Adaptive Control of Cardio-respiratory Training in a Virtual Reality

Hiking Simulation: A Feasibility Study

Rodrigo Lima

1,3 a

, Muhammad Asif

1,3 b

, Honorato Sousa

2 c

and Sergi Berm

udez I Badia

1,3 d

Faculdade de Ci

encias Exatas e da Engenharia, Universidade da Madeira, Funchal, Portugal

Faculdade de Ci

encias do Desporto e Educac¸

ao F

ısica, Universidade de Coimbra, Coimbra, Portugal

NOVA Laboratory for Computer Science and Informatics, Caparica, Portugal

Keywords:

Hiking Simulator, Virtual Reality, Cardiorespiratory Fitness, Biocybernetic Loops, Heart Rate, User Experience, Control,

Adaptation.

Abstract:

Adaptive Virtual Reality applications are a novel way to enhance and promote higher levels of physical activity and car-

diorespiratory ﬁtness, leading to a healthier lifestyle and avoid cardiovascular diseases. In this study, we developed a system

using a virtual hiking simulator, the Levadas from Madeira Island, that aims to increase the compliance of recommendations

levels of exertion by implementing a closed-loop adaptation according to the heart rate. The system was tested with a sample

of twenty healthy young adults on a repeated measures design, comparing the adaptive VR, a non-adaptive VR version of

the software, and a non-VR version. Perceived exertion, presence, usability and intrinsic motivation were assessed. The

results from the study reveal that the adaptive control according to the heart rate promoted approximately 20% more time of

exertion in the recommended target heart rate zone, while perceiving lower levels of exertion by the participants, compared

to the non-adaptive condition.

1 INTRODUCTION

Physical inactivity has been recognized as the fourth

leading cause of death worldwide (Lee et al., 2012).

A sedentary lifestyle is considered a sole risk fac-

tor for cardiovascular diseases, which account for ap-

proximately 30% of global mortality. Thus promot-

ing an increase in physical activity in people of all

ages will help in reducing the risk of cardiovascular

diseases (Hoffmann et al., 2015). A novel way to in-

crease physical activity is to use exergames to exer-

cise and promote health and well-being (Mu

noz et al.,

2018). Exergames are digital games that require the

usage of the whole-body to control a game, increas-

ing the physical activity level, potentially improving

the physical ﬁtness components, such as endurance,

strength, balance and ﬂexibility (Oh and Yang, 2010).

According to the American College of Sports and

Medicine (ACSM), exergaming can be described as

a healthy and beneﬁcial form of exercising by en-

gaging and challenging the participants to play (Dean

et al., 1998). Several studies have shown that ex-

ergames can enhance enjoyment and intrinsic moti-

vation compared to traditional exercises and are efﬁ-

https://orcid.org/0000-0002-4030-9526

https://orcid.org/0000-0002-9091-9381

https://orcid.org/0000-0002-7434-643X

https://orcid.org/0000-0003-4452-0414

cient to promote physical and mental health (Pluchino

et al., 2012; Rosenberg et al., 2010).

Although not strictly a computer game, the system

used in this paper is a Virtual Reality (VR) simula-

tion of a pleasant real-life experience (Ahmad, 2021).

In recent years, VR technologies have made much

progress and many VR systems have been introduced.

The usage of VR technology is trending because it

provides a high level of immersion - the extent to

which the VR system delivers sensations from the

real world to the virtual world (Bailenson et al., 2008;

Stasie

nko and Sarzy

nska-Długosz, 2016). In particu-

lar, systems such as CAVE (Cave Automatic Virtual

Environment) have been reported to be effective to

immerse and engage participants during VR experi-

ences (Gonc¸alves et al., 2021).

VR-based applications are being used for athletic

training, ﬁtness training, and high-intensity interval

training, as the full-body interaction and high im-

mersive experience are the main advantages of using

VR technology for cardiorespiratory training (Shep-

herd et al., 2018). Recent studies have established

that VR applications can increase enjoyment, motiva-

tion and engagement, contrary to traditional exercises,

such as cycling and running. Garcia et al. (Garcia

et al., 2016) investigated the feasibility and efﬁcacy

of Kinect-based stepping exergame, and reported im-

provements of participants in stepping, standing bal-

Lima, R., Asif, M., Sousa, H. and Bermúdez i Badia, S.

Adaptive Control of Cardio-respiratory Training in a Virtual Reality Hiking Simulation: A Feasibility Study.

DOI: 10.5220/0011004400003123

In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 4: BIOSIGNALS, pages 91-99

ISBN: 978-989-758-552-4; ISSN: 2184-4305

ance, gait speed, and mobility. Christos Ioannou et

al. (Ioannou et al., 2019), introduced the concept of

Virtual Performance Augmentation (VPA) of running

and jumping in-place. They reported that VPA can

induce moderate to high physical activity levels, in-

creasing intrinsic motivation and general physical ac-

tivity motivation, and perceived competence and ﬂow.

Huang et al. assessed the effect of immersion on

the perceived exertion by using a stationary bicycle-

based training system (Huang et al., 2008). They

compared perceived exertion in PC-Desktop screen,

a projector, Head Mounted Displays (HMD), and a

non-VR system. A signiﬁcantly lower perceived ex-

ertion was found in the HMD and projector condi-

tions when compared to non-VR. No signiﬁcant dif-

ference was found between HMD and projector con-

ditions. The outcomes of this study were consistent

with (Mestre et al., 2011), who compared immersive

VR with virtual feedback and no 3D computer gener-

ated feedback setups. They reported lower perceived

exertion and higher excitement in the immersive VR

environment.

Although these VR applications have shown the

potential to improve the cardiorespiratory training,

training intensity is often lower than what is expected

to be the ﬁtness recommendations (Dean et al., 1998).

Thus, to maintain the player’s training intensity, one

possibility is to monitor the user and adapt the VR ap-

plicationin response. This adaptation allows changing

the training load required to achieve the desired levels

of exertion (Hoffmann et al., 2015). Here, a closed-

loop control approach was implemented using the

biocybernetic-loop-engine (BLEngine) (Mu

noz et al.,

2017) to monitor the heart rate (HR) and adapt a vir-

tual hike experience to control the intensity of the ex-

ercise performed by the participants. We aimed to ad-

dress the following research questions:

• RQ1: Can an adaptive system successfully manip-

ulate training intensity?

• RQ2: Can an adaptive system effectively keep

participants in the desired target HR zone?

• RQ3: How does an adaptive system compare to

its non-adaptive counterpart?

• RQ4: What is the impact of VR feedback on per-

ceived exertion levels and motivation?

2 METHODOLOGY AND

MATERIALS

2.1 Participants

The participants were recruited from a convenience

sample of volunteer subjects, all university students

and workers. Twenty-two healthy adults (12 females,

10 males) volunteered to participate in this study. Two

participants were excluded from the study: a male

participant due to a technical error and a female par-

ticipant decided to withdraw due to virtual reality

sickness. The sample considered for the analysis was

composed of the remaining 20 participants, with an

average age of 29 years old. The demographical in-

formation of the participants is described in Table 1.

Table 1: Sample statistics.

(n=20) Mean STD Min Max

Age 29,25 5,03 23,00 44,00

Height (cm) 168,95 9,06 159,00 191,00

Weight (kg) 64,35 14,16 47,00 102,00

BMI (kg/m

) 22,26 2,60 17,91 27,96

2.2 Experimental Setup

2.2.1 Hardware

The virtual environment was designed to work in a

CAVE with walls of 2.2 meters width and 2.8 me-

ters of height, using the KAVE software developed

by (Gonc¸alves and Badia, 2018). The display con-

sists of the front and lateral projections, as well as

a ﬂoor projection, thus it requires a computer with a

graphic card capable of displaying 4 screens simulta-

neously. A Kinect V2 (Microsoft, Redmond, USA)

body tracking sensor was used to track the full-body

of the participants (3D position of the 25 joints’ skele-

ton) (Gonc¸alves et al., 2021). A schematic version of

our CAVE setup is shown in Figure

A photoplethysmography (PPG) sensor was

used to measure the heart rate (HR) at rest during

the pre-assessment procedure (section 2.3.2) with

a wearable device, the Biosignalsplux (PLUX

- Wireless Biosignals, Lisboa). Finally the

HR at rest was computed with the Opensignals

software (PLUX-Wireless Biosignals, Lisboa)

(https://biosignalsplux.com/products/software/

opensignals.html).

To measure the HR of the participants during the

experiment, the HR chest band Polar H10 (Polar Elec-

tro Oy, Kempele, Finland) was placed on the par-

ticipants. The Polar H10 was paired with the Acti-

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

Figure 1: Schematic of our CAVE setup.

Graph’s WGT3X-BT accelerometer (Actigraph Cor-

poration, Pensacola, FL, USA) to measure physical

activity during the experiment. The metrics for phys-

ical activity were computed using the ActiLife6 soft-

ware (version 6.13.4, ActiGraph, Cary, NC, USA) to

process the accelerometer data.

2.2.2 Software

The environment simulated a Levada hiking track and

was developed by Ahmad et al (Ahmad, 2021). The

Levada track included computer-generated 3D ob-

jects, such as trees, mountains, tunnels and irriga-

tion canals, created with Unity3D Engine and Blender

software (Blender Foundation, Amsterdam, Nether-

lands) (Ahmad, 2021).

Then, an adaptation of the procedure in

(Gonc¸alves et al., 2016) was implemented in

the VR Levada environment, to determine a target

height at which the knee of the participant had to be

raised during stepping-in-place to progress in the VR

Levada hiking track. The initial height was calculated

by the Kinect, as the middle point between the hip

and the knee, so that it adjusts to people of different

height. The height of the hip and the knee were

set, respectively, as the boundaries of maximum and

minimum required heights during the adaptation.

2.3 Experimental Procedure

2.3.1 Pre-assessment

All participants performed a pre-assessment session

on a different day, previous to the experiment. In this

session, the HR at rest of each participant was mea-

sured to calculate their experimental target HR. The

participants also performed the 3-min YMCA Step

Test (Golding, 2000) to assess their physical ﬁtness.

Figure 2: Example of our VR Levada hiking track. The

red dots represent the knees of the participant. The white

line represents the target height at which the knee had to be

raised.

2.3.2 HR at Rest

Participants were asked to sit and relax in a chair

placed in a quiet room for 5 minutes, without mov-

ing or speaking. Then the HR was computed using a

PPG sensor with the biosignalsplux wearable device.

The PPG sensor was placed on the index ﬁnger of the

left hand. The HR at rest of the participants was cal-

culated as the average HR of the 5 minutes using the

Opensignals software, as mentioned in section 2.2.1.

2.3.3 Target HR

The target HR for the experiment was calculated us-

ing the Karvonen formula (Karvonen and Vuorimaa,

1988). After computing the heart rate at rest, the max-

imum HR (HR

max

) was calculated using Equation 1.

Then the heart rate reserve (HRR), which is the differ-

ence between the maximum heart rate and the heart

rate at rest, was calculated using Equation 2. Finally,

the target HR was calculated using Equation 3, with a

target exercise intensity of moderate to vigorous (ap-

proximately 60% of the HRR), according to (Dean

et al., 1998).

max

= 220 − Age (1)

HRR = HR

max

− HR

rest

(2)

TargetHR = (%intensity ∗ HRR) + HR

rest

(3)

2.3.4 3-min YMCA Step Test

The 3-min YMCA Step Test (Golding, 2000) was

used to assess the cardiorespiratory ﬁtness of the par-

ticipants. To perform this test, a 30 cm step, a digi-

tal chronometer and metronome were used. First the

Adaptive Control of Cardio-respiratory Training in a Virtual Reality Hiking Simulation: A Feasibility Study

test procedure was explained to the participants by

demonstrating the cadence stepping. The metronome

was set to 96 beats per minute, with 4 clicks represent-

ing one step cycle: 1st beat - ﬁrst foot up, 2nd beat -

second foot up, 3rd beat - ﬁrst foot down, 4th beat -

second foot down. The duration of this test was 3 min-

utes. After completing the test, participants immedi-

ately sat down and the average HR for 1 min was as-

sessed using the same sensor as in section 2.3.2. The

classiﬁcation of the cardiorespiratory ﬁtness of our

participants was performed by comparing the scoring

of the 1-min post exercise average HR, with the age

adjusted standard ratings for this test. Our classiﬁca-

tion is shown in Table 2.

Table 2: 3-min YMCA Step Test Classiﬁcation.

Males

(n=9)

Females

(n=11)

Excellent 2 4

Good 1 0

Above Average 0 0

Average 2 1

Below Average 2 3

Poor 0 3

Very Poor 2 0

2.3.5 Protocol

This study is divided into two components: the study

of the adaptive control of HR and the study of the user

experience in the virtual environment. Participants

were provided with informed consent previously to

the pre-assessment session (Section 2.3.1).

After the pre-assessment session, a within-

subjects experimental design was used, in which par-

ticipants performed the three following conditions, on

consecutive days with an approximate time interval

of 24 hours: Adaptive VR Levada experiment (Ex-

perimental condition for VR adaptive control of HR

and user experience), Non-Adaptive VR Levada ex-

periment (Control condition for adaptive control of

HR) and Adaptive Non-VR experiment (Control con-

dition for the user experience).

For each experiment participants were asked to

wear the Polar HR chest band and the ActiGraph ac-

celerometer placed on the hip. The experiment con-

sisted of stepping in the same place at the pace of 125

beats per minute set by a metronome, with 2 clicks

representing one step cycle, as the participants were

stopped in the same place and had to raise their knees

up to a target height, to progress in the virtual hiking

track.

Each experiment had a total duration of 10 min: a

2-min warm-up to drive the participants to the target

HR zone, a 7-min training in the target HR zone, and

a 1-min cool-down. This timeline is shown in Figure

At the end of each experiment, participants an-

swered a sickness/dizziness questionnaire, classiﬁed

the perceived exertion using the Rated Perceived Ex-

ertion Scale (RPE scale) (Borg, 1998), and for the test

and control conditions of the user experience, partic-

ipants answered the Witmer-Singer Presence Ques-

tionnaire (WSPQ) (McCall et al., 2004), the Slater-

Usoh-Steed Questionnaire (Slater et al., 1995) and the

Intrinsic Motivation Inventory (IMI) (Sheehan et al.,

2017; Mcauley et al., 1989).

0 1 2 3 4 5 6 7 8 9 10 min

Warm-up period

Training in target HR zone

Cool-down period

Figure 3: Timeline of each experiment.

2.4 Adaptive Control of HR

The main goal of the adaptive control of HR was to

drive participants to reach the target heart rate zone,

moderate to vigorous intensity (57% - 63%), as men-

tioned in section 2.3.3, and keep them inside that zone

during the whole training phase.

The cardiorespiratory ﬁtness adaptation based on

the HR was performed using an updated version of

the Biocybernetic Loop Engine (BLEngine) (Mu

noz

et al., 2017). The BLEngine received the real-time

HR data from the Polar H10 chest band using UDP

communication.

Then a proportional-integral-derivative controller

(PID controller) was implemented to adapt the height

to which the participants had to raise their knees while

stepping in place. Then, target height was adapted

every 5 seconds, according to the instantaneous HR

(HR

5sec

). For the warm-up phase, a linear regression

was calculated to drive the participant’s HR, between

the initial HR of the participant and the target HR, to

gradually increase the intensity of the exercise, so that

after 2 minutes of exercise, the participant reached the

intended HR zone.

The PID controller followed equation 4, with

= 0.03 is the proportional constant and K

= 5

is the derivative constant. The PID parameters were

tuned by performing several pilot tests prior to this

study, until the desired performance of the controller

was achieved.

These HR adaptive rules implemented on the

BLEngine software, are shown in Figure 4.

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

PID = K

∗



target

− HR

5sec



∗



Error

current

− Error

∆t



(4)

Figure 4: Adaptive HR rules used on the BLEngine soft-

ware.

2.5 Instruments

2.5.1 Physical Exertion Metrics

The data acquisition for the physiological signals was

performed with a custom-made log ﬁle implemented

on the BLEngine, to record all the HR related sig-

nals, and then all the HR metrics were computed us-

ing Python. For the accelerometer signals, the ActiL-

ife6 software provided all the required metrics.

Concerning the HR related metrics, the following

metrics were computed: Average HR, Percentage of

Time in Target HR Zone, considering 100% as being

the 10 min condition, and the root mean square error

(RMSE) between the HR and the target HR.

In terms of the accelerometer metrics, the METS,

vector magnitude, MVPA, and Percentage in Seden-

tary and Light exercise, were computed using the Ac-

tilife6 software.

Finally, a digital version of the OMNI Rated Per-

ceived Exertion (RPE) scale (Borg, 1998) was used to

assess the perception of exertion from the participants

after both conditions, in a 0 to 10 scale (0 - Extremely

Easy, 10 - Extremely Hard).

2.5.2 User Experience

To assess the user experience, the sickness and dizzi-

ness experienced by the participants during the VR

experience, a short brief questionnaire was answered

with a 5-point Likert Scale (1-none, 5-A lot). Also,

the Witmer-Singer’s Presence Questionnaire (WSPQ)

was used to assess the sense of presence. It includes

24 items addressing Involvement, Immersion, Visual

Fidelity, Interface Quality, and Sound, rated on a 7-

point Likert scale. Consistent with other studies,

items 20-22 related to sound were excluded. Items

23-24 related to haptics were not applicable for this

study (McCall et al., 2004).

The Intrinsic Motivation Inventory (IMI) was

used to assess intrinsic motivation. It is a multi-

dimensional measurement questionnaire, which is

comprised of seven sub-scales and used for several

studies including exercising and sports. The question-

naire contains the following sub-scales on a 7-point

Likert scale: Interest/Enjoyment, which is considered

to be the main self-report measure for this question-

naire (7 items), and Pressure/Tension, which is con-

sidered to be a negative predictor of intrinsic motiva-

tion (5 items) (Sheehan et al., 2017; Mcauley et al.,

1989).

The System Usability Scale (SUS), created by

(Brooke, 1995), was implemented to assess the ap-

plication’s usability. SUS comprises ten items and

allows a quick evaluation of the usability of a wide

variety of products and services, including hardware

and software.

2.6 Statistical Analysis

For the physiological signals statistical analysis, the

Kolmogorov-Smirnov normality test was used to as-

sess the normality of the data. Since the data was not

normally distributed, non-parametric statistical tests

were used. The Wilcoxon matched-pair signed ranks

test was used to compare conditions.

Regarding the statistical analysis for the question-

naires, all the data from the questionnaires are of or-

dinal nature (Likert Scale), thus non-parametric tests

were used to assess the signiﬁcance of the results. The

Wilcoxon matched-pair signed ranks test was also

performed for the questionnaires. All the statistical

analysis was performed in SPSS Statistics version 26.

3 RESULTS

3.1 Adaptive Control of HR

Training in a speciﬁc heart rate zone has beneﬁts and

helps improve cadiorespiratory performance, accord-

ing to the ACSM guidelines (Dean et al., 1998). To

measure the accuracy of the algorithm implemented

to drive and maintain the participants in the target

heart zone, the metrics related to the HR mentioned

in section 2.5.1 were analyzed.

When performing the statistical comparison,

we found no signiﬁcant difference in the average

HR, between the Non-Adaptive VR (Mdn=142.02,

Range=70.82) and the Adaptive VR condition

(Mdn=138.90, Range=48.12).

Adaptive Control of Cardio-respiratory Training in a Virtual Reality Hiking Simulation: A Feasibility Study

In terms of the difference between the HR

and the target HR, the RMSE revealed lower val-

ues in the Adaptive VR condition (Mdn=20.57,

Range=18.46) compared to the Non-Adaptive VR con-

dition (Mdn=25.08, Range=27.14). This result was

signiﬁcantly different, T=173,p-value<0.05,r=0.57

(Figure 5).

Finally, the Adaptive VR condition had sig-

niﬁcantly higher percentage of time (T=32.00,p-

value<0.01, r=0.61) in the target HR zone

(Mdn=26.83, Range=67.83) compared to

the Non-Adaptive VR condition (Mdn=7.83,

Range=49.67)(Figure 6).

Figure 5: Boxplot of the RMSE for the Non-Adaptive VR

Levada and Adaptive VR Levada conditions.

Figure 6: Boxplot of the Time in target HR zone in per-

centage for the Non-Adaptive VR Levada and Adaptive VR

Levada conditions.

Figure 7 depicts the time evolution and relation-

ship between participants’ HR and their target HR

over the 10 minutes of the experiment. HR data in

Figure 7a (black line), shows that the Adaptive VR

Levada condition spends more time inside the target

HR zone (red band) compared to the Non-Adaptive

VR Levada. Also, the variability of the data (black

shadow) seems smaller than that of the control.

(a) Adaptive VR Levada.

(b) Non-Adaptive VR Levada

Figure 7: Average HR of all participants throughout the

whole experiment, for the Adaptive and Non-Adaptive VR

conditions. Black line - Average HR, Black Shadow - Av-

erage Standard Deviation, Red Band - Target HR zone, Yel-

low Band - Warm-up Phase, Green Band - Training Phase,

Blue Band - Cool-down Phase.

3.2 Physical Exertion

Concerning the accelerometer metrics mentioned

in section 2.5.1 it is possible to verify that in

the Adaptive VR Levada condition the METS val-

ues were lower (Mdn=1.02, Range=1.59) than the

Non-Adaptive VR Levada condition (Mdn=1.05,

Range=2.33). In terms of the Percentage of

time in Sedentary and Light exercise, in the

Non-Adaptive VR Levada condition the participants

spent more time in Sedentary exercise (Mdn=34.17,

Range=85.83) compared to the Adaptive VR Lev-

ada (Mdn=29.00, Range=95.50). Consequently,

the time spent in Light exercise was higher for

the Adaptive VR Levada condition (Mdn=70.50,

Range=95.00) than in the Non-Adaptive VR Lev-

ada (Mdn=59.67, Range=84.33). For the MVPA,

the Non-Adaptive VR Levada condition values were

higher (Mdn=0.05, Range=4.27) than the Adaptive

VR Levada values (Mdn=0, Range=4.92). Finally,

the magnitude vector values for the Adaptive VR

Levada were higher (Mdn=27168.5, Range=36849.8)

than the Non-Adaptive VR Levada condition values

(Mdn=20891.1, Range=33329.1). Despite these re-

sults, no signiﬁcant difference was found for all the

metrics computed, between the Non-Adaptive VR

Levada and Adaptive VR Levada conditions.

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

The results for the perceived exertion (RPE

Scale) reported by the participants, revealed that the

Adaptive VR Levada condition showed lower val-

ues of perceived exertion (Mdn=4.00, Range=8.00)

compared to the Non-Adaptive VR Levada condi-

tion (Mdn=5.00, Range=6.00) (Fig.8). No signiﬁ-

cant difference was found between the Non-Adaptive

VR Levada and Adaptive VR Levada conditions (p-

value=0.089).

Figure 8: Boxplot of the Rated Perceived Exertion (RPE)

Scale for the Non-Adaptive VR Levada and Adaptive VR

Levada conditions.

3.3 User Experience

Participants reported a higher level of dizziness

(Mdn=2.00, Range=3.00) on the Adaptive VR Levada

condition compared to the Adaptive Non-VR condi-

tion (Mdn=1.00, Range=3.00). For the sickness ques-

tion, the result reported were the same on both con-

ditions (Mdn=1.00, Range=2.00). No signiﬁcant dif-

ference was found between the Adaptive Non-VR and

Adaptive VR Levada condition, regarding the sickness

and dizziness short 5-point questionnaire.

Presence was measured with the WSPQ. The to-

tal mean score for the sum of all sub-scales was

M=93.1(16.1), which indicates a presence level of

70%, similar to the result reported by (Ahmad, 2021),

and higher than the result reported by (Gonc¸alves

et al., 2021) (Table 3). The mean rating score of in-

volvement suggested that the user was engaged with

the experiment while the immersion score shows that

user perceived the environment realistically. The vi-

sual ﬁdelity and interface quality scores represented

the clarity, perception depth and user-friendly appli-

cation. The mean rating score of the sound also indi-

cated the realistic sound coming from the application

environment.

Regarding the results obtained between the Adap-

tive Non-VR and the Adaptive VR Levada conditions

for the IMI questionnaire (Table 4), no signiﬁcant dif-

ferences were found for either Pressure/Tension or In-

terest/Enjoyment.

Table 3: Mean Rating Score for each sub-scale of the

WSPQ. The scale ranges from 1 to 7 in a Likert Scale.

Sub-Scale Mean Rating Score (SD)

Involvement 4.78 (0.95)

Immersion 5.23 (0.79)

Visual Fidelity 4.3 (1.6)

Interface Quality 5.3 (1.1)

Sound 4.7 (1.5)

Table 4: Intrisic Motivation Inventory Results.

Sub-Scale

VR Non-VR

p-value

Mean (SD) Mean (SD)

Interest/Enjoyment 4.62 (1.37) 4.42 (1.19) 0.37

Pressure/Tension 2.65 (0.99) 2.71 (1.26) 0.87

The usability of the application was assessed with

the SUS (M=78.1 (3.2)). In this test, a score of 68 is

the threshold that indicates that user satisfaction level

is above average (Brooke, 1995). Hence, our results

showed the excellent usability of the simulation appli-

cation in terms of comfort and ease of use, and were

similar to the result reported by (Ahmad, 2021).

4 DISCUSSION

Most studies related to exercising with adaptive con-

trol of HR are based on cycle ergometers or ex-

ergames (Mu

noz et al., 2018; Hoffmann et al., 2015;

Kiryu et al., 2001). This study assessed the feasi-

bility of performing an adaptive control of cardio-

respiratory training based on the participant’s HR

while performing a stepping activity in a virtual Re-

ality hiking simulation. The results obtained from

our study showed that the algorithm implemented was

able to drive the participants to reach the target heart

zone within, approximately, the ﬁrst 2-3 minutes of

the exercise, answering the ﬁrst question (RQ1) re-

garding the manipulation of the training intensity us-

ing an adaptive system.

Regarding RQ2 and RQ3, the participants in the

Adaptive VR Levada exerted more than 20% of the

total duration of the experiment (10-min), in the tar-

get heart rate zone, compared to the Non-Adaptive VR

Levada condition, while having an median RMSE of

20 beats per minute between the heart rate and the tar-

get heart rate.

Both of these results agree with the results ob-

tained by (Mu

noz et al., 2018), in which the time that

the participants exerted in the target HR zone was

40% higher in the adaptive condition with a RMSE

of 15 beats per minute, considering the entire experi-

ment of 20 minutes.

Adaptive Control of Cardio-respiratory Training in a Virtual Reality Hiking Simulation: A Feasibility Study

Regarding the RPE scale, although no signiﬁcant

difference was found between the Non-Adaptive VR

Levada and Adaptive VR Levada conditions, partic-

ipants reported lower values of perceived exertion in

the Adaptive VR Levada condition. This indicates that

even though participants were training in the target

HR zone for more time, their perception of effort was

smaller, addressing the impact of VR feedback on per-

ceived exertion levels (RQ4). This could suggest that

lower levels of fatigue may come from training in a

more controlled HR regime. These results are also in

agreement with the results obtained by (Mu

noz et al.,

2018).

In terms of sense of presence, most of the pa-

pers in the literature report treadmills or cycling ergo-

meters, which makes it difﬁcult to compare with our

study. In our experiment, the impact of VR on physi-

cal exertion and intrinsic motivation during the virtual

hiking simulation was investigated in a stepping-in-

place based application. From the results obtained, it

is possible to observe that this simulation of a virtual

hiking activity generated a high sense of presence, ap-

proximately 70%, with the sub-scale Immersion hav-

ing the highest mean rating score of M = 5.23 com-

pared to the other sub-scales. Regarding the intrin-

sic motivation, participants reported a higher value

for intrinsic motivation in the Adaptive VR of ap-

proximately 66% compared to 63% in the Adaptive

Non-VR Levada for the sub-scale Interest/Enjoyment,

and a lower value of Pressure/Tension of 38% for

Adaptive VR Levada and 39% for the Adaptive Non-

VR Levada. Even though no signiﬁcant effect was

found for intrinsic motivation, the results found for

the Interest/Enjoyment are in agreement with other

studies. Buchner et al. (Buchner and Zumbach,

2018) also reported higher values for the sub-scale

Interest/Enjoyment in augmented reality than a non-

Augmented Reality application.

Finally, regarding the levels of dizziness and sick-

ness while using the VR system, despite having a par-

ticipant that was dropped from the study due to sick-

ness problems, the median value reported for the sick-

ness was of 1, and for the dizziness was of 2, in a

5-point scale, thus our system is not prone to induce

virtual sickness or dizziness to the participants.

5 LIMITATIONS

Although the main goal for this study was achieved

with success, there are some limitations implicit with

this study. The effect of VR is speciﬁc for the de-

sign of our VR Levada experiment and for our par-

ticular CAVE, so we do not have information on what

can happen on other VR delivery technologies or sim-

ulations, or the effect of adding gamiﬁcation to the

task. This study targeted a training intensity of 60%

with a mandatory pace of stepping of 125 beats per

minute set by a metronome. In order to generalize the

application of this system to other situations, higher

or lower training intensities, with a different pace set

by the metronome should be tested. Also, the single

variable adapted during the entire study was the tar-

get height at which the participants had to raise their

knees. A variable could be added to adjust the pace

set by the metronome to combine the adaptation with

the target height. Finally, the acquisition of HR was

performed on a consumer-grade device, the Polar H10

chest band.

6 CONCLUSION

This study aimed to create a stepping-based VR ap-

plication simulating a hiking track, the Levadas, that

could adapt to the physiological signals of the partic-

ipant to provide adequate levels of exercise intensity.

Our data indicates that the adaptation rules created on

the closed-loop, according to the participant’s HR, us-

ing BLEngine, could drive the participant to the de-

sired target heart rate zone. Thus, successfully ad-

justing the intensity of training within the target heart

rate zone of optimal effectiveness. This adaptation in-

creased the time in which the participants were in the

target heart rate zone by 20% compared to the Non-

Adaptive VR Levada condition. In addition, partici-

pants perceived lower levels of exertion in the adap-

tive condition. In conclusion, we highlight the po-

tential of personalized and adaptive VR applications

to improve cardiorespiratory ﬁtness, engagement and

motivation of the participants.

ACKNOWLEDGEMENTS

This work was co-ﬁnanced by ”Madeira 14-20” under

the project MITIExcell - Excel

encia internacional de

IDT&I nas TIC (M1420-01-0145-FEDER-000002)

and supported by the NOVA Laboratory of Computer

Science and Informatics (UID/CEC/04516/2020) and

by the ARDITI - Ag

encia Regional para o Desen-

volvimento da Investigac¸

ao, Tecnologia e Inovac¸

ao.

REFERENCES

Ahmad, M. A. (2021). Efﬁcacy of augmented reality-based

virtual hiking in cardiorespiratory endurance : A pilot

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

study.

Bailenson, J., Patel, K., Nielsen, A., Bajscy, R., Jung, S. H.,

and Kurillo, G. (2008). The effect of interactivity on

learning physical actions in virtual reality. Media Psy-

chology, 11(3):354–376.

Borg, G. (1998). Borg’s Perceived Exertion And Pain

Scales.

Brooke, J. (1995). Sus: A quick and dirty usability scale.

Usability Eval. Ind., 189.

Buchner, J. and Zumbach, J. (2018). Promoting intrinsic

motivation with a mobile augmented reality learning

environment.

Dean, A., Sciences, H., Kingston, R. I., Island, R., Edith,

P. C., Heart, B. F., Henry, C., Hospital, F., Kingston,

R. I., Island, R., and Coordinator, P. (1998). Advanced

ﬁtness assessment & exercise prescription, volume 35.

Garcia, J., Schoene, D., Lord, S., Delbaere, K., Valenzuela,

T., and Felix Navarro, K. (2016). A bespoke kinect

stepping exergame for improving physical and cogni-

tive function in older people: A pilot study. Games for

health journal, 5.

Golding (2000). YMCA Fitness Testing and Assessment

Manual. YMCA of the USA.

Gonc¸alves, A. and Badia, S. B. I. (2018). KAVE: Building

kinect based CAVE automatic virtual environments,

methods for surround-screen projection management,

motion parallax and full-body interaction support.

Proceedings of the ACM on Human-Computer Inter-

action, 2(EICS).

Gonc¸alves, A., Cameir

ao, M., Berm

udez i Badia, S., and

Gouveia, E. (2016). Automating senior ﬁtness testing

through gesture detection with depth sensors. pages 6

.–6 .

Gonc¸alves, A., Montoya, M. F., Llorens, R., and Berm

udez

i Badia, S. (2021). A virtual reality bus ride as an

ecologically valid assessment of balance: a feasibility

study. Virtual Reality, (0123456789).

Hoffmann, K., Sportwiss, D., Hardy, S., Wiemeyer, J., and

obel, S. (2015). Personalized Adaptive Control of

Training Load in Cardio-Exergames - A Feasibility

Study. Games for Health Journal, 4(6):470–479.

Huang, S.-F., Tsai, P.-Y., Sung, W.-H., Lin, C.-Y., and

Chuang, T.-Y. (2008). The comparisons of heart

rate variability and perceived exertion during simu-

lated cycling with various viewing devices. Presence,

17:575–583.

Ioannou, C., Archard, P., O’Neill, E., and Lutteroth, C.

(2019). Virtual performance augmentation in an im-

mersive jump & run exergame. pages 1–15.

Karvonen, J. and Vuorimaa, T. (1988). Heart Rate and Exer-

cise Intensity During Sports Activities: Practical Ap-

plication. Sports Medicine: An International Journal

of Applied Medicine and Science in Sport and Exer-

cise, 5(5):303–311.

Kiryu, T., Sasaki, I., Shibai, K., and Tanaka, K. (2001).

Providing appropriate exercise levels for the elderly.

IEEE Engineering in Medicine and Biology Maga-

zine, 20(6):116–124.

Lee, I.-M., Shiroma, E. J., Lobelo, F., Puska, P., Blair, S. N.,

and Katzmarzyk, P. T. (2012). Effect of physical in-

activity on major non-communicable diseases world-

wide: an analysis of burden of disease and life ex-

pectancy. The Lancet, 380(9838):219–229.

Mcauley, E., Duncan, T., and Tammen, V. (1989). Psycho-

metric properties of the intrinsic motivation inventory

in a competitive sport setting: A conﬁrmatory factor

analysis. Research quarterly for exercise and sport,

60:48–58.

McCall, R., O’Neil, S., and Carroll, F. (2004). Measur-

ing presence in virtual environments. Conference on

Human Factors in Computing Systems - Proceedings,

pages 783–784.

Mestre, D., Ewald, M., and Maiano, C. (2011). Virtual real-

ity and exercise: Behavioral and psychological effects

of visual feedback. Studies in health technology and

informatics, 167:122–7.

noz, J. E., Cameir

ao, M., Berm

udez i Badia, S., and Ru-

bio Gouveia, E. (2018). Closing the loop in exergam-

ing - Health beneﬁts of biocybernetic adaptation in se-

nior adults. CHI PLAY 2018 - Proceedings of the 2018

Annual Symposium on Computer-Human Interaction

in Play, pages 329–339.

noz, J. E., Gouveia, E. R., Cameir

ao, M. S., and Ba-

dia, S. B. I. (2017). The biocybernetic loop engine:

An integrated tool for creating physiologically adap-

tive videogames. PhyCS 2017 - Proceedings of the

4th International Conference on Physiological Com-

puting Systems, (PhyCS):45–54.

Oh, Y. and Yang, S. (2010). Deﬁning exergames & ex-

ergaming.

Pluchino, A., Lee, S. Y., Asfour, S., Roos, B., and Signo-

rile, J. (2012). Pilot study comparing changes in pos-

tural control after training using a video game balance

board program and 2 standard activity-based balance

intervention programs. Archives of physical medicine

and rehabilitation, 93:1138–46.

Rosenberg, D., Depp, C., Vahia, I., Reichstadt, J., Palmer,

B., Kerr, J., Norman, G., and Jeste, D. (2010). Ex-

ergames for subsyndromal depression in older adults:

A pilot study of a novel intervention. The Ameri-

can journal of geriatric psychiatry : ofﬁcial journal

of the American Association for Geriatric Psychiatry,

18:221–6.

Sheehan, R., Herring, M., and Campbell, M. (2017). Moti-

vation measures in sport: A critical review and biblio-

metric analysis. Frontiers in Psychology, 8.

Shepherd, J., Carter, L., Pepping, G.-J., and Potter, L. E.

(2018). Towards an operational framework for design-

ing training based sports virtual reality performance

simulators. Proceedings, 2:214.

Slater, M., Steed, A., and Usoh, M. (1995). The Vir-

tual Treadmill: A Naturalistic Metaphor for Naviga-

tion in Immersive Virtual Environments. (December

2013):135–148.

Stasie

nko, A. and Sarzy

nska-Długosz, I. (2016). Virtual

Reality in Neurorehabilitation. Postepy Rehabilitacji,

30(4):67–75.

Adaptive Control of Cardio-respiratory Training in a Virtual Reality Hiking Simulation: A Feasibility Study