s starting position is next

to the side wall. To seamlessly steer them away from

it, we use the Redirected Walking (RDW) technique,

as initially introduced by (Razzaque, 2005). Redi-

rected walking with gains involves subtly manipulat-

ing a user’s movements in VR by applying rotation,

translation, or curvature gains. When being applied

below a perception threshold (Lutfallah et al., 2024),

(Steinicke et al., 2010), these gains alter a user’s vi-

sually perceived direction, speed, and curvature with

regard to the real walking trajectory, enabling them to

navigate a larger VE within a limited physical space.

This creates a seamless and immersive experience by

preventing collisions with real-world boundaries.

While some research exists on the inﬂuence of

VEs on a human’s decision-making process for path

selection, no research exists on the inﬂuence of the

real environment, which the user sees before putting

on the headset, on their decisions regarding path se-

lection in the VE. Thus, our research ﬁndings will

sensitize future research on redirection by addressing

unwanted biases stemming from the real environment.

This paper ﬁrst describes the methodology of our ex-

periments, followed by a description of the user study.

We then discuss the results and give an outlook on fu-

ture work.

2 METHODOLOGY

The goal of this work was to investigate the inﬂuence

of the perceived starting position in the physical space

on user decisions when making decisions in a virtual

maze. To achieve this, we developed a VE with a T-

shaped corridor where the user traverses it using nat-

ural walking and must make a turn at the junction.

We developed the environment using Unity version

2022.3.16f1 with the OpenXR library. The VE was

streamed from a PC to the Pico 4 Enterprise via WiFi.

The available tracking space was 7m × 10m, in which

users always starting next to a physical wall (see Fig-

ure 1.

The layout of the implemented corridor can be

seen in Figure 2. We used a monochromatic texture to

eliminate any potential inﬂuences stemming from the

VE. The side passageways were identical and dimly

lit, so the user could not see the end. For the experi-

ments, one side of the corridor led to a physical wall

(critical turn (CT) in Figure 2) since the user starts

next to the environment boundaries. To allow the user

to walk a certain distance if that CT corridor is cho-

sen, we used curvature gains, which represent the in-

jection of rotation while the user perceives to walk

straight, subtly directing them away from the physi-

cal wall. The employed curvature gain was restricted

to a radius of 16 meters, which is more conservative

than the thresholds described (Steinicke et al., 2010).

Using such a wide radius also avoids triggering the

SAP behavior in humans, as described by (Rothacher

et al., 2020).

Since the awareness of the physical boundaries

and thus their effect on decision making in VR might

decay over time and walked distance in VR, the ef-

fects of different corridor lengths were studied in this

work. The lengths used were 7, 5, and 3 meters, and

the same experiment was conducted on both the left

and right sides of the room. Figure 3 shows the user

path along the corridor when curvature gains are ap-

plied. This resulted in total in 6 conditions for the user

study, and in 5 games between them.

Impact of Starting Position on Decision-Making in Virtual Reality

605

Figure 1: The tracking space where the user study has been conducted. Depending on the user’s starting position, the physical

wall is either on the left or the right side, hence resulting in left or right critical turn (CT).

Starting Position!

Physical Boundary!

Critical Turn

Figure 2: Placement of the VE corridor with respect to the

physical space boundaries.

Starting Position!

Physical Boundary!

Critical Turn

Starting Position!

Physical Boundary!

Critical Turn

Figure 3: Actual user path in the physical space due to the

employment of curvature gains.

3 USER STUDY

To gather data about user decisions in the imple-

mented VE, a user study was conducted. Participants

were recruited through a poster shared at the univer-

sity. The only criterion for participation was the abil-

ity to walk. The procedure of the study is shown in

Figure 4.

After welcoming the participants, they had to ﬁll

out a consent form. Then, they received an explana-

tion of the overall study sequence, consisting of the

introduction, the two parts of the questionnaire, and

the VR sessions. They then ﬁlled out the ﬁrst part

of the questionnaire which consisted of demographic

questions and the simulator sickness questionnaire

(SSQ) as presetned by (Kennedy et al., 1993). Fol-

lowing this, the user received a short tutorial about the

Pico 4 Enterprise and its controllers, and the guide-

lines were described, as well as the VEs and necessary

Introduction

Consent Form

Demographic Questions

Pre-Study SSQ

Hardware Tutorial

7m, wall on the left

5m, wall on the left

3m, wall on the left

7m, wall on the right

5m, wall on the right

3m, wall on the right

Game

Post-Study SSQ

VR Session Questions

7m, wall on the right

5m, wall on the right

3m, wall on the right

7m, wall on the left

5m, wall on the left

Game

3m, wall on the left

Switch Sides

Games

T-junctions

Pre and Post

VR Session

Figure 4: User study procedure and the two conditions of

the nearby physical wall being on the left or right side of

the starting position. The game of each round depends on

the assigned order.

steps. Next, participants were instructed to wear the

headset, adjust it, and adjust the correct interpupillary

distance if necessary. Then, they were guided to the

starting position without wearing the HMD so they

could see the nearby wall. The experiment started

with the 7-meter corridors, followed by the 5-meter

and then 3-meter corridors. This sequence was re-

peated for the other side of the physical room (see Fig-

ure 2). The starting side was divided equally among

the users, with half starting on the right and half on the

left. The starting positions were placed at a 0.7m dis-

tance from the corresponding wall and marked on the

ﬂoor to simplify the positioning of the users before the

start of each experiment round. The descending order

HUCAPP 2025 - 9th International Conference on Human Computer Interaction Theory and Applications

606

of lengths was chosen because starting with 7 meters

allowed the user to be redirected the most, enabling

them to walk the furthest in the critical direction be-

fore reaching the boundaries again. This created the

impression that they would be able to walk similarly

in future conditions, while in the 3-meter condition,

only two steps are possible in the CT direction.

Once the user had walked through the corridor,

made a decision on which side to go, and walked a

few steps in that direction, the experiment was ﬁn-

ished, and a stop sign appeared. Then, the user is

taken to a new scene where they can play a game.

The games served two purposes: ﬁrst, to prevent users

from remove the HMD immediately after ﬁnishing

the experiment to prevent them from realizing the use

of curvature gains; second, the games required walk-

ing to help the user forget the decision they made in

the previous condition. After ﬁnishing the game, the

user had to take off the HMD and walk again to the

starting position next to one of the walls, allowing the

second experiment to start. This led the user to see

the starting position again in the physical space.

The games used were rather simple and only en-

sured that the user needed to walk. There were in

total 5 games for the 6 conditions. In the ﬁrst game,

the user had to pick apples from trees and place them

in a box in the center of the room. The second and

third game involved shooting balloons, where the user

had to walk to the reloading station after shooting 10

bullets. In one game, the balloons were spread every-

where, while in the other, there were separating walls

that the user had to navigate to see the targets. The

fourth game involved solving mathematical equations

using cubes placed around the space. The user had to

grab the cubes and place them on a board to complete

an equation. Finally, the last game consisted of grab-

bing colored cubes and placing them in the correct

basket of the same color.

The order of the games between the conditions

was also permuted so that none of the users had ex-

actly the same sequence, ensuring that in the set of

combinations, no pair of games appeared in the same

position twice. After ﬁnishing the complete study in

VR, each user had to answer again the SSQ question-

naire, and some experiment-speciﬁc questions. These

consisted of describing the effects of the walls on their

decision and the strategy followed to make the deci-

sion at the junction. Additional questions were added

to detect the perception of curvature gain. To obscure

the focus on curvature gain, some questions referred

to non-existent aspects, with the key question about

curvature gain hidden among them. This procedure is

inspired by the work of (Suma et al., 2011). For ex-

ample, one of the obscuring questions asked the user

Figure 5: Amount of participants per number of critical

turns.

if they felt they were walking faster or slower between

the environments.

4 RESULTS AND DISCUSSION

A total of 24 individuals were recruited, 21 male and

3 female. All participants had normal or corrected-

to-normal eyesight. They consisted mostly of bach-

elor’s and master’s students. Their ages ranged from

20 to 28 years. Different levels of experience with VR

equipment are represented in the recruited pool, rang-

ing from complete novices to those with more than

100 hours of experience. We recorded 6 paths per

participant, resulting in a total of 144 paths.

The SSQ scores were computed for both the pre-

and post-questionnaire phases. The average differ-

ence between the total scores was 2.65 ± 16.28, with

14 participants having lower or unchanged scores af-

ter the testing phase. This indicates that the experi-

ments did not induce signiﬁcant cybersickness in par-

ticipants. It should be noted that there was one outlier

with a difference of 44.9. However, during this ex-

periment, some connection problems occurred, which

likely led to this result. Since the participant did not

report any discomfort in the informal feedback, we

decided to retain their data.

4.1 General Findings

Out of the 144 paths obtained, 93 were Non-Critical

Turns (NCTs) (see Table 1). That corresponds to

approximately 64.58%, i.e., nearly two-thirds of the

paths. On average 3.875 out of 6 of the participants’

paths avoided CT. So, each person chose on average 2

CTs. In Figure 5, is shown the number of participants

who chose a certain amount of CTs. No user chose

5 or more CTs. In addition, the majority chose 4 or

more paths involving NCTs.

Impact of Starting Position on Decision-Making in Virtual Reality

607

Table 1: Overall Recorded Path Decisions.

Non-critical Turns Critical Turns Total

Amount 93 51 144

Percentage 64.58% 35.42% 100%

4.2 Correlation Between CTs and Wall

Location

When looking at the number of paths leading away

and towards the close wall grouped by the wall loca-

tion (see Figure 6), we see that when the wall is on the

left, exactly 1/3 of the paths are CT. When the wall is

on the right, a similar behavior is observed, though

slightly more paths (37.5%) involving the CT. Our

null hypothesis is formulated in the following way:

: The side on which a participant starts the exper-

iment (wall to the left or to the right of the user) does

not have a signiﬁcant impact on the decision-making.

Using Fisher’s exact test, we get that p > 0.05. There-

fore, H

cannot be rejected, which proves that the ini-

tial position of the wall (to the left or to the right of

the user) did not signiﬁcantly affect the number of the

taken critical turns.

4.3 Correlation Between CTs and

Corridor Length

Grouping the results by corridor length, an interest-

ing behavior is shown. The results for the corridors

with lengths 3m and 7m show the tendency of the par-

ticipants to avoid critical turns. However, for 5m, we

have exactly 50% of CTs. Our null hypothesis is for-

mulated in the following way:

: The longer the distance up to the decision point,

the more paths will involve a CT, as the participants

tend to forget the physical boundary after walking a

longer distance.

A pairwise Fisher’s test was conducted to test our

hypothesis, with a Bonferroni correction applied due

to multiple comparisons. Firstly, Fisher’s exact test

for the conditions with corridor lengths 3m and 7m

yields that p > 0.01, and thus is not statistically sig-

niﬁcant. For conditions with lengths 5m and 7m, we

get p > 0.01, and the same goes for lengths 5m and

3m. In conclusion, H

cannot be rejected for any of

these. Hence, it cannot be proven that the lengths

had an impact on the decision-making. This conﬁrms

the ﬁnding by Ngyuen et al. (Nguyen et al., 2017),

who also stated that alternation behavior neither de-

cays nor increases over time.

Figure 6: Path decisions by wall location.

Figure 7: Path decisions by corridor length.

4.4 Correlation Between CTs and

Physical Boundaries Effect

From the 24 participants, 7 stated that the walls did

not affect their turn decision. Comparing the percent-

ages of path directions in relation to the effect of the

physical walls (see Table 2), we observe that the per-

centage of paths heading towards the wall is higher

for people who stated that they were not affected by

the walls than those who were. Still, the majority of

paths diverted from the nearby wall.

4.5 Correlation Between CTs and the

Round Number

The results yielded a clear preference to avoid the wall

when it comes to the ﬁrst, third, fourth, ﬁfth, and sixth

rounds, where the majority of participants chose the

path avoiding the CT. For the ﬁrst and last rounds,

HUCAPP 2025 - 9th International Conference on Human Computer Interaction Theory and Applications

608

Table 2: Path decisions based on awareness of boundaries.

Decision affected by

the Physical Walls

Non-critical Turns Non-critical Turns [%] Critical Turns Critical Turns [%]

No 24 57.14% 18 42.86%

Yes 69 67.65% 33 32.35%

Table 3: Alternation rates of critical and non-critical turns.

Rounds 1st - 2nd 2nd - 3rd 4th -5th 5th - 6th

P(CT |NCT ) 0.6842 0.3 0.4375 0.3571

Table 4: Alternation Rates of the Turn Direction.

Rounds 1st - 2nd 2nd - 3rd 4th -5th 5th - 6th

P(X|Y ) 0.7083 0.5 0.5 0.5833

Figure 8: Path decisions by round number.

nearly 80% chose to move away from the physical

boundary. Whereas, for the second round the behav-

ior is vastly different, with the majority choosing to

walk towards the obstacle. Still, in ﬁve out of the six

rounds, the majority of paths avoid the CT (see Fig-

ure 8). Table 3 shows the different alternation rates

(P(CT |NCT )) of choosing a CT given that the pre-

vious choice was not a CT (NCT) when comparing

two consecutive rounds of the T-junction experiment.

The probability for the second round is signiﬁcantly

higher, while for the remaining rounds, the probabil-

ity stays below 0.5.

4.6 Correlation Between the Current

and Previous Turn Choices

Table 4 shows the different alternation rates when

comparing the two consecutive rounds of the T-

junction experiment. P(X|Y ) is the probability of

choosing direction X given that the previous choice

was direction Y . We implemented the mini-games be-

tween different rounds to prevent the SAB in two con-

secutive rounds. However, as it is seen from the Table

4, P(X|Y ) between rounds 1 and 2 is rather high. Our

assumption is that this is linked to the curiosity of the

participants to explore another turn of the VE and see

what will happen. As the majority of the participants

tried to avoid the CT in the ﬁrst round due to their

awareness of the physical wall nearby, some of them

reported their interest in taking it in the next round.

5 CONCLUSION

In this paper, we presented an approach to investi-

gate the effect of the starting position on the walking

path and decision-making in VR. We hypothesized

that starting next to the physical wall triggers partici-

pants to take the non-critical turn and steer away from

the real wall to avoid collision. To implement such an

environment, we used a redirected walking technique

that steered participants away from the wall to prevent

them from colliding with it if a critical turn was taken.

To mitigate participants’ spatial awareness, we inte-

grated mini-games in between different rounds. Our

results showed that in the initial round, participants

tended to avoid critical turns due to an awareness of

their starting position near the wall. However, in the

second round, more than half of the participants chose

to take a critical turn, mentioning during interviews

that they were aware of the environment but wanted

to explore the consequences of taking a critical turn.

Our research highlights the need to consider the

spatial awareness of the users when natural walking

is required. Further research can be broadened to not

only consider the starting position next to the wall but

also investigating the inﬂuence of other obstacles that

are present in the physical environment.

REFERENCES

Abu-Saﬁeh, S. F. (2011). Virtual reality simulation of ar-

chitectural clues’ effects on human behavior and deci-

sion making in ﬁre emergency evacuation. In Pea-

Impact of Starting Position on Decision-Making in Virtual Reality

609

cock, R. D., Kuligowski, E. D., and Averill, J. D.,

editors, Pedestrian and Evacuation Dynamics, pages

337–347, Boston, MA, USA. Springer US.

Interrante, V., Lindquist, J., and Anderson, L. (2007). Eluci-

dating factors that can facilitate veridical spatial per-

ception in immersive virtual environments. In 2007

IEEE Virtual Reality Conference, pages 11–18.

Kennedy, R. S., Lane, N. E., Berbaum, K. S., and Lilien-

thal, M. G. (1993). Simulator sickness questionnaire:

An enhanced method for quantifying simulator sick-

ness. The International Journal of Aviation Psychol-

ogy, 3(3):203–220.

Lutfallah, M., Buciunas, K., and Kunz, A. (2024). Revisited

threshold detection in redirection techniques. In Pro-

ceedings of the 2024 8th International Conference on

Virtual and Augmented Reality Simulations, ICVARS

’24, page 28–34, New York, NY, USA. ACM.

Nguyen, A., Kunz, A., Rothacher, Y., Brugger, P., and

Lenggenhager, B. (2017). Spontaneous alternation be-

havior in humans. In Proceedings of the 23rd ACM

Symposium on Virtual Reality Software and Technol-

ogy, VRST ’17, New York, NY, USA. ACM.

Razzaque, S. (2005). Redirected walking. The University of

North Carolina at Chapel Hill, Chapel Hill, NV, USA.

Rothacher, Y., Nguyen, A. N., Lenggenhager, B., Kunz,

A., and Brugger, P. (2020). Walking through virtual

mazes: Spontaneous alternation behaviour in human

adults. Cortex, 121(1):1–16.

Steinicke, F., Bruder, G., Jerald, J., Frenz, H., and Lappe,

M. (2010). Estimation of detection thresholds for redi-

rected walking techniques. IEEE Transactions on Vi-

sualization and Computer Graphics, 16(1):17–27.

Suma, E. A., Clark, S., Krum, D., Finkelstein, S., Bolas, M.,

and Warte, Z. (2011). Leveraging change blindness

for redirection in virtual environments. In 2011 IEEE

Virtual Reality Conference, pages 159–166.

Usoh, M., Arthur, K., Whitton, M. C., Bastos, R., Steed,

A., Slater, M., and Brooks, F. P. (1999). Walk-

ing > walking-in-place > ﬂying, in virtual environ-

ments. In Proceedings of the 26th annual conference

on Computer graphics and interactive techniques -

SIGGRAPH ’99. ACM Press.

Vilar, E., Rebelo, F., Noriega, P., Teles, J., and Mayhorn,

C. (2013). The inﬂuence of environmental features

on route selection in an emergency situation. Applied

Ergonomics, 44:618–627.

Xie, B., Liu, H., Alghofaili, R., Zhang, Y., Jiang, Y., Lobo,

F. D., Li, C., Li, W., Huang, H., Akdere, M., et al.

(2021). A review on virtual reality skill training appli-

cations. Frontiers in Virtual Reality, 2:645153.

HUCAPP 2025 - 9th International Conference on Human Computer Interaction Theory and Applications

610