Towards large scale smell display
for enhancing immersion during
walking in virtual environments
DIPLOMARBEIT
zur Erlangung des akademischen Grades
Diplom-Ingenieur
im Rahmen des Studiums
Media and Human-Centered Computing
eingereicht von
Christoph Götz, BSc.
Matrikelnummer 51825978
an der Fakultät für Informatik
der Technischen Universität Wien
Betreuung: Univ.Prof. Mag.rer.nat. Dr.techn. Hannes Kaufmann
Mitwirkung: Univ.Ass. PhD Hugo Brument
Univ.Ass. Dr. Francesco De Pace
Wien, 15. August 2025
Christoph Götz Hannes Kaufmann
Technische Universität Wien
A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 www.tuwien.at

Towards large scale smell display
for enhancing immersion during
walking in virtual environments
DIPLOMA THESIS
submitted in partial fulfillment of the requirements for the degree of
Diplom-Ingenieur
in
Media and Human-Centered Computing
by
Christoph Götz, BSc.
Registration Number 51825978
to the Faculty of Informatics
at the TU Wien
Advisor: Univ.Prof. Mag.rer.nat. Dr.techn. Hannes Kaufmann
Assistance: Univ.Ass. PhD Hugo Brument
Univ.Ass. Dr. Francesco De Pace
Vienna, August 15, 2025
Christoph Götz Hannes Kaufmann
Technische Universität Wien
A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 www.tuwien.at

Erklärung zur Verfassung der
Arbeit
Christoph Götz, BSc.
Hiermit erkläre ich, dass ich diese Arbeit selbständig verfasst habe, dass ich die verwen-
deten Quellen und Hilfsmittel vollständig angegeben habe und dass ich die Stellen der
Arbeit – einschließlich Tabellen, Karten und Abbildungen –, die anderen Werken oder
dem Internet im Wortlaut oder dem Sinn nach entnommen sind, auf jeden Fall unter
Angabe der Quelle als Entlehnung kenntlich gemacht habe.
Ich erkläre weiters, dass ich mich generativer KI-Tools lediglich als Hilfsmittel bedient
habe und in der vorliegenden Arbeit mein gestalterischer Einfluss überwiegt. Im Anhang
„Übersicht verwendeter Hilfsmittel“ habe ich alle generativen KI-Tools gelistet, die
verwendet wurden, und angegeben, wo und wie sie verwendet wurden. Für Textpassagen,
die ohne substantielle Änderungen übernommen wurden, haben ich jeweils die von
mir formulierten Eingaben (Prompts) und die verwendete IT- Anwendung mit ihrem
Produktnamen und Versionsnummer/Datum angegeben.
Wien, 15. August 2025
Christoph Götz
v

Danksagung
Es gibt viele Personen, die mich im Laufe des Studiums und im Rahmen dieser Masterar-
beit unterstützt haben, und auch wenn ich sie hier nicht alle namentlich nennen kann,
möchte ich mich auf diesem Weg für die große Hilfe bedanken. Allen voran gilt mein Dank
besonders meinen Eltern, Susanne und Alexander, und meiner Lebensgefährtin, Diana,
die mich fortlaufend motiviert haben und mir in jeder Lebenslage zur Seite standen. Ich
bin besonders dankbar dafür, dass ihr mir stets das Vertrauen entgegengebracht habt,
meinen eigenen Weg und mein eigenes Tempo zu gehen, auch wenn sich das Studium etwas
länger gezogen hat. Diese Geduld und Unterstützung waren für mich von unschätzbarem
Wert.
Zusätzlich möchte ich mich bei meiner Kommilitonin Katharina für die vielen Ratschläge
im Rahmen des Studiums und der Masterarbeit bedanken. Mit dir habe ich nicht nur eine
sympathische Kollegin für Gruppenprojekte gefunden, sondern auch eine gute Freundin
für die Zeit nach dem Studium.
Mein Dank gilt auch meinen weiteren Freundinnen und Freunden, denen ich inzwischen
schon viel zu oft von meiner Masterarbeit und meinem Studium erzählt haben muss –
danke für eure ehrlichen Meinungen, eure Ratschläge und eure Motivationsschübe. Auch
wenn eure ständige Neugier auf meinen Fortschritt mir manchmal die Nerven geraubt hat,
hat sie mir doch gezeigt, dass ihr euch, im besten Sinne, wirklich für meine Masterarbeit
und für mich interessiert habt. Dafür bin ich euch sehr dankbar.
Schlussendlich möchte ich mich bei meinen Betreuern Hannes Kaufmann, Hugo Brument
und Francesco De Pace bedanken. Ihr habt mich bei meinen, wie ihr wisst, unzähligen
Fragen und Herausforderungen unterstützt und mir stets Geduld entgegengebracht, wenn
ich mal gestockt habe. Besonders Hugo möchte ich an dieser Stelle ein großes Dankeschön
aussprechen, sowohl für die engagierte Unterstützung bei allen Masterarbeits-relevanten
Themen, als auch für die angenehme, motivierende Atmosphäre während der gesamten
Projektbetreuung.
Für alles, was ihr für mich getan habt, und immer noch tut, danke ich euch von Herzen.
vii

Acknowledgements
There are many people who have supported me throughout my studies and during the
course of this master’s thesis, and even though there are too many to name individually,
I would like to take this opportunity to sincerely thank you all for your tremendous
support.
First and foremost, I would like to express my heartfelt gratitude to my parents, Susanne
and Alexander, and to my partner, Diana, who have continuously motivated me and
stood by my side through all phases of life. I am especially thankful that you always
trusted me to follow my own path and pace, even when it took me a little longer to
complete my studies. Your patience and support have been invaluable to me.
I would also like to thank my fellow student Katharina for her many helpful insights
throughout my studies and this thesis. In you, I not only found a reliable project partner
but also a true friend beyond the academic journey.
My thanks also go to my friends, who by now must have heard far too much about
my thesis and studies – thank you for your honest feedback, your advice, and your
motivational boosts. Even if your persistent curiosity about my progress sometimes got
on my nerves, it always reminded me that you genuinely cared – and for that, I am truly
grateful.
Finally, I want to express my deep gratitude to my supervisors, Hannes Kaufmann, Hugo
Brument, and Francesco De Pace. You supported me through countless questions and
challenges, always responding with patience when I felt stuck. A very special thank-you
goes to Hugo for your dedicated support on every thesis-related topic, as well as for the
kind conversations and the motivating atmosphere throughout the entire project.
I thank all of you from the bottom of my heart for everything you have done, and continue
to do.
ix

Kurzfassung
Die Einbindung olfaktorischer Reize in die virtuelle Realität (VR) entwickelt sich zuneh-
mend zu einem vielversprechenden Forschungsfeld, das neue Möglichkeiten zur Steigerung
der Immersion und des emotionalen Erlebens bietet. Bestehende Lösungen für olfak-
torische Displays sind derzeit oft durch stationäre Duftquellen oder tragbare Geräte
limitiert, die entweder die Bewegungsfreiheit der Nutzer:innen einschränken oder als
störend empfunden werden. Diese Arbeit präsentiert einen neuartigen Ansatz für groß-
flächige Geruchsinteraktionen in VR durch ein robotergestütztes, mobiles Duftsystem,
das eine ortsbezogene Geruchsausgabe während natürlicher Gehbewegungen ermöglicht.
Das System basiert auf einem mobilen Boston Dynamics Spot-Roboter in Kombination
mit einem Olorama-Duftgenerator, navigiert autonom zu vordefinierten Duftzonen und
synchronisiert die Geruchsemission mit der räumlichen Nähe der Nutzer:innen. Dadurch
werden zeitlich und räumlich abgestimmte olfaktorische Reize ermöglicht, ohne dass
zusätzliche Hardware getragen werden muss.
Das System wurde in einer Nutzerstudie in einem raumgroßen virtuellen Waldszenario mit
drei verschiedenen Geruchserlebnissen evaluiert, dabei wurden die Erkennungszeit, die Er-
kennungsgenauigkeit, die wahrgenommene Intensität sowie der wahrgenommene Komfort
während der Interaktionen erfasst. Die Ergebnisse zeigen, dass die meisten Düfte korrekt
erkannt wurden, wobei die Reaktionszeiten durch Unterschiede in der Duftintensität
beeinflusst wurden. Insgesamt belegen die Ergebnisse, dass eine großflächige, roboter-
gestützte Geruchsausgabe sowohl technisch umsetzbar als auch wahrnehmungsmäßig
effektiv und gut verträglich ist und von den Nutzer:innen gut toleriert wird.
Durch den Verzicht auf tragbare Komponenten, permanente Installationen oder raum-
spezifische Infrastruktur bietet das System eine skalierbare und flexible Lösung für
immersive Geruchsanwendungen. Es unterstützt gehbasierte VR-Szenarien, ohne die
Bewegungsfreiheit von Nutzer:innen einzuschränken, und ermöglicht eine natürlich wir-
kende, kontextbezogene Geruchsausgabe für vielfältige Anwendungsfelder wie Storytelling,
Therapie oder Ausstellungen. Insgesamt unterstreicht diese Arbeit das Potenzial mobiler
Duftsysteme als vielseitige und praktikable Erweiterung immersiver VR-Erlebnisse.
xi

Abstract
Incorporating olfactory stimuli into Virtual Reality (VR) has become a growing area of
interest, offering new ways to enhance immersion and emotional engagement. However,
current olfactory display solutions are often constrained by stationary emitters or wearable
devices, which either restrict user mobility or cause discomfort. This thesis presents a
novel approach to large-scale smell interaction in VR through a robot-mounted olfactory
display that enables encounter-based scent delivery during natural walking. Built on
a mobile Boston Dynamics Spot robot and an Olorama scent generator, the system
autonomously navigates to predefined smell zones and synchronizes scent emission with
user proximity, allowing for timed and spatially aligned olfactory cues without requiring
users to wear additional hardware.
The system was evaluated with a user study in a room-scale VR forest scenario with
three distinct olfactory events, where the detection time, recognition accuracy, perceived
intensity, and comfort of users during the interactions were assessed. Results show that
users were generally able to detect and identify the smells correctly, with reaction times
being influenced by differences in scent intensity. The results demonstrate that large-scale,
robot-assisted scent delivery is not only technically feasible but also perceptually effective
and well tolerated by users.
By avoiding wearable gear, permanent installations, or room-specific infrastructure, the
system offers a scalable and portable alternative for immersive olfactory interaction. It
supports walk-based scenarios without confining users to a small area, enabling natural,
spatially grounded scent delivery in various applications, including storytelling, therapy,
or exhibitions. Overall, this work highlights the potential of mobile scent systems as a
flexible and practical modality for enhancing immersion in VR.
xiii

Contents
Kurzfassung xi
Abstract xiii
Contents xv
1 Introduction 1
1.1 Background and Motivation . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Aim of the work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Related Work 7
2.1 Human Smell Sensory System . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Importance and Impact of Smell . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Effects on Emotion . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Effects on Learning and Memory . . . . . . . . . . . . . . . . . 10
2.3 Olfactory Technologies and Evaluation in VR . . . . . . . . . . . . . . 10
2.3.1 Wearable Olfactory Systems . . . . . . . . . . . . . . . . . . . . 11
2.3.2 Room-Based and Static Emitters . . . . . . . . . . . . . . . . . 17
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 System Design and Implementation 21
3.1 System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Hardware and Software Setup . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.1 Boston Dynamics Spot . . . . . . . . . . . . . . . . . . . . . . . 24
General Description . . . . . . . . . . . . . . . . . . . . . . . . 24
Motion Control and Spatial Alignment . . . . . . . . . . . . . . 26
3.2.2 Olorama Scent Generator . . . . . . . . . . . . . . . . . . . . . 28
3.2.3 Software Architecture . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Evaluation Task and Study Logic . . . . . . . . . . . . . . . . . . . . . 31
4 Evaluation and Results 37
4.1 Study Design and Hypotheses . . . . . . . . . . . . . . . . . . . . . . . 37
xv
4.2 Participants and Apparatus . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4 Data Collection and Metrics . . . . . . . . . . . . . . . . . . . . . . . . 43
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.5.1 Time and Trajectories . . . . . . . . . . . . . . . . . . . . . . . 45
4.5.2 Smell Perception . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.5.3 Questionnaires . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5.4 Post-Experiment Feedback and Additional Observations . . . . 49
5 Discussion 51
5.1 Reflections on Study and Hypotheses . . . . . . . . . . . . . . . . . . . 51
5.2 Design Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6 Conclusion and Future Work 59
Overview of Generative AI Tools Used 61
List of Figures 63
List of Tables 65
Bibliography 67
Appendix 73
CHAPTER 1
Introduction
The sense of smell plays a fundamental role in how humans experience and navigate
the world, as the ability to sense odors in the environment affects our daily decisions.
Smell accompanies humans when deciding if a particular item of food is still edible; it
is an indicator for environmental hazards, a smell can induce strong emotional feelings,
alter behavior, and the way humans communicate with each other, depending on their
smell, and is even able to act as a stimulus to enhance memory. The importance of
smell in daily life is gaining recognition from various commercial sectors, such as food,
cosmetics, and cleaning products. These industries work to create scents that effectively
convey the qualities of their products. This effort clearly affects how consumers view
their attractiveness and therefor make the products seemingly more desirable [ABE+22].
However, in the fields of Virtual Reality (VR), robotics, and human-computer interac-
tion, olfaction remains one of the most underexplored sensory modalities, especially in
comparison to using sounds or the visual sense. This is particularly true for large-scale
VR environments, where delivering smell in sync with user actions remains a challenge.
Reasons for this are that VR environments are often not designed realistically enough to
be able to convey the realistic feeling of being able to smell in VR, that methodological
limitations exist in study design or the development of smells, and that the mechanisms
of delivering the odors to users in VR are too unsophisticated [ABE+22].
This thesis investigates how mobile robots can be used to create encounter-based, location-
aware olfactory interactions in VR environments, focusing on the integration of the Boston
Dynamics Spot robot with the Olorama scent generator. By combining spatial movement
with controlled scent delivery, this work presents a proof-of-concept system that enables
users to experience scents in immersive, multi-sensory ways. The following chapters
outline the background, related work, system architecture, and practical evaluation of
this approach.
1
1. Introduction
1.1 Background and Motivation
Virtual Reality makes the simulation of real-life situations in a computer-generated
environment possible. Through stimulation of different senses such as the sense of smell,
by using vibrations, by increasing the sense of movement by blowing air into users faces,
by adding touch or taste or auditive feedback, or even by using a bright light source
to simulate the feeling of sun on the skin to stimulate touch, as well as by using many
other possibilities that stimulate the different human senses, in combination with using
VR technology, it is possible to immerse into a completely different scene [SBB16]. In
general, Virtual Reality, Augmented Reality (AR), and Mixed Reality (MR) provide the
means to evoke a sense of presence for users, where they can feel as if they are physically
present in a virtual environment [Hol23].
While the sense of immersion is commonly achieved through adding 3D visuals, spatial
audio, or haptic feedback, or in more rare cases by expanding the application with
olfactory feedback, VR experiences have so far predominantly focused on stimulating
the visual and auditory sense of users. In recent years, however, multiple studies have
explored the effects of stimulating additional senses beyond sight and hearing to enhance
the sense of presence in VR further. Their findings consistently demonstrate that the
more senses are being engaged, the stronger the sense of presence users report [DWH+99].
For example, Hoffman et al. [HHSR98] investigated the role of tactile sensations and
confirmed that adding haptic feedback improves the sense of presence in virtual environ-
ments. Similarly, other studies have proposed that olfactory stimuli in VR can not only
strengthen the feeling of immersion but also evoke specific emotional responses, which
directly influence user behavior. Tortell et al. [TLD+07] demonstrated that the inclusion
of smell in a VR experience significantly enhanced the ability of participants to memorize
visual items that are encountered during their journey through the virtual environment,
compared to participants who experienced the same environment without any olfactory
input [TLD+07].
Nevertheless, despite these promising findings, the implementation of olfactory feedback
into large-scale VR environments remains challenging due to multiple fundamental hurdles
that remain unsolved in current literature. One of the most significant problems is that
stationary emitters are limited to a dispersion range of one to two meters, requiring
near-nose placement of the sensors to maintain a detectable concentration of the smell.
In addition to that, temporal precision issues manifest through a mixture of latency
delays between the different components that are being used, resulting in a delay between
scent activation and user perception, which proves to be a critical barrier to effective
multi-sensory experience synchronization [IBM+14]. Moreover, controlling the timing
and intensity of scent releases poses significant challenges, even more so when using them
in multi-user scenarios with multiple smell releases, where cross-contamination between
different scents can occur [SH21]. These constraints restrict the scalability of smell in
larger VR environments, making it challenging to deliver olfactory stimuli effectively in
scenarios where users are allowed to move around freely [SBB16].
2
1.2. Problem Statement
The lack of scalable olfactory solutions for VR is particularly noteworthy, as no existing
work has successfully addressed the challenge of delivering scent cues across physical
spaces in a VR context. This limitation confines current olfactory VR solutions to being
small-scale applications, despite their potential to transform into various fields, and
prevents deployment in applications that require synchronized multi-user feedback or
large-scale environmental priming, such as collaborative training simulations. Similarly,
smell could be used for large-scale spaces, such as in museums, where immersive exhibit
spaces could employ pre-delivered ambient odors to generate environmental smell contexts
before visual scene transitions appear, making users more immersed in the exhibition.
However, recent advancements in robotic mobility platforms and programmable odor
generators offer promising avenues for overcoming these challenges. Emerging research
in multi-sensor robotic navigation demonstrates the feasibility of combining olfactory
tracking with autonomous mobility for scent delivery. Studies, for example, already show
that robots can localize odor sources with an accuracy of two meters, using bio-inspired
algorithms [SLW11]. Although current implementations primarily focus on odor source
localization rather than scent delivery, these systems lay the groundwork for navigation
frameworks that adapt to their environments, serving as a solid foundation for further
development. While quadrupedal robots, such as the Boston Dynamics Spot robot, have
not yet been explicitly tested for scent delivery, their autonomous navigation capabilities,
as demonstrated in the example for odor localization, suggest a viability for maintaining
optimal emission distances when integrated with programmable generators such as the
Olorama [LYZ+23].
Successful implementations of robotic mobility platforms, such as the Spot, in combi-
nation with programmable scent generators, such as the Olorama, could improve smell
interaction in VR. By combining both technological aspects, it may be possible to achieve
precise spatiotemporal control over scent deliveries, enabling applications that were
previously unfeasible. Successful implementation could have far-reaching implications,
from revolutionizing therapeutic VR through graded exposure environments for anxiety
disorders to creating new paradigms in architectural design by enabling scent-augmented
virtual walkthroughs of unbuilt spaces, pre-delivering smell in immersive museum exhibits,
as well as creating more immersive training scenarios, which, for example, firefighters
could use.
1.2 Problem Statement
The integration of olfactory stimuli in VR environments has gained attention for its
potential to enhance immersion, emotional engagement, and realism.
However, most existing implementations rely on static scent emitters, such as, for
example, the Olorama Smell generator, which is generally a static device, or wearable
olfactory displays, such as the one that has been built throughout the study by Yamada
et al. [YYT+06], which are typically cumbersome. Stationary systems are typically
fixed to a single location and cannot adapt to a users movement. In contrast, wearable
3
1. Introduction
solutions, although mobile, often suffer from limited scent intensity, discomfort, and
diffusion range, making them unsuitable for multi-user or large-scale environments. These
limitations restrict the use of olfactory interaction in spatially dynamic VR scenarios that
involve walking or physical navigation, since in such cases, scent delivery should ideally
adapt in real time to a users position and their actions within the physical environment
immediately. The lack of flexible, mobile, and scalable solutions for delivering smell in
large VR spaces remains a significant challenge and presents a gap that this thesis aims
to address.
Current research and applications often rely on fixed scent emitters positioned in the
environment or integrated into head-mounted displays. While these solutions can pro-
vide localized olfactory feedback, they are often cumbersome for users, as they require
wearing additional hardware or being tethered to a specific area. This makes them
impractical for larger environments, multi-room setups, or experiences that involve free
movement [LYZ+23]. Moreover, delivering scents at the correct time and place, in sync
with a user’s actions and the corresponding virtual stimuli, remains a significant technical
challenge. Many existing systems struggle with precise timing, resulting in delayed or
lingering odors from emitted smells that are still in the air and that do not align with
visual or spatial cues. Such mismatches can break the immersion or lower the amount of
immersion that users feel and reduce the effectiveness of the experience [PD20].
To address these challenges, this thesis is guided by two overarching research questions:
• RQ1 (System Design): How can a mobile robotic platform be designed to deliver
olfactory stimuli dynamically in large-scale virtual environments?
• RQ2 (System Evaluation): How can the usability, effectiveness, and user
experience of a mobile olfactory display system be systematically evaluated in VR?
1.3 Aim of the work
The goal of this thesis is to design, develop, and evaluate a mobile, location-based
olfactory interaction system that combines the Boston Dynamics Spot robot with the
Olorama scent generator in VR. The central idea is to use Spot as a dynamic carrier
of scent, enabling real-time olfactory experiences that respond to user movement and
interaction within a physical environment that is simultaneously experienced through a
VR headset.
To achieve this, the system architecture integrates multiple technologies, including Unity
for scene and interaction control, a Python backend to manage the communication
between the VR application and the physical devices, especially the Spot robot and the
Olorama scent generator, and VR hardware for spatial tracking as well as user immersion.
The aim is not only to demonstrate the technical feasibility of such a setup but also
to investigate its usability and flexibility for future use cases with a proof-of-concept
application.
4
1.4. Structure of the thesis
Through the implementation of a forest-based virtual scenario, the work further explores
how different scents and scent delivery patterns influence the user experience, while
also assessing the technical performance of the system. In particular, it is important to
evaluate whether the Spot robot can deliver scents with sufficient precision and reliability.
In addition, the study examines practical considerations that arise when deploying such
a system repeatedly in real-world conditions.
1.4 Structure of the thesis
This thesis is divided into several chapters that collectively describe the background,
design, implementation, and evaluation of the proposed system.
Chapter 2, Related Work, provides an overview of the scientific and technological
foundations relevant to this thesis. It summarizes olfactory output technologies across
wearable and room-based systems, and highlights current evaluation methods used in
VR studies. The chapter discusses trade-offs in mobility, spatial accuracy, and timing. It
concludes by identifying a research gap in mobile scent delivery and positions this work
as a novel contribution addressing that gap.
In Chapter 3, System Design and Methodology, introduces the overall system setup
and describes the hardware components, including the Boston Dynamics Spot robot, the
Olorama scent generator, the HTC Vive setup, as well as the integration of all parts.
Following, the software architecture is discussed in detail, combined with a breakdown of
the scenario logic and interaction design. The chapter ends with a discussion of the key
limitations of the current prototype setup.
Chapter 4, Evaluation and Results, is dedicated to the evaluation of the system. It
begins by outlining the design and objectives of the user study, followed by the user
study, which includes a description of the experimental procedure and the collected
data. The analysis combines quantitative metrics with qualitative feedback to assess
user experience, realism, and perceived immersion. Afterwards, results are presented
along defined categories, reflecting the hypotheses and evaluation goals of the study. The
chapter concludes with an integrated summary of outcomes based on statistical trends
and the feedback of participants.
Chapter 5, Discussion, critically reflects on the findings and outcomes presented in
the previous evaluation chapter. It contextualizes the results in relation to the initial
research questions and existing literature, highlighting key insights and interpretations.
Limitations of the system and study design are discussed, alongside suggestions for
improvement. The chapter concludes by outlining implications for future research and
potential applications of mobile olfactory VR.
Chapter 6, Conclusion and Future Work, summarizes the primary outcomes of the
thesis and reflects on the research questions. It concludes with a critical reflection on
the prototype and suggests potential avenues for future development, both in terms of
technical enhancements and research directions.
5

CHAPTER 2
Related Work
The integration of olfactory stimuli into immersive systems is a growing research field
at the intersection of Virtual Reality, robotics, and multi-sensory interaction. Although
visual and auditory modalities are well-established in VR, the sense of smell remains
underutilized despite its potential to deepen presence, evoke memory, and influence
behavior [TLD+07]. This chapter reviews the existing literature and the technological
foundations relevant to this work. It begins by outlining the fundamentals of olfactory
perception and its role in human experience, followed by an overview of scent delivery
technologies, including electronic noses and programmable diffusers. Subsequently, it
discusses prior attempts to incorporate smell in VR and AR environments, as well as
the challenges associated with spatially accurate and timely scent deployment. Finally,
related approaches involving robotics and mobile scent systems are reviewed, highlighting
the need for location-based, real-time olfactory interaction systems, which is precisely
the gap this thesis aims to address.
2.1 Human Smell Sensory System
Senses can perceive information and allow humans to interact with the environment more
intimately. By absorbing photons of light, the rods and cones in the human eyes are
activated. The ears use compressions and rarefactions of sound waves to perceive and
understand sound. At the same time, acidic is responsible for the taste and sense of smell
in the tongue, probably making it the most intimate sense, turning odorant molecules
into electrical messages that the brain interprets [Bro10].
The olfactory system is responsible for the sense of smell and is made up of several
specialized structures and organs, visible in Figure 2.1. Odorant molecules enter the
body through the nostrils and travel into the nasal cavity, where they encounter the
olfactory epithelium, which is a specialized sheet of tissues located deep within the
upper part of the nasal cavity. The olfactory epithelium contains three main types
7
2. Related Work
of cells, which are olfactory receptor neurons, which detect odorants, supporting cells,
responsible for providing structural and metabolic support, and basal cells, needed to
serve as stem cells to regenerate olfactory neurons. Olfactory receptor neurons are bipolar
cells with cilia that extend into the mucus that covers the epithelium. These cilia contain
olfactory receptors that bind to dissolved odor molecules in mucus, initiating the sensory
transduction process [KGS+24][HKC+23]. The axons of the olfactory receptor neurons
bundle together to form the olfactory nerve, which passes through to reach the olfactory
bulb, a structure located at the base of the brain just above the nasal cavity. Within
the olfactory bulb, the axons of receptor neurons converge in spherical structures and
synapse with the primary projection neurons of the olfactory bulb. These projection
neurons transmit processed olfactory information through the olfactory tract to various
regions of the brain, including the olfactory cortex, amygdala, and other structures of
the limbic system involved in perception, memory, and emotion of odors. This complex
organization allows the olfactory system to rapidly detect, discriminate, and process a
vast array of smell molecules [MSR06].
Figure 2.1: The human olfactory system. An olfactory cue enters the human smell
system, travels through the nasal complex and the mucus to the cilia, where the olfactory
receptors perceive the odor, start a chemical transformation, and turn the odor into an
understandable signal for the human brain [Bro10].
When smell is derived from their unique chemical properties, the molecules travel through
nasal passages at different speeds, leading to different interactions with the receptors
within the nose of a human and can make the olfactory system perceive different olfactory
signals [CF23]. The smell molecules are small enough to reach deep into the nasal cavity
and diffuse through a 10 to 40 micrometer thick mucus layer and then interact with the
8
2.2. Importance and Impact of Smell
different elements that the olfactory sensory neurons deploy [Bro10]. As can be seen
in Figure 2.1, the mucus layer is secreted by acinar cells and acts as the final barrier
to odorant access. Molecules must pass from the air to the mucus, diffuse through it,
and then reach the receptor sites with the olfactory receptor neurons, which expel the
necessary chemical elements to perceive the odor. The mucus circulation functions not
just as a gatekeeper of molecules, but also has the important role of assisting in the
removal of odorants from the receptor sheets after a transduction has occurred, so that
new smells can be derived and do not overlap with smells which have already been
perceived [Lin98]. There are approximately 347 related and various olfactory receptors
which can react to odorants [Bro10]. Furthermore, there are around 1000 types of receptor
cells, with which people can distinguish around 20,000 different smells, thus creating a
ratio of smells to receptors of about 20:1 [PBH17]. Some of them are specifically set to
perceive specific molecules. When receptors receive an odorant, they release a protein
unit, initiating a chemical process that converts the olfactory sensory neurons message
from an odorant molecule into an electric signal interpretable by the brain [Bro10].
In general, humans constantly encounter odors in daily life, often perceiving the quality of
a smell without being able to identify its source specifically. Smells are a combination of
molecules that vary in thousands of physicochemical dimensions, while odor concentrations
in the everyday environment of humans can differ by ten thousand times due to distractions
and external factors such as air quality, wind, or distance [Bro10]. Airflow rates and
turbulence can also be affected by actions by the human being itself, such as by active
sniffing, which boosts the air velocity and can therefore have a critical influence on how
the receptors in the nose perceive a smell [Lin98].
2.2 Importance and Impact of Smell
The human sense of smell is important for vital functions such as warning and protection
from environmental hazards, eating behavior, nutrition, and social communication. By
using this sense, humans can extract nutritional information from olfactory food cues,
which can trigger specific emotions, such as in this case, a particular appetite and direct
food choices, while not directly impacting the actual eating behavior. Furthermore, smell
can transfer and regulate emotional conditions, and therefore has a massive impact on
social relationships [BP21].
2.2.1 Effects on Emotion
When people smell the fragrance of flowers or well-smelling food, they feel relaxed and
happy, whereas they feel disgusted when they smell contaminated food. Therefore,
exposure to odor affects the mood of people, and correspondingly, unique smells could
potentially influence human behavior. Since brain structures such as the amygdala,
hippocampus, orbitofrontal cortex, and insula are involved in olfactory processing while
also being the primary structures for emotional processing, the connection between these
characteristics becomes logical [LJW20].
9
2. Related Work
Research has shown that odors that are liked or disliked have a congruent impact on
the mood and cognition of people. Ambient odors can elicit mood effects on cognition
and behavior in a simple manner, which could be used for daily life. The presence of a
pleasant ambient odor has been found to help in solving creative problems in relation
to an unpleasant odor condition. In contrast, the presence of a reeky odor reduces the
likelihood that participants can make objective judgments. Furthermore, a foul-smelling
odor lowers the tolerance for frustration, which also has an impact on the cognitive effort
of a human being. Herz et al. [HSB04] argue that associative learning is responsible for
such emotional responses since the different smells can act as cues to past emotional
experiences and therefore exert the same type of cognitive and behavioral effects that
the emotions would cause. Therefore, not only is the sense of smell closely linked to
our emotional experiences, but odors, in addition to being pleasantly or unpleasantly
smelling, can become linked to emotions in humans’ memory [HSB04].
2.2.2 Effects on Learning and Memory
Improving learning abilities is an important aspect for every human being in nearly every
situation.
Studies have found that presenting cues, such as odors or sounds, during both learning
and sleep can significantly enhance memory performance. In the study by Smith et
al. [SSdM92], participants learned a list of 24 words while being exposed to one of two
odors, which were either jasmine incense or Lauren perfume. Subsequently, the subjects
had to relearn the list with either the same or the alternative odor present. The results
showed that superior memory was found when the same odor was used for the initial
learning phase and the relearning session, demonstrating that the odor makes a difference
when trying to remember something [SSdM92].
Smell can also be used for associative learning, which is the process where one item or
event is linked to another through an experience. This art of learning is involved in
human cognition and behavior, and is specifically interesting regarding how smell can
change the setting of such learning experiences. If one were to smell the odor of a rose in
an unpleasant setting, such as at a funeral, for the very first time, this smell would likely
keep reminding the person about the funeral, even if the smell occurred at a different
time and place in the future [HSB04].
2.3 Olfactory Technologies and Evaluation in VR
Historically, the foundations for immersive multi-sensory environments were laid by
Morton Heilig, who in the 1950s built the first virtual environment, which provided users
with a multi-sensory experience in the form of a simulated motorcycle ride through New
York, combining 3D visuals, stereo sound, wind, vibrations, and even aroma cues [AAS20].
He later developed the Telesphere Mask, the first head-mounted device with 3D vision
and stereo sound. Developments continued with Comeau and Bryan’s “Headsight” in
10
2.3. Olfactory Technologies and Evaluation in VR
1961, incorporating head-tracking and dual video screens. By 1966, Furness III introduced
the first VR flight simulator, and internal motion sensors were added to HMDs by 1968.
Over time, VR technology has expanded across various domains, from immersive gaming
experiences to teaching new skills, to therapy, which assists in treating psychological
disorders, and even allows terminally ill patients to embark on virtual journeys through
the world [AAS20].
In past studies, it was already noted that surgical training simulations, which are one of
the most advanced and important developing VR applications, feel incomplete without
olfactory input. This has led to increasing research into scent delivery systems that can
release odors in real-time, reproducibly, and in sync with events and user actions. Ideally,
these systems must support multiple odorants, maintain spatial and temporal accuracy,
avoid contamination across trials, and operate without tactile distractions [DWH+99].
In general, humans tend to underestimate the importance of smell as a source of informa-
tion and an interaction tool with our environment. However, olfactory cues can evoke
memories, convey meaning, such as by warning us of danger, and can promote instinctive
behavior, such as keeping a distance from a bad-smelling object. These qualities make
smell a powerful tool for designing spatial interactions, especially in VR, where scent can
significantly enhance the sense of presence in multi-sensory experiences. While there is
growing interest in integrating olfactory stimuli into VR to enrich user experience and in
developing wearable scent interfaces, the exploration of smell in relation to its spatial
characteristics and interaction logic remains limited [DMP+21].
2.3.1 Wearable Olfactory Systems
Wearable systems bring scent emitters close to the user’s nose, often integrated into or
mounted on HMDs. This configuration offers spatial precision and immediacy but often
introduces design and comfort challenges.
Sezille et al. [SMB+13] and Nakamoto and Minh [NM07] developed early wearable
olfactory displays capable of releasing 15 and 30 distinct smells, respectively. While those
devices looked promising in terms of odor diversity, their systems relied on significant,
heavy components, long tubes, and suffered from delivery noise, making them impractical
for consumer VR.
To overcome these constraints, Yanagida et al. [YKN+04] introduced a novel system that
used a kind of “air cannon” to project “clumps” of scented air into the area beneath the
user’s nose. Unlike tube-based emitters, this setup avoided direct contact or attachments
on the face. The goal was to provide spatio-temporal control over olfactory cues, avoiding
residual scent clouds in the surrounding environment.
On the downside, this system only worked with a very limited number of smells, as each
new smell had to overlap with and overcome the previous one. This led to users not being
able to recognize the low-concentrated smells. Systems that are not mounted are often
placed at a distance from the user, which automatically produces latency in the system.
11
2. Related Work
When the extraction of the smell happens, possible airflow fluctuations or positional
changes of the user can make the smell less effective and probably even unconceivable,
which makes it very complicated to adjust smells to the amount and intensity that an
odor would require so that the recipient can recognize the smell accordingly [IBM+14].
Another study, by De Paiva Guimarães et al [DPGMD+22], presented a non-intrusive,
mobile, and low-cost wearable olfactory display combined with a software service that
enables developers to integrate olfactory stimuli into VR applications easily. The system
uses Google Cardboard, a Raspberry Pi 3, and Unity as the game engine. A small olfactory
device was mounted on the front of the Google Cardboard headset, as illustrated in
Figure 2.2.
Figure 2.2: Google Cardboard VR headset equipped with a wearable olfactory display.
The system integrates a Raspberry Pi, an odor repository, and a diffuser controller for
scent delivery during VR interaction [DPGMD+22].
De Paiva Guimarães et al. [DPGMD+22] also conducted a user study to evaluate how
comfortable and usable the wearable system was, and how users perceived the integration
of scent into VR. In the virtual scene, as can be seen in Figure 2.3, users walked across a
bridge toward a set of flowers. Upon reaching the flowers, a stream of vapor containing
either lavender cologne or plain water was released. After the experiment, participants
were then asked to complete a questionnaire. Results showed that 94 percent of users who
experienced the lavender fragrance agreed or strongly agreed that the odor was pleasant,
and 94 percent agreed that odors should be part of VR applications. In contrast, of the
group that smelled the pure water odor, while 81 percent still agreed with the statement
that odors should be part of VR applications, only 25 percent rated the smell as pleasant,
further supporting the role of scent in enhancing VR experiences [DPGMD+22].
12
2.3. Olfactory Technologies and Evaluation in VR
Figure 2.3: Virtual Reality scene used in the user study. Participants walked toward a
flower field, where scented vapor, lavender or plain water, was emitted to evaluate the
impact of olfactory cues in immersive environments [DPGMD+22].
In parallel to systems integrated into VR headsets, other wearable devices offer standalone
solutions for context-aware scent delivery. In general, wearable olfactory systems are
gaining interest, with numerous use-cases emerging beyond notification or integration into
VR systems, including support for relaxation, well-being, and enhancement of physical
activities. Dobbelstein et al. [DHR17], and Amores et al. [AHD+18], propose wearable
olfactory devices that are not integrated with VR, but worn around the neck, and which
both rely on the atomization of liquid phase odorants.
Amores et al. [AHD+18] present BioEssence, which is a wearable olfactory display that
can be worn as a clip or as a necklace, visible in Figure 2.4. It can also easily be attached
to various objects, including pockets, shirts, necklines, jewelry threads, and cords. The
BioEssence device was developed in response to research showing that essential oils such
as lavender, citrus, or vanilla can reduce anxiety, improve sleep, and alleviate pain and
depression. BioEssence is a novel wearable olfactory display capable of releasing up to
three different scents based on the users physiological state. For example, lavender scent
is emitted when the heart rate exceeds 80 beats per minute. The device is lightweight,
wireless, and designed to be worn daily, allowing for seamless, context-aware scent delivery
throughout the day.
13
2. Related Work
Figure 2.4: BioEssence: A wearable olfactory display that can be clipped onto clothing
or worn as a necklace. The device delivers scents based on real-time physiological signals
such as the heart rate of wearers [AHD+18].
In addition, Amores has published another paper with Maes [AM17] on the topic of
how olfactory interfaces can, for users unconsciously, influence the mood and cognitive
performance of wearers of the olfactory device. Amores and Maes presented Essence,
which is an olfactory computational necklace, which can be seen in Figure 2.5, that can
be remotely controlled through a smartphone. It was possible to change the intensity
and the frequency of the released scent, based on biometric or contextual data. The
paper discusses the role of smell in designing pervasive systems that affect the mood
and cognitive performance while being asleep or awake, and a set of applications for the
device is given and reviewed. Due to its lightweight and wireless design, the necklace can
be worn throughout the day and night, making it suitable for various scenarios.
Amores and Maes identified their necklace as beneficial not only for sleep, memory, and
learning, but also for improving cognitive performance and social interactions, particularly
in immersive environments such as VR or MR. It is mentioned that in a physical social
interaction, people are constantly using their sense of smell and unconsciously responding
to the present odors [AM17].
Figure 2.5: The Essence Necklace. A contains the fragrance, while B displays the device
which manages the function [AM17].
14
2.3. Olfactory Technologies and Evaluation in VR
Dobbelstein et al. [DHR17] on the other hand, built the inScent device, which can be
worn on a necklace, that contains up to eight different smell cartridges. inScent also
refers to the effect of smells being used to create or enhance experiences and is used to
utilize artificially emitted scents to invoke emotions and experiences for users in everyday
life situations. Primarily, the use case of the olfactory device in the Dobbelstein et
al. study is to complement and amplify received mobile notifications by using scent as
an additional emotional notification channel, which they call scentification. Messages
that the life partner of the wearer sends can, for example, be emphasized by emitting a
pleasant, related scent, such as flowers or the other person’s perfume aroma, to reflect
the emotional link to that person.
In Figure 2.6, a user can be seen wearing the inScent smell device on a necklace. When
the user receives a message, an odor is deployed. In pleasant anticipation, the user then
reaches for her phone to read the message.
Figure 2.6: A user wearing the inScent wearable olfactory display. The sequence shows
the process of receiving a message via scent (a–c), followed by reaching for the phone in
pleasant anticipation (d) [DHR17].
The main contribution of the inScent device is that it is a novel, unobtrusive, wearable, and
miniaturized olfactory display that allows a passive amplification of received notifications
with scents, by using an open-source hardware and software framework that allows
developers and researchers to add scents to their mobile applications and use cases. In
addition, the study finished with a qualitative user study, investigating the perception of
such scentifications of users in public [DHR17].
The use of scents can be applied in almost any context, as demonstrated by Zhou et
al. [ZZAL25], whose system, the E-scent Coach, provides olfactory cues to assist users
with breathing control during yoga exercises. While the correct breathing technique is
essential for every sport, the coordination of yoga postures with deep and slow breathing
15
2. Related Work
is one of the core elements for practitioners. When a yoga posture expands the chest or
the abdomen, such as when stretching, backwards bending, or when moving upward, it
requires inhalation to enhance the effect of the pose, while compression of the same areas,
such as when practitioners twist, bend forward, or move downwards, requires exhalation.
Due to the poses being challenging, practitioners often become too focused and, as a
result, neglect the rhythm of their breathing, which affects the effectiveness and increases
the risk of exercise-related issues due to poor breathing technique.
Many studies have already tried resolving this issue by leveraging visual, auditory, and
haptic feedback during yoga sessions. However, these methods require users to constantly
focus on additional feedback, making it even more challenging for them to focus on
everything at once. Due to olfaction being processed rapidly and unconsciously by the
human brain, olfactory feedback for guiding breathing has a lower cognitive load for users,
allowing them to focus better on their physical activity. Studies have shown that scents
can indeed alter breathing patterns, as demonstrated by [VBMRS14]. Furthermore,
scents are becoming increasingly embedded into applications to enhance multi-sensory
experiences, such as through the use of tools from Aromatherapy [ZZAL25].
Figure 2.7: The E-scent Coach is a wearable olfactory device designed to support correct
breathing during yoga sessions. Many practitioners struggle with proper breathing
technique. Using a webcam to analyze the users posture in real-time, their pose is
classified into either expansion or compression. Based on that, the wearable releases or
pauses scents to guide inhalation and exhalation, promoting a stable and deep breathing
rhythm [ZZAL25].
Zhou et al. [ZZAL25] followed up their research with a user evaluation with 16 participants.
Participants practiced 12 different postures, each for a total of 5 minutes, in 5 different
scent conditions. While one condition included having no additional smell present, the
others were based on different smell families, which are listed in Figure 2.8.
16
2.3. Olfactory Technologies and Evaluation in VR
Figure 2.8: Four Scent Families and their Representative Scents for the Experiment of
Zhou et al. [ZZAL25]
The study by Zhou et al. [ZZAL25] finishes by discussing that participants found all
scented conditions to be effective and usable, emphasizing the potential of scents to
influence breathing patterns. The results showed that high-arousal scents can activate the
olfactory system more effectively, including physiological changes in users and enhancing
perceptual experiences. Furthermore, the results indicate that the inclusion of smell as a
guiding tool has the advantage of being seemingly implicit and unconscious for users,
which allows them to focus on their breathing and thereby enhance their immersion
naturally.. It was verified that the use of scents in yoga can not only stabilize the emotion
and enhance the concentration of practitioners, but also serve as a medium of information,
resulting in enhanced respiratory responses [ZZAL25].
Wearable displays remain a key area of research, particularly when exploring mobile and
embodied olfactory interactions, as proposed in the present work, due to their portability
and potential for synchronized delivery in personal VR scenarios.
2.3.2 Room-Based and Static Emitters
Room-based or static olfactory systems involve devices placed at fixed positions in the
environment. They are commonly used in labs or exhibitions and offer a non-intrusive
alternative to wearables.
A prominent example is the oPhone DUO, see Figure 2.9, a table-top transmitter that
could send digital scent messages over the internet. While innovative, it lacked integration
with user position or timing, limiting its effectiveness in immersive environments [DVO16].
17
2. Related Work
Figure 2.9: oPhone DUO [DVO16]
Another example is Scentee, visible in Figure 2.10, a scent-based add-on for iOS devices
like the iPhone. It connects via the headphone jack and uses replaceable scent cartridges
that are activated through an app. Users can control the duration and interval of scent
releases, which are emitted as a misted cloud that requires proximity (around 10–15cm)
to be detected. While Scentee allows user control, it supports only one scent at a time,
limiting its expressiveness in immersive environments [DVO16].
Figure 2.10: Scentee on an iPad [DVO16]
Lastly, also prominent, is the Aroma Shooter device by Aromajoin, which is shown in
Figure 2.11. It connects via USB and includes six scent cartridges. These cartridges can
be precisely activated with directional airflow, enabling controlled scent transmission up
to 30cm after about 3seconds. The system requires the user to be accurately positioned, as
even slight deviations may result in missed scent perception. Developers can adjust which
scent is deployed, how intensely it is released, and the blower activation time. According
to [DVO16], the Aroma Shooter achieves high distance and volume performance with a
high-speed output.
To improve spatial targeting, Ischer et al. [IBM+14] proposed directional scent jets that
project odors at the user, reducing dispersion time. However, even these setups suffer
from air-based delivery delays, making it challenging to match scent release to precise
virtual events. Additionally, when using multiple smells in such manner, smells start to
overlap, where previously released odors still linger in the air, which poses a significant
challenge in multi-smell sequences [LYZ+23]. These mismatches can break immersion or
create confusion for users.
18
2.3. Olfactory Technologies and Evaluation in VR
Figure 2.11: Aroma Shooter device by Aromajoin [DVO16]
Static emitters also struggle to support personalized or multi-user VR sessions and cannot
follow users through space. This limits their use in large-scale, free-roaming experiences,
where mobility is crucial.
An example of a projection-based setup is the Brain and Behavioral Laboratory –
Immersive System (BBL-IS) introduced by Ischer et al. [IBM+14]. This system featured
a fully interactive visualization environment designed to deliver complex scents at specific
times and spatial locations without contaminating the broader environment. As illustrated
in Figure 2.12, the setup included coated projection screens, a dedicated scenting chamber,
and real-time stimulus synchronization. Such precise and isolated delivery was essential
to avoid the blending or lingering of odor traces that can disrupt multi-scent sequences
and degrade realism.
Figure 2.12: Schematic representation of the "BBL-IS"-system [IBM+14]
Although highly controlled, room-based emitters like the BBL-IS or oPhone DUO struggle
to scale beyond isolated experimental setups. They often require careful positioning,
limited mobility, and environmental constraints that reduce their applicability in dynamic,
multi-user VR scenarios.
In addition to research prototypes, several commercial room-based olfactory systems exist
today. For instance, the "Moodo Smart Diffuser"1 offers a modular design with swappable
1https://moodo.co/
19
2. Related Work
scent capsules and remote app control, allowing users to personalize ambient scent profiles.
The "Aromajoin Aroma Shooter"2, though often used in research, is also sold commercially
and provides targeted scent delivery with fast-switching capabilities. Another example
is "Scentys"3, a French company offering programmable scent diffusion systems used
in retail and event installations. These systems reflect growing commercial interest in
olfactory display technologies, even though they are typically optimized for ambient use
and lack the real-time tracking or interactive control required for VR immersion.
2.4 Summary
Olfaction remains an underutilized sensory channel in both robotics and VR, despite
its well-established role in influencing human emotion, memory, and spatial perception.
While technological advances have made it possible to integrate scent into immersive
environments, most systems still face limitations in terms of personalization, spatial
accuracy, and mobility.
This chapter reviewed a wide range of olfactory systems, categorized into wearable,
room-based, and robot-mounted solutions. Each approach offers trade-offs in terms of
timing, localization, and user freedom. Recent efforts have moved toward more dynamic
setups that allow for controlled, contextual scent release in sync with virtual or physical
events.
To our knowledge, no prior research has explored robot-mounted olfactory displays in
immersive VR. This work introduces a novel approach to mobile scent delivery using a
quadruped robot, enabling precise spatial alignment with users during locomotion.
The novelty of this thesis, therefore, lies in combining robotic mobility with olfactory
output. By using the Boston Dynamics Spot robot as an active scent carrier, it becomes
possible to deliver smells at specific locations and times, supporting user-dependent,
real-time interactions. This approach moves beyond static or head-mounted solutions and
opens up new application areas such as therapy, storytelling, and large-scale interactive
installations.
2https://aromajoin.com/products/aroma-shooter
3https://www.scentys.com/en/
20
CHAPTER 3
System Design and
Implementation
This chapter outlines the design and implementation of the mobile olfactory system
developed for this thesis. It provides a comprehensive overview of the core components,
both hardware and software, and describes how the Boston Dynamics Spot robot, the
Olorama scent generator, and the Unity control system were integrated to enable large-
scale smell interaction in VR. The chapter details how the system enables location-based
scent delivery synchronized with user position, and lays the groundwork for the evaluation
conducted later in the study.
3.1 System Overview
This thesis investigates how olfactory interaction can be effectively deployed in large-
scale physical environments using a mobile, multi-sensory system that combines robotic
mobility with olfactory output to create dynamic, location-based scent interactions in
physical and virtual environments.
To explore this, a custom system was developed, which integrates the Boston Dynamics
Spot robot and the Olorama scent generator through a unified control architecture
implemented in Unity, serving as the central hub for coordinating mobility, interaction,
and scent output.
By doing so, the exploration of different scents and intensities, which are controllable
through Unity’s interface by simply adjusting the timing, duration, and frequency of
scent releases, becomes possible. The feature allows for future experimentation with
different perceived scent strengths in a diversity of test settings and enables future studies
on how users react to variations in olfactory outputs under different conditions.
21
3. System Design and Implementation
Lastly, the system is designed to be modular, allowing for flexible scenario design and
scalable experimentation in real-world conditions. The technical setup integrates multiple
components, which are more detailed in section 3.2, such as a Windows 11 PC equipped
with an Intel Core i9-9900K @3.6GHz and an NVIDIA RTX 2080Ti (W-PC), serving as
the central control and computation unit. Other components include:
• The Boston Dynamics Spot robot acts as the mobile transportation tool for the
scent generator as well as for the tracker. The robot can navigate a wide range of
environments with precision and safety while carrying the tools, thereby enabling
the delivery of scents to specific physical locations or users. The robot is controlled
through TCP commands received from a Python-based server, enabling precise
mobility within the environment.
• The Olorama Scent Generator, a compact electronic scent device capable of releasing
multiple scents on command, is mounted on top of Spot. It receives trigger signals
via Wi-Fi from a Unity application, which allows precise control of the type, timing,
and duration of scent release.
• An HTC Vive Tracker, mounted on Spot for real-time position tracking of Spot
Robot within a defined VR space, used to calculate the distance between the robot
and the user’s HMD in real time for proximity-based scent triggering.
• A VR Setup, including an HTC Vive Head-Mounted Display (HMD) and VR
controllers, is used to place users within a multi-sensory, spatially coordinated
environment. In addition, Vive Base Stations, which are mounted on the walls in
the room where the user study takes place, are used to enable real-time tracking of
both users and the robot in the physical space.
• The central control logic is implemented in the Unity Engine, which coordinates
robot movement, scent deployment, and user interaction. Unity connects to both the
Spot SDK and the Olorama device, serving as a tool for designing the environment
and executing interaction scenarios in the application.
• A Python-based backend server acts as a communication bridge between Unity and
Spot. It handles messaging, command forwarding, and timing coordination between
these two components, ensuring low-latency and synchronized operation, acting as
a middleware that translates the relative goal position into Spot’s coordinate frame
before sending the command. Olorama, on the other hand, is directly controlled
from Unity via UDP, without any backend server.
22
3.1. System Overview
Figure 3.1: Overview of the hardware components (top) and software system (bottom).
Red arrows indicate TCP-based communication (Unity to Spot via Python), while blue
arrows represent direct UDP commands (Unity to Olorama).
Altogether, these components, as illustrated in Figure 3.1, form a modular and tightly
integrated platform for exploring mobile, multi-sensory interaction through scent in
real-world and virtual environments. The system coordinates multiple hardware and
software components in real time, ensuring a synchronized experience across physical
robot movement, olfactory output, and user interaction in VR.
A typical execution loop proceeds as follows:
1. Unity continuously monitors the user’s position in the virtual environment and
determines when the user approaches an interactive object or smell zone.
2. Upon reaching a predefined distance threshold, Unity sends a TCP message to a
custom Python server, specifying the next target position for the robot.
3. The Python server, acting as a middleware layer, parses the command and uses
the Spot SDK to control the Boston Dynamics Spot robot, instructing it to walk
toward the designated position.
4. As Spot arrives near the goal, it readjusts its orientation to face the user, ensuring
optimal delivery conditions for the upcoming scent emission.
23
3. System Design and Implementation
5. Simultaneously, Unity transmits a UDP message directly to the Olorama scent
generator, containing the spray parameters such as scent channel, duration, and
intensity.
6. The Olorama device releases the scent into the environment. At the same time,
Unity triggers a visual feedback system, spawning a subtle particle effect above the
object, accompanied by a light glow and an auditory cue, to indicate its activation.
This process happens with near real-time responsiveness and requires precise coordination
between spatial logic, network communication, physical hardware, and VR feedback.
Section 3.2.3 provides a more detailed technical breakdown of the communication flow
and implementation details.
3.2 Hardware and Software Setup
The following sections present the technical setup of the system in more detail, beginning
with the robotic platform and continuing with the scent hardware and supporting software
components.
3.2.1 Boston Dynamics Spot
General Description
Boston Dynamics is known for developing bioinspired robots with agility, stability, and
complex motion capabilities that push the latest technology to new heights. One of these
robots is the quadruped Boston Dynamics Spot Robot, which, in this thesis, serves as the
central mobile platform in this project and is responsible for carrying the Olorama scent
generator, which can be seen in Figure 3.2, thus enabling mobile scent deployment while
simultaneously acting as as a moving tracked unit within a VR-aligned space, enabling
dynamic and spatially accurate smell delivery.
24
3.2. Hardware and Software Setup
Figure 3.2: The Boston Dynamics Spot robot equipped with the Olorama scent generator
and the HTC Vive Tracker. The scent device is mounted on the rear plate for spatial
alignment and synchronized scent emission based on the user’s location.
Spot’s hardware is almost entirely custom-designed and includes processing boards for
more control as well as advanced onboard computation and perception systems. Sensors
are mounted on the front, rear, and on the sides of the robot. Stereo cameras, wide-angle
lenses, and structured light projectors are distributed across the whole body, enabling
robust 3D sensing and autonomous operations. Spot is capable of forward and lateral
25
3. System Design and Implementation
walking, turning in place, and climbing, with a top speed of 1.6m/s. Its legged morphology
allows navigation across uneven terrain, making it suitable for complex or unpredictable
physical environments.
In the scope of this study, Spot was used to navigate between two spawned target objects
and respond to proximity-based user interactions. Since users wore VR headsets and
could not see the robot physically, the motion system had to be tightly coupled with
virtual content. Spot was responsible not only for locomotion but also for stopping,
rotating to face the user, and releasing scents at predefined spatial points. This setup
created spatially contextual olfactory interactions that would not be possible with static
scent emitters [Gui19].
Spot has a payload capacity of up to 14kg, and due to its precise trajectory execution
ability and the possibility of mounting the Olorama onto the robot, the Boston Dynamics
Spot was an ideal choice for this study. The mobile scent generator setup directly
supported the thesis objective of investigating scalable and flexible smell delivery in VR
environments. In addition, a Vive tracker was mounted to Spot’s rear plate, enabling
the robot’s spatial tracking within the system. Unity used this tracker to localize Spot
in real-time and trigger scent emissions based on the users location and the predefined
procedure logic. Particular attention was given to the mounting configuration to ensure
both visibility for the tracker and unrestricted motion for the robot, while keeping the
robot, the tracker, and the scent generator safe.
Spot was then controlled indirectly from Unity using a custom Python server, which
interpreted JSON messages over TCP and translated them into motion commands via
the Boston Dynamics SDK. For more on the control pipeline and messaging architecture,
see Section 3.2.3. The robot’s movement execution is explained in more detail in the
following subsection.
Motion Control and Spatial Alignment
To control Spots movements in a precise and spatially aligned way, all motion was defined
using relative poses in something named “SE(2)”. SE(2) is a combination of multiple
notations and stands for the mathematical space of Special Euclidean transformations
in 2D. In SE(2), “S” stands for “Special” and implies that the transformation preserves
the original orientation of an object without any flipping. “E” is short for “Euclidean”,
which means that the distance is preserved, without any stretching applied. And lastly
(2) correlates to working in the 2D space with an x-position, a y-position, and a heading.
In other words, SE(2) combines all possible 2D movements that translate and rotate an
object, but do not scale or skew it, and, in this study, therefore, is the math space for
moving Spot around on a flat surface to move the robot forward, sideways, and to make
it turn. In general, SE(2) is commonly used in robotics for mobile navigation. Each pose
can be encoded into a 2D position (x,y) and a heading angle θ, compactly represented as
26
3.2. Hardware and Software Setup
a 3×3 matrix:
T =
cos(θ) − sin(θ) x
sin(θ) cos(θ) y
0 0 1

This matrix-based representation allows combining transformations via simple multipli-
cation. In this thesis, Spot’s relative goal poses were computed in Unity and transformed
into the robot’s global coordinate system using:
T global
goal = T global
body · T body
goal
This composition demonstrates that the global position of a goal can be determined
by taking the robots current global pose and applying a relative offset from its body
frame. In essence, the goal’s global position is determined by transforming its local
offset via Spot’s current world pose. The following SDK code was used to perform this
transformation:
out_tform_body = get_se2_a_tform_b(transforms, frame_name, BODY_FRAME_NAME)
out_tform_goal = out_tform_body × body_tform_goal
In Unity, the relative position of a goal object was computed using matrix multiplication.
Spot’s tracked transform (via the Vive Tracker) and the goal object’s transform were
used to calculate their relative offset:
Matrix4x4 relativeMatrix = t1Matrix.inverse * t2Matrix;
Vector3 local_position = relativeMatrix.GetColumn(3);
However, since Unity and Spot use different coordinate conventions, a frame conversion
was necessary. Unity uses a left-handed coordinate system (with +Z forward), while Spot
uses a right-handed system (+X forward). The terms “left-handed” and “right-handed”
refer to how the three axes (X, Y, Z) are oriented in a 3D space, and specifically how to
determine the third axis depending on the first two. A practical example can be shown
with everyone’s own hands. When making an L-shape with the left, by pointing the index
finger forward, which resembles the Z axis, and by pointing the thumb to the right, which
stands for the X axis, the middle finger, the Y axis indicator, will point upward. On,
literally, the other hand, when doing the same procedure, the middle finger, which again
resembles the Y axis, will point downward. This is the indicator that the coordinate
system is adjusted to “right-handed”.
This transformation therefore affects how objects are rotated, how angles are measured,
and also how coordinates are converted. By ignoring it, everything will move in the
wrong direction or be mirrored falsely. Regarding this thesis, when Unity computes a
position relative to Spot in VR, it does so in left-handed space, but Spot’s SDK expects
27
3. System Design and Implementation
right-handed inputs. Consequently, the coordinate axes must be converted accordingly
before sending the motion commands. That is precisely what the next few lines of code,
which are essential, do.
jsonObject["x"] = -local_position.y; jsonObject["y"] = -local_position.x;
jsonObject["yaw"] = 360 - local_rotation.eulerAngles.z;
This flips and swaps the axes, so that Spot walks in the correct real-world direction.
In addition, in the last line, another transformation was applied to match Unitys local
offset to Spots format, which, in total, mathematically can be displayed by the following
equation.
xspot = −yunity, yspot = −xunity, θspot = 360◦ − θunity
Then, once the converted pose is calculated, it is sent to a custom Python server using
TCP. The server then generates an SDK-compatible Spot movement command, with
which the robot knows what to do next. Furthermore, to control Spot’s walking speed,
Unity included a speed parameter in the JSON message. This is translated into a velocity
constraint within the Python server:
vel_limit = SE2VelocityLimit(max_vel=SE2Velocity(x=speed, y=speed),
angular=1.0)
Before scent emission, Spot was also instructed to rotate and face the user. This
orientation was computed based on the relative angle between Spot’s current pose and a
designated Unity object, which in this case was the HMD that the users wore during the
study experiment. A yaw-only command was issued to ensure that the scent generator
pointed in the user’s direction, increasing olfactory clarity. A yaw-only command tells
Spot to turn around on its vertical axis, basically spinning the Spot. Spot was therefore
instructed to only rotate in place and to change its facing direction, to point towards the
user’s position, but not to move forward or sideways, so that the Olorama scent generator
is aimed directly at the user’s location. This ensures that the smell is emitted towards
the nose of the user, instead of into a random direction. This improves how clearly and
quickly users can perceive the scent.
Spot’s entire behavior is based on a self-written predefined Unity procedure, which
controls the sequence of calibration, movement, rotation, and scent release. This structure
ensured consistent interaction timing across participants and made the movement logic
reproducible, with each part working independently yet synchronously.
3.2.2 Olorama Scent Generator
In this study, the Olorama scent generator was physically mounted on the Spot robot to
turn it into a mobile scent-delivery unit. Instead of having a static smell device, fixed in
just one spot, the idea was to make the olfactory output move through space, triggering
28
3.2. Hardware and Software Setup
scents dynamically and depending on the user’s position in the virtual environment. This
setup allows for more context-sensitive and spatially aligned smell experiences.
The Olorama system was selected due to its modularity, its wireless communication
capabilities with different plugins and SDKs over Wi-Fi, as well as for its ease of integration
with third-party software such as Unity. It is possible to release up to 10 different scents
on command, making it a good match for this mobile VR scenario where multiple smells
are deployed, without having to use extra cables or an onboard PC.
Communication with the Olorama is possible through direct UDP messages that are
being sent from Unity. During this thesis, a custom UDP sender script was implemented
in C#, which formats and transmits the control strings according to Olorama’s command
structure. Messages include the definition of the output channel, the intensity, spray
time, fan duration, and a few other parameters. Unity sends this message over the local
Wi-Fi network directly to the device’s IP address. Unlike the robot, which is controlled
via a separate Python server, Olorama does not require any backend, since it is controlled
entirely from within Unity.
In the experiment, the scent trigger is tied to user proximity in the VR environment.
When the user gets close, below a certain threshold distance, to a designated smell object,
Unity sends the corresponding UDP message to release the scent in the real world. While
the Olorama hardware does not allow direct control over intensity or fan speed, these
effects can still be approximated by adjusting the timing and duration in the Unity
Inspector.
The Olorama Scent generator was then mounted onto the Spot using 3M Dual Lock
adhesive strips, which provided a non-invasive, vibration-resistant method of attachment
that could be adjusted easily and did not require drilling. This ensured that the device
remained stable during the robot’s movement without obstructing the robot’s sensors or
its locomotion ability.
3.2.3 Software Architecture
The software architecture was designed to enable coordinated operation between multiple
components: the mobile Spot robot, the Olorama scent generator, the virtual environment
in the Unity game engine, and user interactions within the virtual world. Unity serves
as the central orchestration layer, managing user tracking, triggering Spot’s movement
through a defined and automated procedure, handling the scent release logic based on
user behavior, and providing both automatic and manual control options to monitor and
adjust the application. All components are synchronized in real time via message-based
communication protocols but remain largely independent from one another. For example,
if a user does not approach an object linked to scent deployment, the scent generator
will not release any odor.
To coordinate these interactions and ensure that all components are informed when a
scent should be released, the system introduces the concept of “Smell Zones.” Smell Zones
29
3. System Design and Implementation
are predefined positions in the virtual environment where the Spot robot must arrive in
order to trigger a scent. These zones are placed in Unity’s world space and are linked
to specific virtual objects, such as small rocks used in the experimental scene, in this
thesis. When a user approaches a smell zone, detected through the real-time tracking of
their head-mounted display (HMD), and enters a defined distance threshold, the system
initiates the next procedural step by sending the corresponding commands to Spot and
Olorama.
Upon detecting such an event, Unity generates a structured JSON command that
describes the desired action. This message is sent via TCP to a custom Python-based
server running locally, which parses the command and translates it into instructions
for the appropriate hardware. Commands related to Spot, such as movement, posture
changes, or yaw rotation, are forwarded to the Boston Dynamics SDK and executed
through its interface. In contrast, scent-related commands are sent as UDP messages
directly to the Olorama device, containing the scent channel, spray duration, fan timing,
and related parameters.
Rather than relying on Unity’s classic trigger zones, which depend on collider-based
physics events, the system uses a custom distance-based approach. This method computes
the real-time proximity between the user and each Smell Zone without requiring any
physical colliders or rigid bodies in the scene. It allows for precise and context-sensitive
activation of both the scent release and Spot’s orientation logic. Once the user enters the
defined proximity range, a scent is emitted, and before each release, Spot consistently
realigns its yaw to face the user, ensuring the scent is directed toward the users nose for
maximum perceptibility. This orientation is calculated in Unity and sent as a dedicated
rotation command to Spot via TCP.
To support the development and allow for detailed debugging and calibration, a manual
control interface was implemented directly in Unity. This inspector-based interface
enables researchers to send predefined Spot commands, override automatic behavior, or
verify system functionality during testing. The User Interface includes numeric fields
for defining Spot’s walking speed and rotation angle, as well as interactive buttons for
commonly used actions. Each button sends a structured TCP command to the Python
server. The available commands are summarized in the following table, Table 3.1, and
can also be seen in Figure 3.3:
Button Description
Boot Initializes and connects Spot to the SDK
Move Sends a test movement command with a relative pose
Rotate Rotates Spot based on a user-defined yaw value
Bend Changes Spot’s body posture (e.g., to a lowered stance)
Bend Reset Resets Spot’s posture to the default upright configuration
Stop Interrupts the current Spot activity or walking routine
Table 3.1: Manual control functionalities available in Unity’s Spot interface.
30
3.3. Evaluation Task and Study Logic
Figure 3.3: Inspector-based interface in Unity for manual Spot control.
While this manual interface greatly aided development and debugging, it also represents
a key layer in the system architecture. During user studies, it served as a fallback
mechanism in case of tracking failures or unexpected robot behavior. Most importantly, it
ensured that every action Spot performed was intentional, either by being triggered by a
well-defined automatic procedure or manually initiated by the experimenter. This hybrid
design supports both flexibility and safety, which are essential for real-time, user-facing
robotic systems.
3.3 Evaluation Task and Study Logic
The goal of the evaluation task was to assess the technical feasibility and timing accuracy
of a mobile scent delivery system integrated into a spatial VR experience. To do so,
a walking-based scenario was designed in which users moved through a small forest
environment and encountered three olfactory interaction points. Each of these points
was linked to a specific virtual rock and triggered the release of a scent when the user
entered a proximity range during a predefined sequence. The idea for this project was
to create a natural setting in which olfactory interactions would make intuitive sense
and where scents like wood, wet ground, or freshly cut grass would feel immersive rather
than artificial, such as in realistic scenarios. The system was designed to coordinate the
robot’s movement, orientation, and scent release based on the user’s position and task
progression.
The entire virtual forest environment was implemented in Unity and sized to fit within a
5×5 meter tracked space. As shown in Figure 3.4, users were surrounded by a square-
shaped wall of trees, forming the boundaries of the forest scene, and small rocks, which
represented the interactive smell goals that were positioned at the four edges of the
virtual environment, were placed to define the olfactory targets. The Spot robot would
move to the small rocks, the smell goals, in a predefined global sequence that stayed
the same throughout all trials with all participants. One additional large rock in the
center served as a visual obstacle and safety measure, ensuring that users moved around
it instead of walking directly towards the robot. This central rock also acted as a starting
position for the robot before the sequence began, due to users being told not to step
straight into the rocks.
31
3. System Design and Implementation
Figure 3.4: Top-down view of the virtual forest environment with predefined smell targets
and boundary trees.
The experience begins with the user standing at the starting rock on the left side of
the forest scene. No scent is released at this point. After the study examiner manually
initiates the procedure, the Spot robot autonomously moves to the first scent location.
Once the robot reaches its designated pose, a soft blue smoke particle effect appears above
the first target rock to signal the active smell zone. A subtle glow effect is also used to
guide the user’s attention. The smoke never appears before Spot is in position, ensuring
users do not approach the robot prematurely. In general, users are initially instructed to
start walking only when a new smoke effect appears. During the procedure itself, after
having done the first smelling, they can hear a subtle sound cue, which clarifies that a
new smoke has spawned somewhere in the environment, which they have to find, which
signals them that they can start moving again. As shown in Figure 3.5, this particle
32
3.3. Evaluation Task and Study Logic
system rises gently and fades upward to provide a clear but unobtrusive signal. Together
with the glow, it helps users visually locate the scent interaction point in the forest scene.
Figure 3.5: Subtle blue smoke with glow effect marks the active scent zone in the VR
scene.
When the user walks towards the smoke and enters a defined distance threshold of 75
cm from the target object, the smoke effect disappears. At this moment, Spot aligns its
body to face the user’s head position, ensuring that the scent is sprayed directly toward
the user. Users are instructed to remain stationary once the smoke disappears, both to
avoid walking into Spot and to ensure consistent delivery conditions. However, when
Spot is moving and the user is not, it moves around objects and the user, making this
measure essential for both user safety and the robots safety.
Figure 3.6 depicts the synchronized real-world and virtual implementation of the scent
emission mechanism. On the left, the Spot robot releases the scent upon proximity
threshold detection; on the right, a corresponding visual particle effect is rendered in the
virtual environment to indicate scent delivery.
33
3. System Design and Implementation
Figure 3.6: Visualization of synchronized real and virtual scent emission once the 75cm
threshold is crossed.
Once the user detects the scent, they confirm the perception by pressing the button on
the back of their VR controller. If no confirmation is received within 20 seconds, the
system automatically continues to the next step. This setting can be seen in Figure 3.7.
In such cases, an auditory cue is played to signal that the time has expired, and a new
smell interaction is starting. The robot then moves to the following target location, and
the particle glow system is reactivated at the new smell zone to guide the user. Users are
told that they may resume walking once a new smoke zone appears.
Figure 3.7: Inspector settings in Unity defining the smell intensity and timeout logic for
the evaluation sequence.
This cycle is repeated for three olfactory interaction points, each associated with a different
smell, which are "wet ground", "wood logs", and "roses". The objects are arranged to
34
3.3. Evaluation Task and Study Logic
encourage a walking path from the left to the bottom, then to the top, and finally to the
right. After the third interaction, and once the scent is confirmed or the timeout elapses,
Spot performs a final behavior by stopping and sitting down. This indicates the end of
the trial, and users can take off the HMD.
This entire sequence was repeated across all participants under identical conditions and
serves as the foundation for the study design discussed in Chapter 4.
35

CHAPTER 4
Evaluation and Results
This chapter presents the evaluation and results of the Spot-based scent delivery system,
focusing on its effectiveness and on how users perceive smell within a VR environment.
The main objective of the evaluation was to investigate how different smells, with varying
intensities and smell types, affect the perception of participants and overall experience in
VR. Furthermore, the evaluation aimed to assess whether such a mobile scent system
could be integrated into future studies, enabling it to be applied to more complex use
cases and features. By considering various measurements, this chapter aims to inform
about the usability and impact of future large-scale olfactory VR systems, laying the
groundwork for potential improvements and broader deployment in similar applications.
4.1 Study Design and Hypotheses
The user study was designed as an initial evaluation of the Spot-based scent delivery
system, aiming to investigate how different smell intensities and smell types influence the
ability of participants to detect smells, as well as their overall experience. An additional
goal was to assess the system’s applicability for future studies involving more complex
use cases, interactions, and scenarios.
Three naturalistic scents, already described in the section 3.3, were used to match the
forest theme and provide sensory contrast. Additionally, three scent intensity levels
(200, 400, and 600) were selected based on the Olorama specifications for low, medium,
and high intensity, according to their user documentation. These values were chosen
to explore how the strength of olfactory input may affect detection performance and
perceived intensity throughout the whole thesis.
Deciding on the three different types of smells was a long process, ranging from choosing
the right environment, such as in this case a forest, rather than, for example, a beach
or a meadow full of flowers, to determining which forest elements should be displayed
37
4. Evaluation and Results
throughout the procedure. In early iterations, a wider range of smells was explored using
both preset scent capsules from the Nature Pack, such as lavender, jasmine, wet ground, or
rosemary, and a custom selection tailored to fit the designed environment, including scents
such as forest, peppermint, lemon, or coconut. This flexibility in smell selection, provided
by Olorama’s modular cartridge system, allowed for thematic alignment and emotional
tone adjustments based on insights from aromatherapy and multi-sensory experience
design. The three specific smells, namely wet ground, wood logs, and roses, were chosen
to represent a balanced and realistic range of forest-related olfactory impressions that
differ clearly in character. Each scent reflects a distinct semantic category within the
natural environment:
• Wet ground was chosen as a subtle, earthy base note often associated with freshness
and outdoor nature.
• Wood logs represented a more neutral, dry, and woody tone, typical of forest
surroundings.
• Roses added a floral and more intense contrast to the other two, providing a
distinguishable scent with higher perceptual salience.
This selection ensured that the user experience remained thematically coherent with the
forest scene, while also enabling the study to explore how smell intensity and character
affect detectability and perception. The combination allowed for comparative analysis
between a subtle (wet ground), medium (wood), and intense (rose) scent profile, helping to
understand which types of natural smells are more effective or salient in mobile olfactory
VR contexts.
In addition, it had to be tested which fan duration should be chosen as a constant, as
well as which distance the Spot should maintain from users before emitting the smell
to, on the one hand secure a safe environment, but on the other hand provide sufficient
testing possibilities. In the end, the fan duration was set to a constant of 4 seconds,
which was recommended as a value from the Olorama constructor for this study, and the
75 cm proved ideal.
A within-subjects experimental design was selected to ensure that each participant
experienced a range of smell intensities and smell types, thereby reducing the impact of
individual differences in olfactory sensitivity and VR familiarity.
To systematically vary the presentation of smell intensities and smell types, while
avoiding overwhelming each participant with all possible combinations, a Latin Square
counterbalancing design was implemented. In this approach, each participant was assigned
one specific combination of each smell type with one smell intensity per block, ensuring
that all smells and intensities were represented across all participants.
Each participant completed three experimental blocks, each block corresponding to one
specific smell intensity (low, medium, high). During each block, participants experienced
38
4.2. Participants and Apparatus
all three smell types, which were also all delivered at the same intensity. Since users went
through the experiment three times, namely one block per intensity, they experienced
each smell type at each intensity level across the whole session. This design ensured
that all intensities were tested for each participant, not within a single block, but across
the entire user test. The order of the smell intensities was counterbalanced, so that one
participant might have received the smell intensities in the order of low, medium, high,
while another participant might have experienced them in the order of medium, high,
low. This approach allowed us to systematically investigate both the intensity and the
smell type effects while minimizing bias due to the block intensity order.
The study was designed to test the following hypotheses:
• H1: Higher smell delivery intensities would result in faster detection times, due to the
increased concentration of scent molecules.
• H2: Different smell types would result in varying detection times, reflecting on how
easily specific scents can be recognized.
• H3: Higher smell intensities would lead to higher perceived smell strength ratings.
• H4: The addition of olfactory stimuli would not significantly increase VR sickness
symptoms.
These hypotheses guided the analysis of both the objective detection times and the
subjective ratings collected during the study. The aim was to systematically investigate
how users interacted with the system and how different smell intensities and types shaped
their overall experiences, ultimately providing valuable insights to guide the development
of future olfactory VR systems.
4.2 Participants and Apparatus
A total of 14 participants (7 female, 7 male) participated in the study. The ages of the
participants ranged from 21 years to 60 years (Mean M = 29.5, Standard Deviation SD
= 4.15). The distribution was pretty even across ages, with no significant clustering in
any particular range. None of the participants reported having olfactory disorders or any
prior experience with olfactory display technologies, making their impressions throughout
the study even more unbiased, genuine, and reflective of their first-time user experience.
At the same time, participants had different levels of exposure and familiarity with VR.
5 of the 14 users were first-time VR users, 5 had tried VR at least once before, and 4 of
them were regular VR users. All participants signed an informed consent form before
participating in the study and were naïve to the purpose of the study. The study was
carried out in a dedicated 5m×5m workspace where participants, equipped with an HTC
Vive Pro HMD, could move freely. To enhance immersion and safety, the setup included
a Vive Wireless Adapter, which eliminated the need for physical cables. This not only
allowed users to explore the environment unimpeded but also prevented potential tripping
hazards for both participants and the Spot robot. The battery for the Vive wireless
39
4. Evaluation and Results
adapter was stored in a small belt pocket or securely mounted on the user’s clothing, as
shown in Figure 4.1.
Figure 4.1: Left - Battery stored in a belt pocket. Right - Battery is securely mounted
on the users clothing.
The VR environment was developed in Unity3D (version 2021.3.25f1) and designed to
simulate a small forest scene, featuring trees, rocks, and ambient forest sounds. The
rocks were placed as obstacles for the users in such a way as to guide them through the
environment while preventing them from colliding with the Spot robot throughout the
navigation process, as well as waypoints for the Spot robot since the smells were placed
next to those objects, as can be seen in Figure 4.2. A large rock was placed in the middle,
under which the Spot robot started at the beginning of the study, while four small rocks
were placed 2 meters from the center of the workspace in each direction.
40
4.3. Procedure
Figure 4.2: Smell was released near designated small rocks in the environment, with the
Spot robot stopping and facing the user before scent emission.
In addition, the tree line was intentionally placed this way, as this prevented users from
going further than the 5m×5m real-life workspace. Although users were thoroughly
monitored during the experiment to make sure that they were as safe as possible anyway
and, for example, to allow them to walk as freely as possible without endangering them,
it was a great way to visually tell users that they have reached the end of the room
without interrupting their VR experience.
Ambient forest sounds were continuously played in the background to mask any auditory
signals that could reveal the timing of the scent release.
4.3 Procedure
Before the experiment began, each participant was welcomed, informed about the general
concept of the study, and briefed about safety regulations, such as the fact that Spot
moves through the room but is programmed to walk around users if necessary, safely. They
were then asked to sign a consent form, which they all did. Afterwards, they completed
a demographic questionnaire and filled out the pre-experiment Virtual Reality Sickness
Questionnaire (VRSQ) to establish a good baseline for physical well-being and sensitivity
to VR and smell exposure. Participants were then equipped with the HTC Vive Pro Eye
Headset and wireless adapter to perform a short VR training. They entered the virtual
41
4. Evaluation and Results
forest environment and familiarized themselves with walking in VR as well as with the
visual elements and obstacles. This phase ensured that participants were comfortable with
the interface, movement mechanics, and the immersive VR experience before the actual
experiment began. The main experiment consisted of three blocks, each corresponding
to one of the smell intensity conditions, described in Section 4.1. During each of those
blocks, participants completed the navigation task by locating three predefined scent goals
where they performed the smelling. The specific positions of the goals were unknown
to participants beforehand. As previously outlined, Spot autonomously navigated to
each goal, rotated to face the user, and triggered the corresponding scent at the assigned
intensity when the user approached. Participants then confirmed scent detection by
pressing a button on the back of the VR controller or were prompted to proceed if no
detection occurred within the time limit.
Throughout the session, participants were closely monitored to ensure their safety,
particularly as they moved freely within the 5m×5m physical space. The spot robot was
programmed to avoid close-range collisions with participants, or other objects, for that
matter. It actively rerouted its path around users when they were in its path between
one Spot goal and the next, without making physical contact.
After each block, which consisted of recognizing three smells at the same intensity,
resulting in a total of nine smell interactions per participant, participants filled out a brief
post-block questionnaire, assessing perceived scent intensity, recognition, and confidence
in detection. While participants completed the digital survey, which was administered
using Google Forms, the room was aired out for at least five minutes before beginning
the following condition. The reasoning for that was to diminish any residual odors from
the previous block and to minimize odor adaptation from users [BBEL71].
This procedure was repeated for the remaining two intensity conditions. Finally, upon
completion of all three blocks, participants completed a final post-experiment question-
naire, which included repeated VRSQ questions and open-ended feedback prompts. At
the end of the session, participants were debriefed on the purpose and goals of the study,
as well as on potential future applications for such mobile scent systems. They were
encouraged to share any final thoughts or observations they might have made.
In the following image, Figure 4.3, the procedure is visualized as a flowchart.
42
4.4. Data Collection and Metrics
Figure 4.3: Overview of the study procedure, including training, three scent intensity
blocks, 5-minute breaks between experimental blocks, final questionnaires, and the debrief.
Each session lasted approximately 45 minutes per participant.
4.4 Data Collection and Metrics
To evaluate the feasibility and effectiveness of the mobile olfactory VR system, multiple
data sources were collected throughout the study, which can be seen in Figure 4.4. By
continuously recording the position and orientation of both Spot and the participant in
43
4. Evaluation and Results
the virtual environment at 90Hz, and combining this with subjective self-report measures,
a comprehensive understanding of the user interactions and perceptions was obtained.
Figure 4.4: Overview of the collected evaluation metrics, their descriptions, and respective
data sources.
Different subjective measurements have been collected, one of them being the smell
detection time. For each of the three experimental blocks, the time it took participants
to detect a smell was recorded. The time measurement began when the Spot robot
triggered the scent and ended when the participant pressed the designated button on the
VR controller to confirm the detection of the odor. If a scent was not identified within
20 seconds by the user, a timeout was logged, and the experiment moved forward to the
next step of the procedure.
In addition, participants were asked to identify the perceived scent via an open-ended
text field in the questionnaire. This enabled the analysis of recognition accuracy for each
smell type across different intensities and smell types, providing great insight into how
reliably distinct smells could be identified in a VR setting. Furthermore, participants
completed a brief post-block questionnaire after every one of the three blocks, where
they rated the perceived intensity of the scents and indicated their confidence level in
identifying the smells. These metrics help to determine how well users can distinguish
between low, medium, and high-intensity settings and how sure they feel about their
ability to smell the correct odors.
To ensure that users would always feel safe and comfortable, they were monitored
throughout the whole study. Nevertheless, to analyse if the time in the VR environment
would have any impact on participants, they were asked to fill out a VRSQ [KPCC18]
before and after the experiment. This measured symptoms such as eye strain, dizziness,
44
4.5. Results
or disorientation, helping to assess the impact of olfactory stimuli and mobile robotic
interaction on VR sickness. VRSQ scores were then computed considering the delta score
from the pre-VRSQ scores.
In the post-experiment questionnaire, data were collected by asking users the following
subjective questions: “How much did you trust the VR system’s ability to simulate smells
accurately? (1 - Did not trust at all, 7 - Trusted very much)"; "How far were you (in
cm) from the smell source during the experiment?". Participants were also encouraged
to leave open comments to provide qualitative feedback on usability and the perceived
realism of the system.
In parallel to the subjective data, system-level logs were automatically recorded. These
included timestamps of scent release, positional tracking of both the robot and the user,
the distance to the scent source at the time of emission, and timeout events. These
objective metrics allowed cross-validation of self-reported responses with the actual
behavior.
4.5 Results
To analyze the collected data, non-parametric statistical methods were applied, as
the majority of dependent variables did not meet the assumption of normality. This
was confirmed using the Shapiro-Wilk test. For interval-scaled measurements such as
smell detection time, an Aligned Rank Transform (ART) ANOVA test [WFGH11] was
conducted, treating Smell Intensity and Smell Type as within-subject factors. Post-hoc
comparisons were performed using contrast tests based on the ART-ANOVA framework
[EKHW21]. For ordinal data and Likert-type responses, including the VRSQ scores,
perceived intensity ratings, and confidence levels, a Friedman ANOVA was used, with
the experimental block as the within-subject factor. Here, each participant completed
three experimental blocks, one for each smell intensity (low, medium, high), allowing for
the within-subject comparison across the conditions. When significant effects were found,
follow-up pairwise comparisons were conducted using Conover’s test with appropriate
adjustments.
4.5.1 Time and Trajectories
There was no significant effect of block order (χ2(2) = 4.00, p = 0.13) or Smell intensity
(χ2(2) = 0.57, p = 0.75) on the overall duration of a trial. On average, each block lasted
approximately 211±99 sec.
Participants generally stayed within the boundaries of the 5m×5m tracked space, and
Spots movement accuracy was continuously logged and compared to the coordinates
of each scent goal. Figure 4.5 below shows aggregated trajectories of both the Spot
robot and participants, revealing clearly visible clusters at (0, 2), (–2, 0), and (0, –2),
which correspond to the scent goal locations where Spot released the odors. The figure
also illustrates how participants navigated the virtual forest environment while avoiding
45
4. Evaluation and Results
central obstacles and the origin zone. To assess the precision of Spot’s positioning during
scent delivery, we compared its final pose with the target locations. On average, Spot
reached the scent goals with a positional deviation of M=10.87cm (SD=7.92cm) and an
orientation deviation of M=3.76° (SD=6.12°). Across all trials, the robot consistently
approached the goal from the correct direction and triggered the scent within a 75cm
radius of the user. To assess how accurately Spot reached the designated scent goal
locations, we analyzed its final position and orientation relative to the predefined targets.
On average, Spot stopped with a positional offset of M = 10.32 cm (SD = 8.51 cm) and
an orientation difference of M = 3.11° (SD = 8.61°) compared to the intended Smell Goal
alignment, confirming its consistent and safe alignment, facing the user for each scent
emission.
Figure 4.5: Combined movement paths of Spot (orange) and the user (blue) within the
workspace throughout the study
4.5.2 Smell Perception
Figure 4.6 visualizes the average time and standard deviation for participants to detect a
smell during each trial, grouped by Smell Intensity and Smell Type. An ART-ANOVA
revealed a significant main effect for Smell Intensity F (2, 99.57) = 9.94, p < .001, η2
p = .16
46
4.5. Results
as well as for the Smell Type F (2, 99.05) = 8.62, p < .001, η2
p = .14. However, no
significant interaction effect between the two factors was observed, F (4, 99.63) = 0.53,
p = .54, indicating that the effect of smell intensity on detection time was consistent
across all smell types, and vice versa. Therefore, the two factors did not influence each
other’s impact on detection performance. Post-hoc comparisons indicated that the high
intensity led to the fastest detection time (M = 7.67, SD = 2.80), followed by the
medium intensity (M = 8.90, SD = 3.05) and the low intensity (M = 10.47, SD = 4.40),
which resulted in the slowest detection overall. The difference between low and medium
intensity, however, was not statistically significant. Regarding scent types, Flower smells
were associated with the longest detection time (M = 10.16, SD = 3.39), compared to
Grass (M = 8.80, SD = 3.83 ), and Wood (M = 8.06, SD = 3.47), which were detected
faster. Beyond that, there was no statistically noteworthy detection time difference
between Wood and Grass.
Figure 4.6: A Boxplot visualizing the mean and standard deviation of smell detection
times, grouped by scent intensity levels (200, 400, 600) and scent type. Colors indicate
scent type: red for Flower, green for Grass, and blue for Wood.
Despite these measurable differences in detection speed, the subjective perception of scent
intensity did not show any significant variation across the multiple conditions. Users did
not report any effects of the different smell intensities. A Friedman ANOVA therefore
showed no main effect of intensity levels on perceived strength of users (χ2(2) = 2.22, p =
0.33), since the users rated all intensities similarly:
47
4. Evaluation and Results
• Low (200): M = 5.14, SD = 1.09
• Medium (400): M = 5.04, SD = 1.46
• High (600): M = 4.85, SD = 1.91
This suggests that participants were not consistently able to distinguish between intensities
on a subjective level, despite objective differences in detection times.
Participants were also asked to describe the smells as a free-text answer after the
experiment. The responses were sanitized for analysis by applying lowercase formatting,
removing plural/singular inconsistencies, correcting typos, and excluding generic filler
terms. Figure 4.7 shows the word clouds that have been generated for each of the three
scents. Across all experiments, the most commonly mentioned scent-related words were:
“flowers” (5 mentions), “wood” (4 mentions), “forest” (3 mentions), “grass” (3 mentions),
and “roses” (2 mentions).
Figure 4.7: Word clouds illustrating participant responses to the free-text question on
perceived smells for each scent type "Please describe the smells you experienced?". From
left to right: Flower, Grass, and Wood. Larger and more centered words indicate higher
mention frequency across participants.
4.5.3 Questionnaires
To evaluate the subjective experience of the participants, questionnaire data were collected
before, during, and after the experiment. The VRSQ was administered before and after the
three individual experimental blocks to assess any physical discomfort or symptoms such
as disorientation (χ2(2) = 6.25, p = 0.51) or oculomotor strain (χ2(2) = 1.00, p = 0.60).
A Friedman NOVA did not reveal any significant effect across the blocks (χ2(2) =
1.80, p = 0.40), suggesting that no substantial increase in symptoms was due to the
mobile scent device and VR.
48
4.5. Results
In addition, at the end of the experiment, users were asked to estimate how far away the
smell was displayed. Users reported that the distance between them and the Spot robot
with the smell was about 108±63 cm, whereas the actual distance was 150 cm.
Table 4.1 summarizes the average scores per block.
Table 4.1: Average VRSQ scores per block, reported separately for the disorientation,
oculomotor, and total subscales.
Disorientation Oculomotor Total
Block 1 5.23±11.45 5.14±10.26 6.19±10.55
Block 2 6.14±10.92 7.73±10.05 8.44±9.57
Block 3 6.94±11.42 8.33±11.23 7.73±10.43
Additionally, as a response to the question “How much did you trust the VR system’s
ability to simulate smells accurately?”, participants were asked to rate their confidence
in scent recognition on a 7-point Likert scale after each block. A score of 5.21±1.47 on
the 7-point Likert scale was evaluated. Similarly, to the question “How much did you
trust Spot Robot for walking safely in the workspace?”, participants reported a score of
5.23±1.21 on a 7-Likert scale. While slight trends suggested higher confidence ratings
for more intense scent conditions, no statistically significant difference across blocks was
found.
4.5.4 Post-Experiment Feedback and Additional Observations
Most participants described the VR experience as something pleasant, enjoyable, and
refreshing, particularly those who were using VR for the first time. The experience
was perceived as something new and left a positive impression overall (user 1). The
environment itself was considered visually pleasing, although slightly simplistic in its
design (users 2 and 3). The ambient soundscape, which included bird sounds, contributed
to immersion. Nevertheless, some users noted that they could still hear the robot, which
was slightly disruptive and occasionally broke their sense of immersion (users 2, 4, 6).
Several participants reported being able to perceive the scents that were deployed by the
smell sensor and that they generally found them pleasant, even if the specific smells were
not always easy to identify for the users (users 10, 11, 13). Suggestions for improvement
included using more recognizable or stronger scents, such as lavender or bakery aromas
(user 5). One user noted that one scent was much stronger than the other smells (user
11). One participant also questioned why blue smoke was used and reported difficulty
understanding the spatial relationship to the robot, which users could not locate spatially,
since it was not visible through the headset (user 6).
49

CHAPTER 5
Discussion
This chapter discusses the key findings of the evaluation in light of the initial research
questions and hypotheses. It reflects on how the system performed in terms of smell
detection, recognition, and user comfort, and explores potential reasons behind differences
observed across conditions. Design implications, technical insights, and directions for
future work are derived from both quantitative results and qualitative feedback.
5.1 Reflections on Study and Hypotheses
This study aimed to investigate the feasibility, reliability, and user perception of a mobile
scent delivery system embedded in a spatial VR environment. Four core hypotheses were
tested, each addressing a different aspect of system performance and human perception.
These were evaluated based on the statistical analyses outlined in Chapter section 4.5 and
are interpreted below in terms of both quantitative results and qualitative observations.
The first hypothesis (H1) proposed that higher smell intensities would lead to faster
detection. This was confirmed by a significant main effect of scent intensity on detection
time, as shown by the ART-ANOVA in Figure 4.6 and discussed in subsection 4.5.2.
Users reacted more quickly to stronger smells, consistent with prior findings that higher
molecular concentrations accelerate olfactory detection [LHK+24]. This result validates
the technical ability of the system to modulate perception through intensity control
reliably, and it highlights intensity as a crucial parameter for time-sensitive scent design.
The second hypothesis (H2) suggested that different smell types would result in different
detection times, assuming that certain smells are more immediately recognizable than
others. This was partially supported, as again can be seen in Figure 4.6. While detection
times did differ across scents, with “Wood” being detected more quickly than “Rose”
or “Wet Ground,” these differences were not always statistically significant. Qualitative
comments suggested possible causes, including that users found some smells artificial or
51
5. Discussion
difficult to describe, while others evoked stronger emotional associations. Interestingly,
negative associations seemed to increase detection speed, as evidenced by the “Wood”
smell. In contrast, pleasant or ambiguous smells like “Rose” were sometimes missed or
confused, resulting in a longer time to detect the smell. Since the “Wood” scent drew
more negative feedback and was described as uncomfortable, along with intense, spicy,
or woody, participants would not want to be exposed to it for extended periods and
therefore may have subconsciously prioritized its detection to avoid prolonged exposure.
These associations are further illustrated in Figure 4.7, which shows participant word
associations for each scent type, revealing common descriptors and emotional connotations.
This could also explain why it was the fastest to detect, as prior research noticed that a
negative or unpleasant olfactory experience could influence the perception and attention
of users [BB17]. Additionally, the speed at which users recognize a smell is known to
improve when the scent is familiar or easier to identify [PD20]. Due to “Wet Ground”
being more easily identified than the other two smells by the users, this could explain
the faster detection time. Nevertheless, this suggests that both emotional valence and
perceptual clarity affect olfactory responsiveness in VR. Finally, it is important to note
that the physical distance between the user and the Spot robot may also have influenced
perception. In this study, only one fixed distance, which was approximately 75cm, was
tested, leaving open questions about how proximity affects intensity perception and
response times.
The third hypothesis (H3) assumed that higher smell intensity would also lead to stronger
perceived smell strength. However, this was not supported. While the detection time
varied across intensity levels, with higher intensities leading to faster detections, which also
can be seen in Figure 4.6, participants rated the intensity levels as relatively similar across
all blocks. This discrepancy highlights that users did not consciously perceive higher-
intensity scents as stronger, despite their quicker reactions. One possible explanation
is that the differences were not significant enough to be consciously distinguished, or
that users adapted quickly to the smells. It may also reflect a ceiling effect, where
participants have already perceived the lowest intensity as sufficiently strong, making
further increases feel negligible. Additionally, it should be noted that intensity perception
was only assessed after each block and based on a limited number of trials. This suggests
that participants may have either learned or adjusted their perception over time, or
lacked sufficient repeated comparisons to notice differences, particularly when users
experienced different intensity orders, which likely led to divergent subjective learning
perceptions across participants. This finding suggests that increasing intensity alone
does not necessarily translate into a more substantial subjective experience, which is a
valuable insight for the design of future multi-sensory systems.
The fourth hypothesis (H4) proposed that mobile scent delivery would not increase VR
sickness symptoms. This was confirmed, as discussed in subsection 4.5.3, and as can be
seen in Table 4.1. VRSQ scores remained stable across all interaction blocks, and no
significant discomfort was reported. Participants reported feeling safe and unaffected,
with the experience not interfering with their overall comfort or daily well-being. This
52
5.1. Reflections on Study and Hypotheses
suggests that integrating a moving robot with synchronized scent emission can be done
safely in immersive environments. The system did not introduce any additional strain
beyond the base VR experience, even though users were walking and turning in a spatially
active scenario. This confirms the system’s suitability for extended real-time applications.
However, it must be acknowledged that users were exposed to the system for only a short
period. The absence of symptoms may therefore be partly due to the limited exposure
duration, meaning that systems involving longer user experiences still require further
investigation to understand potential long-term effects.
These outcomes underscore the complexity of multi-sensory interaction. While the system
reliably delivered scent and enhanced timing accuracy, not all perceptual outcomes,
namely the users perceived intensity ratings and their recognition accuracy, aligned with
the technical parameters. The divergence between objective detection and subjective eval-
uation, especially regarding perceived intensity, highlights the need for careful calibration
and individualized tuning in future designs. Such tuning should consider individual fac-
tors, including users olfactory sensitivity, emotional associations with specific scents, prior
scent familiarity, effects of VR sickness or motion sensitivity in larger user experiences,
and susceptibility to sensory or motion-related discomfort in VR environments.
A summary of the tested hypotheses and the observed outcomes can be seen in Table 5.1.
Table 5.1: Overview of tested hypotheses, their rationale, and observed outcomes
Hypothesis Description / Rationale Observed Outcome
H1 Higher smell delivery intensities
would lead to faster detection times,
due to higher concentrations of scent
molecules.
Confirmed. Detection times signif-
icantly decreased with higher inten-
sity across all scents.
H2 Different smell types would result
in varying detection times, based on
how easily they were recognized.
Partially supported. Wood was
identified faster than Rose or Wet
Ground, but results were inconsis-
tent due to individual variation.
H3 Higher smell intensities would lead
to stronger perceived smell strength.
Not supported. Despite faster de-
tection, subjective ratings of inten-
sity did not increase significantly.
H4 Olfactory stimuli would not signif-
icantly increase VR sickness symp-
toms.
Confirmed. VRSQ scores remained
stable; no notable discomfort was
reported.
Beyond the hypothesis-level findings, this study also provides broader insights into the
design and application of mobile olfactory systems in VR:
• The combination of trajectory recording, synchronized robot movement, and
condition-based scent emission proved technically robust and flexible, operating
under real-time constraints without causing sensory conflict.
53
5. Discussion
• The study demonstrated that smell can be introduced at precise moments during
spatial navigation, without confusing users or increasing simulator sickness, making
it suitable for dynamic interaction scenarios.
• Participant feedback suggested that emotional and semantic associations of smells
may have influenced recognition and detection. Some scents were identified more
quickly or confidently, possibly due to prior familiarity or stronger emotional
responses. While these impressions were not quantitatively confirmed, they point to
the potential value of personalization or adaptive scent selection in future systems.
Ultimately, this research demonstrates that mobile olfactory systems are not only tech-
nically viable but also perceived as meaningful and non-intrusive in immersive VR
settings.
Taken together, the findings highlight mobile scent delivery as a promising component
of immersive system design, particularly when carefully adapted to spatial and user
constraints.
5.2 Design Implications
The findings of this study offer several design implications for future implementations of
mobile scent delivery in immersive environments. First, the successful synchronization
between robot movement and scent emission underlines the potential of integrating
olfactory interaction into real-time VR experiences. The system’s ability to trigger smells
precisely and reliably during user navigation confirms that spatial scent cues can be
introduced without overwhelming or confusing users. This opens up design opportunities
for enhancing presence, guiding attention, or marking transitions in interactive scenarios.
Notably, the results also highlight the feasibility of large-scale mobile scent systems
that can dynamically accompany users across space. Unlike stationary emitters, a
mobile platform enables scent delivery that flexibly adapts to the users position and
movement, making it suitable for expansive, walkable environments such as room-scale VR
installations, location-based storytelling, exhibitions, or therapeutic scenarios. Since the
system does not require fixed wall-mounted emitters, permanent wiring, scent tubing, or
room-specific tracking setups, it can be easily deployed across various venues or temporary
installations. This portability not only broadens the range of possible applications but
also emphasizes the value of mobile scent systems as flexible and immersive tools, enabling
them to overcome the spatial constraints of static setups.
Moreover, the study highlights the importance of emotional and semantic associations
with specific smells. These associations affected recognition speed and subjective comfort,
even when the technical intensity was held constant. Designers of future systems should
therefore not only focus on the chemical properties of smell, such as molecule concentration
or intensity, but also consider the social, emotional, and contextual meanings users attach
54
5.3. Limitations
to certain scents. Adaptive scent selection or user-specific profiles could help accommodate
individual preferences and avoid discomfort from ambiguous or negatively perceived odors.
Another key implication concerns the autonomy and behavioral coordination of the
robot. To support user trust and avoid confusion, the robot’s movement and orientation
changes should occur smoothly and predictably, aligning with user expectations and the
narrative or the existing interaction flow. In the current system, Spot moves between
predefined Smell Zones triggered manually or via script. However, future iterations could
benefit from greater autonomy, such as gaze-based or intent-driven predictions that allow
the robot to anticipate user behavior and deliver scents just before the user reaches a
point of interaction. This would minimize delays and create a more seamless, immersive
experience.
Safety considerations are also essential when designing mobile olfactory systems. To
avoid collisions or interruptions in presence, future systems could dynamically restrict
the robot’s access to user-only zones or implement proximity-based logic to adjust
its path. This spatial coordination is especially relevant in multi-user settings, where
multiple users or robots may share the same environment. Supporting individualized
scent scheduling and positioning can help prevent cross-contamination and improve
user-specific interactions.
Finally, the modularity and mobility of the system open up promising application areas
such as location-based entertainment, education, storytelling, or rehabilitation. In
scenarios where workspace constraints limit physical locomotion, scent-triggered cues
could serve as natural, immersive redirection mechanisms. For instance, by introducing
or removing certain smells, users could be guided toward or away from specific areas
without breaking immersion. Similarly, a single mobile scent unit could be shared among
multiple rooms or users, enabling flexible deployment without the need for stationary
infrastructure. Taken together, these insights position mobile scent delivery systems
not just as technical tools but as powerful experiential components. Their ability to
influence perception, behavior, and memory through subtle yet context-sensitive cues
offers exciting opportunities for immersive system design, particularly when adapted to
the spatial logic and sensory expectations of different users and applications.
5.3 Limitations
As with any research prototype that integrates a diversity of tools, plugins, and tech-
nologies, including real-time robotics and multi-sensory interaction, several technical,
practical, and perceptual challenges emerged during development, deployment, and
testing. While the system functioned reliably and fulfilled its core objectives, several
important limitations are worth acknowledging when interpreting the results or planning
future improvements.
One central challenge concerned the inherent complexity of olfactory stimuli. Unlike
visual or auditory output, scent dispersion is influenced by airflow, the shape of the
55
5. Discussion
room, user movement, and the delay from the generator until the scent reaches the nose.
Although the intensity could be adjusted via parameter settings in Unity, precise timing
and localization of smell release were more challenging to control. Smells could linger in
the room longer than intended, occasionally interfering with follow-up interactions. This
issue required careful ventilation and timing between sessions, especially during repeated
trials, and limited the speed and replicability of test runs. However, this limitation is
common to most scent-based systems and not specific to the hardware used.
In total, the integration and control of the Olorama scent generator worked reliably in
most cases but required thoughtful coordination to ensure meaningful scent delivery.
While not a major technical hurdle, its physical placement and interaction timing still
required careful alignment with the user’s position and the robot’s movement to avoid
premature or delayed perception.
A further limitation was the mounting and tracking setup. The HTC Vive Tracker used
to measure the robot’s proximity to the user’s head-mounted display was not fixed in
a dedicated holder but instead taped onto the robot’s surface. Although this approach
worked throughout the study, it introduced potential for minor misalignment. The
position of the tracker, as well as the one of the Olorama scent device and the other tools
that were used, remained constant throughout the study, making any shift impossible.
Nevertheless, a firmer mount would improve positional accuracy and system robustness
in future setups.
From a software perspective, the evaluation was limited to a single 5m×5m forest scene.
While the environment was carefully designed to reflect a natural setting and worked well
for the purpose of the study, it did not allow testing more complex or varied spatial-smell
relationships. Dynamic transitions between scenes or zone-based storytelling were not
implemented. As a result, it might be difficult to generalize and compare the results
to other virtual worlds or applications. Additionally, the study focused on one specific
interaction type involving detection and recognition, and did not explore alternative
use cases or task types such as affective responses, exploration behavior, or narrative
immersion.
Another limitation relates to the use of a within-subjects design, in which each participant
experiences all experimental conditions. While this approach improves statistical power
and allows for controlling individual differences in factors such as smell sensitivity, VR
experience, or reaction speed, it may introduce potential carryover effects. Participants
might become more accustomed to the scent stimuli or anticipate the procedure, which
could affect detection or recognition performance. To mitigate such risks, future studies
should consider incorporating counterbalancing techniques or varying condition orders
more rigorously. Furthermore, the study did not include a control condition without
smell, which would have helped isolate the specific contribution of olfactory input to the
observed outcomes. Likewise, additional outcome variables such as presence, emotional
response, or perceived immersion were not assessed systematically and could provide
more nuanced insights in future evaluations. Future work could also compare olfactory
cues with other sensory modalities, such as visual markers or specialized audio, and
56
5.3. Limitations
examine behavioral aspects like users exploration patterns or time spent near scented
zones, further enriching the understanding of the role of scent in guiding attention and
shaping interaction.
Finally, although the robot–scent coordination was technically stable, the Spot itself
remained audible during operation. Despite participants wearing headphones, the robot’s
mechanical movements were still noticeable. This did not lead to discomfort or break
presence according to participants’ feedback, but it may affect immersion in future
scenarios where audio-olfactory synchronization is critical. Reducing system noise or
better masking it could be considered in future iterations. Future iterations may focus on
using rubber foot pads to reduce the impact noise and the vibration of the robot. In terms
of evaluation, this thesis included measurements of detection time, intensity perception,
and recognition accuracy. However, due to the limited sample size and the relatively
constrained environment, the results should be interpreted with care. Some perceptual
responses varied depending on user proximity, walking speed, or scent familiarity. Larger-
scale or more ecologically diverse studies would help verify how consistent these results
are across settings and user groups.
Summarized, these limitations do not undermine the technical success of the system but
rather highlight areas for future refinement. Most of them offer practical guidance for
scaling and adapting the system beyond this proof-of-concept prototype and into more
robust or applied scenarios.
57

CHAPTER 6
Conclusion and Future Work
This thesis presented a mobile olfactory display system designed for large-scale immersive
VR environments. The system integrates a Boston Dynamics Spot robot, an Olorama
scent generator, a Vive Tracker, and a custom Unity-based interface. Together, these
components enabled location-specific scent delivery by synchronizing robot navigation
with user interactions in VR. A Python-based server was developed to receive real-
time commands from Unity, sending movement instructions to the robot via TCP and
triggering scent emission via UDP. The Vive Tracker mounted on the robot allowed the
system to measure the distance to the users head-mounted display, providing a spatial
reference for scent activation. This approach eliminated the need for wearable scent
hardware and instead allowed for a more ambient and natural integration of smell into
the VR experience by using a mobile robot to deliver odors directly to the intended
location.
The system was tested through a controlled user study in a 5×5 meter virtual forest
environment. Participants encountered three olfactory interaction points, where the robot
stopped and released specific scents when the user approached. A visual cue indicated
the active interaction zone, and participants confirmed detection through a controller.
Results showed that smell delivery could be executed reliably and was generally well-
perceived across different users. Importantly, most participants reported no discomfort
or disruption to their sense of presence, and they were able to detect, recognize, and
respond to the scent cues, suggesting that the system succeeded in introducing olfactory
interaction without breaking immersion.
These results underline that robotic scent delivery is not only technically feasible but also
well-received by users in immersive, mobile VR settings. By enabling proximity-dependent
scent triggering without any wearable device, the system offers a new design direction
for multi-sensory interaction that supports user mobility and natural behavior. This
represents a promising alternative to fixed or body-worn olfactory displays, especially in
scenarios where space, motion, and autonomy are central.
59
6. Conclusion and Future Work
At the same time, several limitations and open questions suggest directions for future
development. While the system worked reliably, it was limited to a single environment and
a predefined interaction structure. Future studies could explore more complex or varied
scenes, incorporate multiple rooms or story-driven layouts, and test how scent perception
holds up under different pacing or environmental conditions. Another promising path is
improving the robot’s autonomy, such as by, for instance, using gaze-based prediction
or behavior modeling to trigger scent emission more dynamically and reduce manual
intervention. Similarly, adapting the robot’s path in real-time to avoid user-robot
collisions or to guide user flow more naturally could improve safety and fluidity, especially
in multi-user scenarios.
Additional technical refinements may further enhance the system. For instance, reducing
the mechanical noise of the robot through foot padding or better masking could support
stronger audio-olfactory synchronization. A more stable mounting solution for tracking
hardware would improve positional accuracy. From a perceptual standpoint, incorporating
user-specific scent preferences or adaptive delivery based on previous responses might
increase comfort and personalization. Lastly, evaluation metrics could be expanded
in future iterations to explore long-term memory effects, cross-modal associations, or
learning outcomes supported by scent-enhanced VR.
In summary, this work offers a proof-of-concept system that demonstrates the feasibil-
ity and potential of robotic scent delivery in VR. It contributes a technically robust,
perceptually subtle, and easily extensible platform for exploring large-scale olfactory
interaction. While still at an early stage, the approach lays important groundwork for
future applications in storytelling, training, education, rehabilitation, or entertainment,
especially where immersive, hands-free, and spatially aware scent delivery could add
value.
Ultimately, this study lays the foundation for future systems that deliver scent dynamically
and non-intrusively in large-scale virtual environments. It is paving the way for shared,
hands-free olfactory experiences that enhance immersion without relying on wearable
hardware, with future work moving towards more adaptive and immersive multi-sensory
experiences.
60
Overview of Generative AI Tools
Used
To support the writing process, I used a few generative AI tools in a helpful but limited
way. Perplexity AI helped me search for academic papers and get an overview of related
work. I used Grammarly to check my grammar and style, and ChatGPT to help with
rewording, editing, and making some parts easier to read, as well as to draw up mockup
images. I made all final content and decisions.
61

List of Figures
2.1 The human olfactory system. An olfactory cue enters the human smell system,
travels through the nasal complex and the mucus to the cilia, where the
olfactory receptors perceive the odor, start a chemical transformation, and
turn the odor into an understandable signal for the human brain [Bro10]. 8
2.2 Google Cardboard VR headset equipped with a wearable olfactory display.
The system integrates a Raspberry Pi, an odor repository, and a diffuser
controller for scent delivery during VR interaction [DPGMD+22]. . . . . . 12
2.3 Virtual Reality scene used in the user study. Participants walked toward a
flower field, where scented vapor, lavender or plain water, was emitted to
evaluate the impact of olfactory cues in immersive environments [DPGMD+22]. 13
2.4 BioEssence: A wearable olfactory display that can be clipped onto clothing or
worn as a necklace. The device delivers scents based on real-time physiological
signals such as the heart rate of wearers [AHD+18]. . . . . . . . . . . . . 14
2.5 The Essence Necklace. A contains the fragrance, while B displays the device
which manages the function [AM17]. . . . . . . . . . . . . . . . . . . . . . 14
2.6 A user wearing the inScent wearable olfactory display. The sequence shows
the process of receiving a message via scent (a–c), followed by reaching for
the phone in pleasant anticipation (d) [DHR17]. . . . . . . . . . . . . . . 15
2.7 The E-scent Coach is a wearable olfactory device designed to support correct
breathing during yoga sessions. Many practitioners struggle with proper
breathing technique. Using a webcam to analyze the users posture in real-
time, their pose is classified into either expansion or compression. Based on
that, the wearable releases or pauses scents to guide inhalation and exhalation,
promoting a stable and deep breathing rhythm [ZZAL25]. . . . . . . . . . 16
2.8 Four Scent Families and their Representative Scents for the Experiment of
Zhou et al. [ZZAL25] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9 oPhone DUO [DVO16] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.10 Scentee on an iPad [DVO16] . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.11 Aroma Shooter device by Aromajoin [DVO16] . . . . . . . . . . . . . . . . 19
2.12 Schematic representation of the "BBL-IS"-system [IBM+14] . . . . . . . . 19
3.1 Overview of the hardware components (top) and software system (bottom).
Red arrows indicate TCP-based communication (Unity to Spot via Python),
while blue arrows represent direct UDP commands (Unity to Olorama). . 23
63
3.2 The Boston Dynamics Spot robot equipped with the Olorama scent generator
and the HTC Vive Tracker. The scent device is mounted on the rear plate
for spatial alignment and synchronized scent emission based on the user’s
location. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Inspector-based interface in Unity for manual Spot control. . . . . . . . . 31
3.4 Top-down view of the virtual forest environment with predefined smell targets
and boundary trees. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Subtle blue smoke with glow effect marks the active scent zone in the VR
scene. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Visualization of synchronized real and virtual scent emission once the 75cm
threshold is crossed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.7 Inspector settings in Unity defining the smell intensity and timeout logic for
the evaluation sequence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Left - Battery stored in a belt pocket. Right - Battery is securely mounted
on the users clothing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Smell was released near designated small rocks in the environment, with the
Spot robot stopping and facing the user before scent emission. . . . . . . 41
4.3 Overview of the study procedure, including training, three scent intensity
blocks, 5-minute breaks between experimental blocks, final questionnaires,
and the debrief. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 Overview of the collected evaluation metrics, their descriptions, and respective
data sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.5 Combined movement paths of Spot (orange) and the user (blue) within the
workspace throughout the study . . . . . . . . . . . . . . . . . . . . . . . 46
4.6 A Boxplot visualizing the mean and standard deviation of smell detection
times, grouped by scent intensity levels (200, 400, 600) and scent type. Colors
indicate scent type: red for Flower, green for Grass, and blue for Wood. . 47
4.7 Word clouds illustrating participant responses to the free-text question on
perceived smells for each scent type "Please describe the smells you expe-
rienced?". From left to right: Flower, Grass, and Wood. Larger and more
centered words indicate higher mention frequency across participants. . . 48
64
List of Tables
3.1 Manual control functionalities available in Unity’s Spot interface. . . . . . 30
4.1 Average VRSQ scores per block, reported separately for the disorientation,
oculomotor, and total subscales. . . . . . . . . . . . . . . . . . . . . . . . 49
5.1 Overview of tested hypotheses, their rationale, and observed outcomes . . 53
65

Bibliography
[AAS20] Yewande Akinola, Oluwatoyin Agbonifo, and Oluwafemi Sarumi. Virtual
Reality as a tool for learning: The past, present and the prospect. Journal
of Applied Learning & Teaching, 3, August 2020.
[ABE+22] Nicholas S. Archer, Andrew Bluff, Andrew Eddy, Chreshall K. Nikhil, Nick
Hazell, Damian Frank, and Andrew Johnston. Odour enhances the sense
of presence in a virtual reality environment. PLOS ONE, 17(3):e0265039,
March 2022. Publisher: Public Library of Science.
[AHD+18] Judith Amores, Javier Hernandez, Artem Dementyev, Xiqing Wang, and
Pattie Maes. BioEssence: A Wearable Olfactory Display that Monitors
Cardio-respiratory Information to Support Mental Wellbeing. In 2018 40th
Annual International Conference of the IEEE Engineering in Medicine and
Biology Society (EMBC), pages 5131–5134, July 2018. ISSN: 1558-4615.
[AM17] Judith Amores and Pattie Maes. Essence: Olfactory Interfaces for Uncon-
scious Influence of Mood and Cognitive Performance. In Proceedings of
the 2017 CHI Conference on Human Factors in Computing Systems, CHI
’17, pages 28–34, New York, NY, USA, 2017. Association for Computing
Machinery.
[BB17] Oliver Baus and Stéphane Bouchard. Exposure to an unpleasant odour
increases the sense of Presence in virtual reality. Virtual Reality, 21(2):59–
74, June 2017. Publisher: Springer Science and Business Media LLC.
[BBEL71] B. Berglund, U. Berglund, T. Engen, and T. Lindvall. The effect of
adaptation on odor detection. Perception & Psychophysics, 9(5):435–438,
September 1971.
[BP21] Sanne Boesveldt and Valentina Parma. The importance of the olfactory
system in human well-being, through nutrition and social behavior. Cell
and Tissue Research, 383(1):559–567, January 2021.
[Bro10] Jennifer C. Brookes. Science is perception: what can our sense of
smell tell us about ourselves and the world around us? Philosophical
67
transactions. Series A, Mathematical, physical, and engineering sciences,
368(1924):3491–3502, August 2010.
[CF23] Sarah Cormiea and Jason Fischer. Odor discrimination is immune to the
effects of verbal labels. Scientific Reports, 13(1):1742, January 2023.
[DHR17] David Dobbelstein, Steffen Herrdum, and Enrico Rukzio. inScent: a
wearable olfactory display as an amplification for mobile notifications.
In Proceedings of the 2017 ACM International Symposium on Wearable
Computers, pages 130–137, Maui Hawaii, September 2017. ACM.
[DMP+21] Nicolò Dozio, Emanuela Maggioni, Dario Pittera, Alberto Gallace, and
Marianna Obrist. May I Smell Your Attention: Exploration of Smell
and Sound for Visuospatial Attention in Virtual Reality. Frontiers in
Psychology, 12, July 2021. Publisher: Frontiers.
[DPGMD+22] Marcelo De Paiva Guimarães, James Miranda Martins, Diego
Roberto Colombo Dias, Rita De Fátima Rodrigues Guimarães, and
Bruno Barberi Gnecco. An olfactory display for virtual reality glasses.
Multimedia Systems, 28(5):1573–1583, October 2022.
[DVO16] Dmitrijs Dmitrenko, Chi Thanh Vi, and Marianna Obrist. A Comparison
of Scent-Delivery Devices and Their Meaningful Use for In-Car Olfactory
Interaction. In Proceedings of the 8th International Conference on Au-
tomotive User Interfaces and Interactive Vehicular Applications, pages
23–26, Ann Arbor MI USA, October 2016. ACM.
[DWH+99] H.Q. Dinh, N. Walker, L.F. Hodges, Chang Song, and A. Kobayashi.
Evaluating the importance of multi-sensory input on memory and the
sense of presence in virtual environments. In Proceedings IEEE Virtual
Reality (Cat. No. 99CB36316), pages 222–228, March 1999. ISSN: 1087-
8270.
[EKHW21] Lisa A. Elkin, Matthew Kay, James J. Higgins, and Jacob O. Wobbrock.
An Aligned Rank Transform Procedure for Multifactor Contrast Tests. In
The 34th Annual ACM Symposium on User Interface Software and Tech-
nology, UIST ’21, pages 754–768, New York, NY, USA, 2021. Association
for Computing Machinery.
[Gui19] Erico Guizzo. By leaps and bounds: An exclusive look at how Boston dy-
namics is redefining robot agility. IEEE Spectrum, 56(12):34–39, December
2019. Conference Name: IEEE Spectrum.
[HHSR98] Hunter Hoffman, Ari Hollander, K. Schroder, and S. Rousseau. Physi-
cally touching and tasting virtual objects enhances the realism of virtual
experiences. Virtual Reality, 3:226–234, December 1998.
68
[HKC+23] Sun A Han, Jin Kook Kim, Do-Yeon Cho, Zara M. Patel, and Chae-Seo
Rhee. The Olfactory System: Basic Anatomy and Physiology for General
Otorhinolaryngologists. Clinical and Experimental Otorhinolaryngology,
16(4):308–316, November 2023.
[Hol23] Shirley Holt. Virtual reality, augmented reality and mixed reality: For
astronaut mental health; and space tourism, education and outreach. Acta
Astronautica, 203:436–446, February 2023.
[HSB04] Rachel Herz, Corrente Schankler, and Sophia Beland. Olfaction, Emotion
and Associative Learning: Effects on Motivated Behavior. Motivation
and Emotion, 28:363–383, December 2004.
[IBM+14] Matthieu Ischer, Naëm Baron, Christophe Mermoud, Isabelle Cayeux,
Christelle Porcherot, David Sander, and Sylvain Delplanque. How incor-
poration of scents could enhance immersive virtual experiences. Frontiers
in Psychology, 5, 2014.
[KGS+24] Elizabeth D. Kirby, Melissa J. Glenn, Noah J. Sandstrom, Christina L.
Williams, Elizabeth D. Kirby, Melissa J. Glenn, Noah J. Sandstrom,
and Christina L. Williams. 8.3 The Olfactory System - Introduction
to Behavioral Neuroscience | OpenStax, November 2024. Book Title:
Introduction to Behavioral Neuroscience Publisher: OpenStax.
[KPCC18] Hyun K. Kim, Jaehyun Park, Yeongcheol Choi, and Mungyeong Choe.
Virtual reality sickness questionnaire (VRSQ): Motion sickness measure-
ment index in a virtual reality environment. Applied Ergonomics, 69:66–73,
May 2018.
[LHK+24] Maxence Lalis, Matej Hladiš, Samar Abi Khalil, Loïc Briand, Sébastien
Fiorucci, and Jérémie Topin. M2OR: a database of olfactory recep-
tor–odorant pairs for understanding the molecular mechanisms of olfaction.
Nucleic Acids Research, 52(D1):D1370–D1379, January 2024.
[Lin98] Roger W. A. Linden. The Scientific Basis of Eating: Taste and Smell, Sali-
vation, Mastication and Swallowing, and Their Dysfunctions. Karger Med-
ical and Scientific Publishers, 1998. Google-Books-ID: VItrjPNGdVMC.
[LJW20] Danyang Li, Jiafeng Jia, and Xiaochun Wang. Unpleasant Food Odors
Modulate the Processing of Facial Expressions: An Event-Related Poten-
tial Study. Frontiers in Neuroscience, 14, 2020.
[LYZ+23] Yiming Liu, Chun Ki Yiu, Zhao Zhao, Wooyoung Park, Rui Shi, Xingcan
Huang, Yuyang Zeng, Kuan Wang, Tsz Hung Wong, Shengxin Jia, Jingkun
Zhou, Zhan Gao, Ling Zhao, Kuanming Yao, Jian Li, Chuanlu Sha, Yuyu
Gao, Guangyao Zhao, Ya Huang, Dengfeng Li, Qinglei Guo, Yuhang Li,
and Xinge Yu. Soft, miniaturized, wireless olfactory interface for virtual
69
reality. Nature Communications, 14(1):2297, May 2023. Publisher: Nature
Publishing Group.
[MSR06] Alan Mackay-Sim and Jean-Pierre Royet. Structure and Function of
the Olfactory System. In Christos Pantelis, David Castle, and War-
rick J. Brewer, editors, Olfaction and the Brain, pages 3–27. Cambridge
University Press, Cambridge, 2006.
[NM07] Takamichi Nakamoto and Hai Pham Dinh Minh. Improvement of olfactory
display using solenoid valves. In 2007 IEEE Virtual Reality Conference,
pages 179–186, Charlotte, NC, USA, 2007. IEEE.
[PBH17] J. Paluchová, J. Berčík, and E. Horská. 2. The sense of smell. In Angel A.
Carbonell-Barrachina, editor, Sensory and aroma marketing, pages 27–60.
Brill | Wageningen Academic, May 2017.
[PD20] Susan Persky and Alexander P. Dolwick. Olfactory Perception and Pres-
ence in a Virtual Reality Food Environment. Frontiers in Virtual Reality,
1, September 2020. Publisher: Frontiers.
[SBB16] Berenice Serrano, Rosa M. Baños, and Cristina Botella. Virtual reality
and stimulation of touch and smell for inducing relaxation: A randomized
controlled trial. Computers in Human Behavior, 55:1–8, February 2016.
[SH21] Rachel S. Herz. Olfactory Virtual Reality: A New Frontier in the Treat-
ment and Prevention of Posttraumatic Stress Disorder. Brain Sciences,
11(8):1070, August 2021.
[SLW11] Kai Song, Qi Liu, and Qi Wang. Olfaction and Hearing Based Mobile
Robot Navigation for Odor/Sound Source Search. Sensors, 11(2):2129–
2154, February 2011. Number: 2 Publisher: Molecular Diversity Preser-
vation International.
[SMB+13] C. Sezille, B. Messaoudi, A. Bertrand, P. Joussain, M. Thévenet, and
M. Bensafi. A portable experimental apparatus for human olfactory fMRI
experiments. Journal of Neuroscience Methods, 218(1):29–38, August
2013.
[SSdM92] D. G. Smith, L. Standing, and A. de Man. Verbal memory elicited by
ambient odor. Perceptual and Motor Skills, 74(2):339–343, April 1992.
[TLD+07] R. Tortell, D. Luigi, A. Dozois, Sean Bouchard, Jacquelyn Morie, and
D. Ilan. The effects of scent and game play experience on memory of a
virtual environment. Virtual Reality, 11:61–68, March 2007.
[VBMRS14] Carlos Velasco, Diana Balboa, Fernando Marmolejo-Ramos, and Charles
Spence. Crossmodal effect of music and odor pleasantness on olfactory
70
quality perception. Frontiers in Psychology, 5, November 2014. Publisher:
Frontiers.
[WFGH11] Jacob O. Wobbrock, Leah Findlater, Darren Gergle, and James J. Higgins.
The aligned rank transform for nonparametric factorial analyses using
only anova procedures. In Proceedings of the SIGCHI Conference on
Human Factors in Computing Systems, CHI ’11, pages 143–146, New
York, NY, USA, 2011. Association for Computing Machinery.
[YKN+04] Y. Yanagida, S. Kawato, H. Noma, A. Tomono, and N. Tesutani. Projec-
tion based olfactory display with nose tracking. In IEEE Virtual Reality
2004, pages 43–50, March 2004. ISSN: 1087-8270.
[YYT+06] T. Yamada, S. Yokoyama, T. Tanikawa, K. Hirota, and M. Hirose. Wear-
able Olfactory Display: Using Odor in Outdoor Environment. In IEEE
Virtual Reality Conference (VR 2006), pages 199–206, March 2006. ISSN:
2375-5334.
[ZZAL25] Leijing Zhou, Yiqing Zhang, Xin An, and Junxian Li. E-scent Coach:
A Wearable Olfactory System to Guide Deep Breathing Synchronized
with Yoga Postures. In Proceedings of the Nineteenth International Con-
ference on Tangible, Embedded, and Embodied Interaction, pages 1–14,
Bordeaux/Talence France, March 2025. ACM.
71

Appendix
73
Demographic Survey
User ID (filled by experimenter):
1. Age
How old are you?
2. Gender
What is your gender?
• ⃝ Female
• ⃝ Male
• ⃝ Prefer not to say
• ⃝ Other:
3. Do you have allergies or sensitivities to scents?
• ⃝ Yes
• ⃝ No
• ⃝ Not sure
4. Previous Experience in Virtual Reality (VR)
What is your level of experience with VR?
• ⃝ None (today is the first time)
• ⃝ Tried once
• ⃝ Novice
• ⃝ Occasional user
• ⃝ Monthly user (at least once per month)
• ⃝ Weekly user (at least once per week)
• ⃝ Daily user
74
5. Sense of Smell
How would you rate your sense of smell?
• ⃝ Very Poor
• ⃝ Poor
• ⃝ Average
• ⃝ Good
• ⃝ Excellent
6. Previous Experience with Smell Detection Technologies
Have you ever used or worked with smell detection technologies before? (e.g., digital scent
technology, olfactory displays)
• ⃝ Yes
• ⃝ No
• ⃝ Not sure
7. Identifying different smells
Please rate your agreement with the following statement: I am confident in being able to
correctly identify specific smells.
1 2 3 4 5 6 7
Strongly Disagree Strongly Agree
75
Post Block Questionnaire
User ID (filled by experimenter):
Block (filled by experimenter): 1 2 3
Symptoms (rated on a 4-point scale):
Symptom Not at all Mild Moderate Severe
General Discomfort ⃝ ⃝ ⃝ ⃝
Fatigue ⃝ ⃝ ⃝ ⃝
Eyestrain ⃝ ⃝ ⃝ ⃝
Difficulty focusing (visually)* ⃝ ⃝ ⃝ ⃝
Headache ⃝ ⃝ ⃝ ⃝
Fullness of the head* ⃝ ⃝ ⃝ ⃝
Dizziness with eyes closed ⃝ ⃝ ⃝ ⃝
Vertigo* ⃝ ⃝ ⃝ ⃝
Participants rated their current symptoms after each interaction block.
Term Definitions:
• * Difficulty focusing (visually) – Challenges in maintaining clear and sharp
visual focus, difficulty in adjusting focus, or the inability to concentrate visually.
• * Fullness of the head – A sensation of pressure or heaviness in the head.
• * Vertigo – Experienced as loss of orientation with respect to vertical upright, e.g.,
the perception of spinning or rotational movement, even when stationary.
How intense were the smells during the trials?
1 2 3 4 5 6 7
Not intense at all Very Intense
76
Post-Experiment Questionnaire
User ID (filled by experimenter):
1. Please describe the smells you experienced during the study:
2. How far were you (in cm) from the smell source during the experiment?
3. I was confident in detecting smells in VR environments correctly.
1 2 3 4 5 6 7
Strongly Disagree Strongly Agree
4. How much did you trust the VR system’s ability to simulate smells
accurately?
1 2 3 4 5 6 7
Did not trust at all Trust very much
5. Additional comments to the experiment:
77