Virtual Reality (2024) 28:12

https://doi.org/10.1007/s10055-023-00904-x

ORIGINAL ARTICLE

An augmented reality application and experiment for understanding

and learning spatial transformation matrices
Zohreh Shaghaghian1 · Heather Burte2 · Dezhen Song3 · Wei Yan1

Received: 30 December 2022 / Accepted: 16 October 2023 / Published online: 6 January 2024
© The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024

Abstract
Understanding spatial transformations and their mathematical representations is essential in computer-aided design, computer
graphics, robotics, etc. This research has developed and tested an augmented reality (AR) application (BRICKxAR/T) to
enhance students’ learning of spatial transformation matrices. BRICKxAR/T leverages AR features, including information
augmentation, physical–virtual object interplay, and embodied learning, to create a novel and effective visualization experi-
ence for learning. In this paper, we evaluated the BRICKxAR/T as a learning intervention using LEGO models for physical
and virtual manipulatives in an experiment. The experiment compared AR (N = 29) vs. non-AR (N = 30) learning workshops
with pre- and post-tests on Purdue Visualization of Rotations Test and math questions to assess students’ learning gains. All
participants math scores significantly improved with the AR workshop tending to show greater improvements. The post-
workshop survey showed students were inclined to think BRICKxAR/T an interesting and useful application, and they spent
more time learning in AR than non-AR.

Keywords Augmented reality · Spatial transformations · Transformation matrices · Visualization · Embodied learning

1 Introduction discipline, assisting in higher-level reasoning activities (Hol-

lebrands 2003). Literature has clearly acknowledged the dif-
Spatial and mathematical thinking are closely allied (Wheat- ficulty of learning and teaching spatial transformations and
ley 1990; Garderen 2006; Rellensmann et al. 2017). Also, the associated math representations such as matrices (Holle-
physical model interactions have shown a significant impact brands 2003; Gülkılıka et al. 2015). Despite the development
on spatial visualization and reduction of extraneous cogni- of Computer Assisted Learning (CAL) in geometry, spatial
tive load in learning spatial and geometry-based problems transformations, and related mathematics, many students
(Kim and Lou Maher 2008; Preece et al. 2013; Hoe et al. still face challenges in solving geometric problems and rely
2019). Understanding tightly coupled spatial transforma- on a trial-and-error process (Garmendia et al. 2007). Fur-
tions and mathematical concepts significantly contributes thermore, it is very challenging to maintain students’ atten-
to STEM (Science, Technology, Engineering, and Mathe- tion in CAD workshops, and students often drop the course
matics) learning in fields of geometric modeling, computer before it ends (Martin et al. 2010).
graphics, computer-aided design (CAD), computer vision, Augmented Reality (AR) as a mediator tool with the abil-
robotics, video games, quantum mechanics, and more, and ity to superimpose digital content and information over a
helps students consider mathematics as an interconnected physical environment supports a context to integrate embod-
ied learning and virtual augmentation of abstract informa-
* Wei Yan tion. AR immersive environment may allow students from
[email protected] different fields of STEM to learn the mathematical logics
and abstractions pertained to 3D modeling functions through
1
Department of Architecture, Texas A&M University, spatiotemporal experiments while removing the extraneous
College Station, TX, USA
cognitive load of using keyboards and mice.
2
Department of Psychological and Brain Sciences, Texas The intellectual merit of this research lies in the devel-
A&M University, College Station, TX, USA
oped innovative learning environment that enables physi-
3
Department of Computer Science and Engineering, Texas cal and virtual interplays to engage students in embodied
A&M University, College Station, TX, USA

Vol.:(0123456789)
12 Page 2 of 18 Virtual Reality (2024) 28:12

learning through interacting with physical models. In this engage in physical interaction, it boosts their creativity in
study, the inherent capability of AR is adopted to realize generating design ideas (Kim and Lou Maher 2008; Viswa-
novel spatial strategies for understanding mathematical con- nathan and Linsey 2012), reduces unnecessary cognitive
cepts through visualizing the concepts and graphical infor- burden during the creative design process (Chandrasekera
mation that are registered (aligned) and synchronized with and Yoon 2015), enhances their spatial skills in compre-
physical motions in the physical environment. Computer hending size relationships among geometries (Sun et al.
graphics (e.g., arrows, tags, highlighting, etc.) matched with 2014), and fosters improved collaboration and communica-
the user’s view could effectively draw students’ attention tion during group work (Seichter 2007; Kendon 1994; Figen
(Tang et al. 2003) and improve their mental imagery by visu- Gül and Halıcı 2015).
alizing difficult invisible concepts. The developed AR app—
BRICKxAR/T—realizes the synchronized visualization of 2.1 AR for learning geometry and mathematics
mathematical concepts with the physical actions. This assists
students in perceiving transformation scenarios (motions and Several AR applications have been developed for learning
the corresponding mathematical functions) within a single descriptive geometry and mathematics (Martin et al. 2010;
comprehensive application. Dünser et al. 2006; Hohenwarter 2018; Ravé et al. 2016),
As a broader impact, this research has the potential to which demonstrate the positive impact of AR intervention in
support many students who struggle with spatial and math geometry perception. Most of the apps have used AR appli-
reasoning, especially those from underrepresented groups cations as visualization tools, displaying 3D geometries and
in STEM. Gaining a fundamental understanding of spatial different representations (for example, images of unfolded
transformations can uniquely contribute to students’ learning geometry) in a spatial environment to help students’ spatial
and development of spatial reasoning and allied mathemati- visualization skills (Martin et al. 2010; Ravé et al. 2016)–(Le
cal skills, leading to improved STEM coursework, STEM and Kim 2017). The mathematical representations in Con-
retention, and degree attainment, and thus supports students’ struct 3D app are limited to certain graphics only (Kaufmann
future development of expertise and career success across and Schmalstieg 2003), and the mathematical representa-
STEM disciplines. tions in the AR app developed by Cahyono et al. (2018)
This research intends to address the mentioned challenges are static text with no real-time interactions (Cahyono et al.
by answering these research questions: 2018). GeoGebra AR is one of the recent AR applications
in learning geometry and mathematics (Hohenwarter 2018).
1. Can an AR-powered learning system improve students’ However, GeoGebra AR does not provide any major physi-
spatial visualization of rotations and their mathematical cal interaction or interplay of physical and digital environ-
representations, specifically, matrices? ments in the learning process. Also, the mathematical sec-
2. Does the AR system help students focus on the targeted tion in GeoGebra AR requires advanced pre-knowledge of
learning subject and decrease the task load compared to mathematical equations to generate forms in AR; otherwise,
a non-AR system? it will be a trial-and-error process. Hence, it may be a chal-
3. Can the AR system stimulate students’ motivation and lenging application for a self-learning process.
engagement in learning mathematical concepts of geo- Although the potential of AR in learning mathematics and
metric transformations? improving spatial skills has been studied in STEM learning,
the technology has not yet been explored deeply in learning
geometric transformations and their mathematics.
2 Literature
2.2 AR vs. CAD and virtual reality (VR)
Studies report that the geometry courses through the US
school grades trend more toward recognizing and naming While CAD and VR may also assist in learning similar spa-
new geometric objects rather than a deep level of analysis tial and math concepts, compared to AR, they have major
(Fuys et al. 1988). Hence, many educators in STEM pro- limitations. For example, the 2D images of 3D objects in
fessions strive to improve academic competency in these CAD models do not match the real-world perspective view
fields. Researchers believe geometric transformations help and require a mental model alignment as well as appropriate
students think of essential mathematical concepts through use of a keyboard and mouse to navigate through the scene.
new approaches with higher-level reasoning (Hollebrands VR can specifically be helpful for simulating experiences
2003). About 20% of students experience difficulty in prob- that do not exist in the real world (e.g., a fictional environ-
lems dealing with spatial skills, such as manipulating figures ment) or are not easily accessible (e.g., walking through a
in space and analyzing complex shapes (Garmendia et al. space) (Nemec et al. 2017; Anifowose et al. 2022). How-
2007; Sorby 2009). Literature indicates that when students ever, the VR experiments detach a user from the physical
Virtual Reality (2024) 28:12 Page 3 of 18 12

environment by replacing it with a complete virtual sur- AR 3D environment, representing the mapping operations.
rounding (Azuma 1997). Modeling a whole new environ- Mathematical functions are synchronized with motion and
ment to simulate a real experience for VR application could mapping. In the current experiment, half of the participants
be time-consuming and computationally expensive (Velev interacted with the AR version of the app and the other half
and Zlateva 2017). The interactions in a VR environment, interacted with a non-AR version.
realized through external hardware controllers, are neither
immediate nor natural, which may impose extraneous cogni- 2.4 AR App
tive loads. Also, some VR experiments report health issues
such as motion sickness and injuries due to wearing VR When the user starts the workshop, the app registers two vir-
headsets such as Oculus (Nemec et al. 2017; Kruijff et al. tual models, one visible and one hidden, through the image
2017). In addition, there are health concerns with sharing a marker and aligned with the physical model. When the phys-
VR headset, especially for children during a pandemic such ical model is moved by the user, the visible virtual model
as COVID-19. These cognitive loads, usability, and health stays in the original location, representing the pre-image of
and safety concerns associated with VR may be reduced the transformation, and the hidden virtual model with a vis-
with AR. For example, the AR-enabled tablets can gener- ible coordinate system follows the physical model, together
ally be more affordable than the VR headsets for personal representing the image of the transformation (Fig. 1). The
use without the need of sharing. Furthermore, AR supports physical model's transformations (translation and rotation)
automatic perspective view alignment with the user’s rela- are thus visualized by the registered and synchronized dis-
tive position, embodied learning, and physical interactions, tance line and rotation angle graphics as well as the trans-
which facilitate tangible manipulatives in 3D space, and con- formation matrix functions (top).
tribute to improving mental 3D visualization (Preece et al. The BRICKxAR/T app has been developed on the AR-
2013), revising mental model misconceptions (Viswanathan enabled iOS mobile device (iPads). In BRICKxAR/T, the
and Linsey 2012), enhancing spatial cognition and design AR registration (alignment between virtual and physical
creativity (Kim and Lou Maher 2008; Chandrasekera and models) is highly accurate, and the occlusions between the
Yoon 2015), improving idea generation (Viswanathan and physical and virtual models are created to realistically reveal
Linsey 2012; Kim and Lou Maher 2008), and encouraging the spatial relationships among the models (Yan 2021).
epistemic action and memory retrieval (Bujak et al. 2013). Hand occlusion is also activated to the AR environment to
augment the virtual renderings with correct depth perception
2.3 Application prototype with respect to the physical objects (Fig. 2).
Figure 3 illustrates that the visible virtual model can
The BRICKxAR/T app (Shaghaghian et al. 2021, 2022) is also be moved through the user interface menus. The 3D
an AR prototype for learning spatial transformations and Cartesian coordinate system (#1) is the World Coordinate
matrix algebra. An equivalent non-AR version of the app is System (WCS) in the AR scene. The WCS gets instantiated
also developed with the same visualization functions without in the beginning but never updated during the play. The 3D
AR features. The app has been developed based on the pro- Cartesian coordinate system (#2) represents the local coor-
gressive learning method introduced in literature for learning dinate system of the physical LEGO model (and an aligned
spatial transformations (Fife et al. 2019). In this technique, hidden virtual model). This coordinate system gets updated
students will learn spatial transformations in three levels of seamlessly by the AR camera tracking the attached image
motion, mapping and function (Fife et al. 2019) as follows: marker and follows the movement of the physical model
applied by the user. The wireframe virtual model (#3) can
• Motion: AR supports physical actions and embodied be transformed (translated, rotated, and scaled) through its
learning. parameter controls on the AR screen by the user.
• Mapping: AR supports visualizing the transformation Figure 4 depicts the relations among the AR camera and
mapping process through demonstration of the image the three components: physical model, virtual model, and
and re-image of transformations. the WCS.
• Functions: AR supports augmentation of mathematical The prototype is an RTS (Rotation, Translation, and
functions of transformation matrices. Scale) puzzle game that helps students understand math-
ematical notions of spatial transformations along with the
We have leveraged the AR technology to realize this mathematical components of transformations, such as vari-
process for spatiotemporal experiments. Using AR inter- ables, parameters, and functions (Shaghaghian et al. 2021).
vention, physical interaction (motion), along with physical The mathematical functions matched with the actual motions
and virtual models’ interplay (mapping), is supported in in real-time assist students to perceive the geometric rea-
the application. Graphical illustrations are displayed in the sonings behind transformation matrices in an integrated
12 Page 4 of 18 Virtual Reality (2024) 28:12

Fig. 1  Seen through an AR-enabled iPad, virtual model (pre-image illustrated by synchronized distance line and rotation angle graphics
of transformations) and physical model (image of transformations) as well as transformation matrix functions (top)
in the AR scene, with the transformations (translation and rotation)

translation and rotation matrices (first row matrices in

Fig. 3). The distance line and the rotation arc and angle show
the graphical representation of the mapping function where
the numbers (distance and angle) are matched with the ele-
ments in transformation matrices (first row).
Students can also play with the virtual model to numeri-
cally compose the transformation matrices (second row) by
directly interacting with the corresponding function param-
eters and associated menu sliders. Translating the virtual
model in all three axes can be done at the same time. The
associated sliders display when touching x, y, and z param-
eters of the point vector (representing a point on the geom-
etry) as shown in red, green, and blue, respectively (Fig. 3).
To rotate or scale the virtual model, students can tap
the Rotation/Scale control. Then, the associated trans-
Fig. 2  A zoom-in view of the distance line and rotation arc and formation matrix and the corresponding slider to change
angle between the physical and virtual models. Occlusion is applied the selected parameter (i.e., rotation angle or scale factor)
between the real hand and virtual objects appears on the screen. The parametric changes applied
on the sliders will be reflected in the 4 × 4 transforma-
spatial scenario. Students can play with the physical LEGO tion matrix of the virtual model as well as the algebraic
model, translate and rotate it and observe the corresponding equations (results of the matrix multiplication), as shown
Virtual Reality (2024) 28:12 Page 5 of 18 12

Fig. 3  AR scene: physical and virtual models and their corresponding transformation matrices for both models (graphics and notations in yellow
and cyan colors are added on top of the AR scene manually for better explanation) (Color figure online)

Fig. 4  Relations among the AR

camera and the three compo-
nents: physical model, virtual
model, and the WCS in the AR
scene
12 Page 6 of 18 Virtual Reality (2024) 28:12

Fig. 5  Rotating the virtual model around y-axis (left) and scale in x-, y-, and z-axes (right)

in Fig. 5. To keep the equations of the matrix multiplica- technology automatically using camera and motion sensors
tion simple for the students to understand, rotation around on the AR device.
other axes gets zeroed out each time the student chooses
a new axis. 2.5 Non‑AR App
Finally, students can learn to compose the transformation
matrix of the virtual model so that it matches the transfor- The features and functions of the equivalent non-AR version
mations of the physical model at its current location and of the app are similar to BRICKxAR/T except that physical
orientation (Fig. 6). This step intends to intuitively describe model interaction is disabled in this version. Due to the lack
the concept behind AR registration (physical and virtual of a physical model interaction in a non-AR environment,
model alignment) as one of the applications of transfor- only two virtual models get rendered in the application,
mation matrices, which is normally done through the AR including a shaded LEGO model and a wireframe model

Fig. 6  Practicing AR registration through playing with function parameters

Virtual Reality (2024) 28:12 Page 7 of 18 12

(i.e., a digital twin). The shaded model remains fixed with a visualization of the rendered graphics, such as distance lines,
Cartesian coordinate system representing the WCS, and the rotation arcs, and corresponding notations. Figure 7 shows a
wireframe model can be transformed using function param- scene from the non-AR versions of workshop1 with manual
eters. In the AR environment of BRICKxAR/T, students can navigations applied.
freely move around the table and interact with the physical
model; however, the non-AR version is similar to a typical
desktop app which only allows common screen navigation 2.6 Research objectives and hypotheses
as follows:
The primary objective of this research is to address the
• Panning: sliding on the screen with two fingers challenges associated with teaching and learning geomet-
• Zooming: pinching two fingers together or apart to zoom- ric transformations. The aim is to leverage AR as a tool to
out or zoom-in, respectively enhance students’ spatial and math skills in understanding
• Orbiting: sliding on the screen with one finger toward left the concepts of transformation matrices. The ultimate goal is
and right to orbit the screen clockwise or counterclock- to provide students with a comprehensive understanding of
wise. these concepts, helping them bridge the gap between math-
ematical abstractions and real-world applications.
Because the physical–virtual model interaction is not pos- Through the development and evaluation of the
sible in the non-AR environment, this version is restricted to BRICKxAR/T application, this research aims to contribute
the relations between the shaded model as the pre-image and to more effective pedagogical approaches for teaching math-
the wireframe model as the image. Also, the non-AR version ematical concepts, ultimately benefiting students in various
does not support an automatic perspective alignment. There- STEM disciplines.
fore, students need to manually navigate through the scene Based on the identified research problem and questions,
(pan, zoom, orbit) for different perspective views and better the study’s hypotheses are as follows:

Fig. 7  Playing with function parameters in the non-AR version of Prototype 1

12 Page 8 of 18 Virtual Reality (2024) 28:12

1. BRICKxAR/T may help students in solving mathemati- Twenty-nine (29) students participated in the AR work-
cal problems, specifically matrices for geometric trans- shop (Fig. 8), among which 31.03% (n = 9) were either very
formations (better improvement in AR than non-AR). familiar or moderately familiar, and 68.97% (n = 20) were
2. The AR version of BRICKxAR/T may impose fewer either slightly or not familiar with digital modeling. 34.48%
task loads compared to the non-AR version of the app. (n = 10) of the students were either very familiar or moder-
3. The AR version of BRICKxAR/T is more exciting and ately familiar, and 65.52% (n = 19) were either slightly or not
engaging to use compared to the non-AR version of the familiar with the AR technology.
app. Thirty (30) students participated in the non-AR work-
shop (Fig. 9), among which 46.67% (n = 14) were either very
In this paper, we present an experiment in which the familiar or moderately familiar and 53.33% (n = 16) were
BRICKxAR/T has been evaluated as a learning interven- either slightly or not familiar with digital modeling (famili-
tion using LEGO models as examples of physical and virtual arity with AR technology was not asked from the non-AR
manipulatives. We conducted an experiment to compare par- group).
ticipants’ learning gains (pre-test to post-test) in spatial visu- Table 1 shows that female and males students partici-
alization and math skills after completing an AR or non-AR pated in the AR workshop have a ratio of approximately 3/1.
workshop. Additionally, we conducted surveys to compare About 51.73% of the AR participants were from STEM col-
the apps’ performance and measure students’ motivation and leges: 24.14% from engineering and 27.59% from science.
play time. The remaining (48.27%) were affiliated with other colleges:
architecture (10.34%) and other non-STEM fields (37.93%).
In the non-AR workshop, female and male participated with
a ratio of approximately 1 to 1. Most participants (73.33%)
3 Methods of the non-AR were from STEM colleges: 50% from engi-
neering and 23.33% from science. Only 26.66% of the
3.1 Participants non-AR participants were from other colleges: architecture
(3.33%) and other non-STEM domains (23.33%).
Upon the human subject research (Institutional Review
Boards, IRB#2020-1213 M) approval, we invited under- 3.2 Design
graduate students, through bulk email messages across Texas
A&M University. Fifty-nine undergraduate students with The current experiment used a mixed design. Participants
normal vision participated in the experiment voluntarily. The were randomly assigned to interact with either the AR ver-
students had at least high school-level knowledge in algebra sion or non-AR version of the BRICKxAR/T, which is a
and geometry. The participants were randomly split into AR between-subjects independent variable. All participants
and non-AR sessions (NAR = 29, Nnon-AR = 30). Data collec- completed a pre-test before interacting with the app and a
tion occurred between Fall 2021 and Spring 2022. post-test afterward, which is a within-subjects independent

Fig. 8  Students playing with the AR version of the app

Virtual Reality (2024) 28:12 Page 9 of 18 12

Fig. 9  Students playing with the non-AR version of the app

Table 1  Demographic information of the participant of AR and non-AR workshops

Group Number Gender (%) Field of study (%)
Female Male Non-binary Engineering Science Architecture Other

AR 29 72.41 (n = 21) 24.14 (n = 7) 3.45 (n = 1) 24.14 (n = 7) 27.59 (n = 8) 10.34 (n = 3) 37.93 (n = 11)

Non-AR 30 46.67 (n = 14) 53.33 (n = 16) 0 (n = 0) 50 (n = 15) 23.33 (n = 7) 3.33 (n = 1) 23.33 (n = 7)

variable. Dependent measures were math test, Purdue Vis-

ualization of Rotations test, NASA TLX ratings, and the
Motivated-Strategies-for-Learning-Questionnaire.

3.3 Materials

In order to have an integrated testing platform, all tests and

surveys were implemented through the online Qualtrics sur-
vey application (https://​www.​qualt​rics.​com/).

3.3.1 Purdue visualization of rotation test (PVRT)

We used the 20-item version of PVRT (Bonder and Guay

1997) to evaluate students’ spatial visualization skills before
Fig. 10  An example question from the 20-item PVRT (Bonder and
participating in the workshops to guarantee that students of Guay 1997)
both groups had similar spatial visualization skills prior to
the workshops. Since prior research shows that spatial skills
may only improve through repetitive sessions over a long
period of time (Casey et al. 2008; Terlecki et al. 2007), we 3.3.2 Math test
did not focus on students’ PVRT post-test scores and their
improvement in a short session workshop. This test consists We designed a math test with 13 questions on transforma-
of 20 multiple choice questions in which students are asked tion matrices, based on learning materials of the Khan
to study how an object is rotated in the sample and select Academy (https://​w ww.​k hana​c ademy.​o rg/​m ath/​geome​
the option that has rotated in the same manner (Fig. 10). In try-​h ome/​t rans​forma​t ions) (Fig. 11). In our analysis,
our analysis, final scores were calculated using an average of final scores were calculated using an average of cor-
correctly answered questions, scaled to 100 like a percentage rectly answered questions, scaled to 100 like a percentage
(range = 0–100, ratio scale). (range = 0–100, ratio scale).
12 Page 10 of 18 Virtual Reality (2024) 28:12

number of times that the factor is selected in the pairwise

questions:
rating score × weight
Adjusted score per factor = (1)
15
The maximum rating score is 100, while the maximum
time that a factor can ever be selected (weight) is 5; hence,
the minimum adjusted score is 0 and the maximum adjusted
score that a factor may achieve is 33.3 (ratio scale) based
Fig. 11  An example of the math test designed based on (https://​www.​ on Eq. (1).
khana​cademy.​org/​math/​geome​try-​home/​trans​forma​tions)

3.3.4 Motivated‑strategies‑for‑learning‑questionnaire
3.3.3 NASA_TLX survey (MSLQ)

We measured the application task workload through the We assessed students’ motivations and interests to play with
NASA-TLX survey (Hart 1980) on six dimensions: mental, the app through a questionnaire based on the MSLQ survey
physical, temporal, effort, frustration, and performance. The (Pintrich and Groot 1990) in three categories of intrinsic
survey contains 6 ranking questions in which each dimen- value, task anxiety, and self-regulated-learning. Each cat-
sion is graded on an interval scale ranging from low 0 to egory consists of multiple questions to evaluate students’
high 10, and 15 pairwise questions in which subscale pairs subjective viewpoints regarding the corresponding item. The
are compared (Fig. 12). questions are scaled in 5 steps starting from 1 to 5 represent-
Based on (Hart 1980), the adjusted score of each factor is ing “strongly agree” to “strongly disagree.” The mean scores
calculated based on its ranking (0–100) and its weight (the of questions within each category are used in the analyses

Fig. 12  Top: An example of the rating questions; bottom: an example of the pairwise questions; adopted from NASA_TLX (Hart 1980)
Virtual Reality (2024) 28:12 Page 11 of 18 12

of this study. Figure 13 shows three example questions from the concepts of the learning materials and an instruction
the intrinsic value category. In our analysis, items in each demo video on using the corresponding apps (AR or non-
category were averaged together (range = 1–5, ordinal scale). AR version). The demo videos were used to minimize the
“Instrument and Misidentification Threat” (Shaghaghian
3.4 Procedure et al. 2021). Then, students were asked to play with the apps
for 20–40 min and record their screens during the play. To
Figure 14 illustrates the workflow that the researchers fol- minimize the “Cross-Group Contamination” effects (Shagh-
lowed to conduct the experiment for evaluating the AR vs. aghian et al. 2022), AR and non-AR workshops were held
non-AR version of BRICKxAR/T. separately, and groups were not aware of each other.
Once recruited, participants completed the pre-session: It is worth noting that the educational content was equal
an online survey (demographic questionnaire) and pre-tests in the AR and non-AR version of the application. Physical
of Purdue Visualization of Rotation Test (PVRT) and a LEGO models were provided for both groups. However, the
matrix algebra math test. To ensure unbiased results, the AR environment allowed students to play and learn from
tests' answers were not disclosed to the students during the abstract information integrated with the physical environ-
pre-session, and randomization was applied to the answer ment. This means that they could move around the room
choices. and move the LEGO model, while the AR version of the
To mitigate the potential Testing and History Threats BRICKxAR/T app responded to those movements and
(Shaghaghian et al. 2021), we conducted the post-sessions dynamically overlaid abstract information. This is because
one to three weeks after the pre-session. This time interval, the LEGO model was attached to image markers for the AR
as recommended by existing literature, strikes a balance by version only. In contrast, the non-AR/desktop version of the
preventing the participants from recalling specific questions app only allowed students to interact with the digital version
(from pre-tests) while also minimizing the likelihood of any of the LEGO model. Any body movement or movement of
unintentional influence from external events or history on the physical LEGO model was not reflected in the digitally
their skills. displayed information.
Participants were randomly assigned to either the AR Following their participation in the workshops, the stu-
or non-AR face-to-face learning workshops. In the post- dents took the same PVRT and math tests (with randomized
session, students first watched an introduction video on answer choices), along with two surveys (NASA_TLX

Fig. 13  Questions of the intrin-

sic value of the motivation ques-
tionnaire inspired by MSLQ
(Garmendia et al. 2007)

Fig. 14  Experiment framework

12 Page 12 of 18 Virtual Reality (2024) 28:12

and Motivated-Strategies-for-Learning-Questionnaire or were statistically different in gender distribution, with the

MSLQ). AR group having more females than the non-AR group. In
this test the only non-binary participant of AR group was
excluded to avoid type I error (caused by a very small sam-
4 Results ple size in one category).
Table 2 shows the pre-test PVRT mean scores of the AR
Before conducting any of the statistical tests, the outliers and non-AR groups. One outlier was excluded from the AR
corresponding to each test were excluded using whiskers data (NAR = 28, Nnon-AR = 30).
box plot. Also, the assumptions, specifically, normality, The between-subjects ANOVA conducted on the PVRT
and homogeneity of variances were verified to determine pre-test scores did not show a significant difference between
whether to apply parametric or an equivalent nonparamet- the AR and non-AR groups (F(1, 56) = 0.68, p = 0.8), which
ric test. If the tests’ pre-assumptions were met, parametric indicates that students of the two groups had similar mental
tests were applied; otherwise, the equivalent nonparametric visualization skills before participating in the workshops for
tests were conducted. We set 0.05 as a threshold for the p learning spatial transformation matrices.
value to accept or reject the null hypothesis associated to
each test. Note that the data illustrated in all tables of this 4.2 Math scores
section reflect descriptive statistics before removing the out-
liers, while statistical tests were conducted after removing Figure 15 left and right illustrates students’ math scores in
the outliers. the pre- and post-sessions of AR and non-AR, respectively.
Table 3 shows that the math mean scores of the AR have
4.1 Demographics and PVRT scores increased after the workshops with more score improvement
in the AR group (29.33%) compared to the non-AR group
All participants of AR and non-AR group were in the age (20.83%). The maximum scores (100) have not changed,
range of 15–30 years old. A Chi-squared test comparing and they represent a ceiling effect (which will be addressed
gender in the AR (males = 7, females = 21) and non-AR in future work).
(males = 16, females = 14) groups found a significant dif- The between-subjects ANOVA t test showed a near sta-
ference, X2 1 = 4.9, p = 0.03. This means that the two groups tistically significant difference on students’ math pre-test
scores between AR and non-AR groups (F(1,56) = 1.6,
p = 0.06). The non-AR group (M = 61.54%) scored margin-
Table 2  PVRT pre-test scores ally higher than the AR group (M = 55.17%), which means
Group AR Non-AR that the non-AR group had slightly better knowledge about
Pre-test score (%) Pre-test score (%) matrix algebra. Because of this nearly statistically significant
difference in pre-test scores, we ran two ANCOVAs below
Min 20 20
that allow us to account for the impact of pre-test scores
Max 100 100
in investigating learning gains differences between the two
Mean 64.14 65
conditions.
SD 23.64 23.27

Fig. 15  Students’ math scores in pre- and post-sessions of workshop 1 AR (left) and non-AR (right)
Virtual Reality (2024) 28:12 Page 13 of 18 12

Table 3  Descriptive statistical Math scores

information of the math scores
Group AR Non-AR
Pre-test score (%) Post-test Score Pre-test score (%) Post-test Score
score (%) improvement score (%) improve-
(%) ment (%)

Min 23.1 38.5 + 66.7 23.1 15.4 − 33.3

Max 100 100 0 100 100 0
Mean 55.17 71.35 + 29.33 61.54 74.36 + 20.83
SD 20.15 20.44 N/A 20.15 23.67 N/A

1. ANCOVA on math post-test, with group as the inde- Table 4  Effect test result of the ANCOVA model on post-test scores
pendent variable having pre-test as the covariate, with group, field of study, and gender
as independent variables

This model examined if the post-test score was a func- Variables F ratio p value
tion of the pre-test score and whether the function changed
Math pre-test 25.56 < 0.001
based on the group (AR vs. non-AR). The ANCOVA
Group 1.89 0.18
model had an adjusted R2 = 0.40, which revealed that 40%
Field of study 0.99 0.4
of the variability of the response variable, i.e., math post-
Gender 0.09 0.77
test scores, can be explained by the linear model fitted to
the data. The results show that the math pre-test is signifi-
cantly predictive of the math post-test, F(1, 56) = 40.06,
p < 0.001. This means that students with high pre-test or field of study. Both groups, both genders, and both fields
scores tended to have high post-test scores, as expected. of study showed similar improvement in math scores.
However, group (AR vs. non-AR) did not have a signifi- Table 4 reports the results of another ANCOVA, which
cant impact, F(1, 56) = 0.56, p = 0.46, on the post-test allowed for differences in pre-test scores or other relevant
scores when considering the pre-test as the covariate. This characteristics, like gender or field of study in our case, to
means that when we statistically control for the impact of be statistically controlled for.
pre-test scores on post-test scores, there is no difference
between the two groups. Therefore, with AR showing 29% 4.3 Task load analysis
improvement and non-AR showing 21% improvement,
both learning workshops improved matrix algebra learn- We compared the task load of the AR vs. non-AR versions
ing similarly. of the app through the NASA_TLX survey. Figure 16 shows
students’ adjusted scores calculated for the corresponding
2. ANCOVA on math post-test, with group as the inde- six factors of NASA_TLX in AR (left image) vs. non-AR
pendent variable and field of study and gender as the (right image) sessions.
extraneous variables Table 5 shows that students generally rated the prototypes
as a low demanding task as the means of the adjusted scores
This model examined if the post-test score was a func- associated with all factors were less than 50% of the maxi-
tion of the pre-test score and whether the function changed mum possible value of 33.3 (see Eq. 1).
based on the group (AR vs. non-AR), considering field of Based on the tests’ pre-assumptions, either ANOVA or
study (STEM vs. non-STEM) and gender as the extraneous the nonparametric Wilcoxon Rank Sum test (also known
explanatory variables. The summary of fit for this ANCOVA as Mann–Whitney U test) with normal approximation was
model showed adjusted R2 = 0.42, meaning that 42% of the conducted on the adjusted scores corresponding to each
variability of the response variable, i.e., post-test scores, factor. Outliers were removed if existed (4 outliers were
can be explained by this model. The results show that pre- removed from the data of physical demand factor and
test is highly correlated with the post-test score (again, as 1 outlier was removed from the data corresponding to
expected), F(1, 55) = 25.56, p < 0.001, while neither group, the performance factor). The results showed that students
field of study, nor gender has a statistically significant impact adjusted scores were not significantly different in men-
on the post-test scores. This means that when we statistically tal demand (F(1,57) = 2.22, p = 0.14), temporal demand
control for the impact of pre-test scores on post-test scores, (Z = − 0.76, p = 0.45), frustration (Z = 1.68, p = 0.09),
there is no difference between the two groups, two genders, and performance (Z = − 0.1, p = 0.92) between AR and
12 Page 14 of 18 Virtual Reality (2024) 28:12

Fig. 16  Students’ adjusted scores of the six factors of the NASA_TLX survey corresponding to AR (left) vs. non-AR (right) groups

Table 5  Minimums, maximums, means, and standard deviations (std) 4.4 Motivation analysis
of adjusted scores per factor between AR and non-AR
We evaluated students’ motivations in three categories
Factors Group Min Max Mean SD
through a motivation questionnaire (adopted from MSLQ
Mental demand AR 2 30 14.4 14.41 Pintrich and Groot 1990) (Table 6).
Non-AR 0 26 11.4 7.34 The result of the nonparametric Wilcoxon Rank Sum tests
Physical demand AR 0 23 5.4 5.38 on students’ answers to each category did not show any sig-
Non-AR 0 16 2.2 3.93 nificant difference between AR and non-AR groups with the
Temporal demand AR 0 10 2.4 2.41 following test results: intrinsic value: Z = − 1.02, p = 0.31;
Non-AR 0 13 3.5 4.07 task anxiety: Z = − 0.6, p = 0.55; self-regulated-learning
Effort AR 1 24 11.8 11.83 result of Z = − 1.62, p = 0.11. For both groups, the results of
Non-AR 0 21 8.4 6.1 the intrinsic value category show that students agreed (total
Frustration AR 0 26 7.4 7.41 MAR = 2.1, ­SDAR = 0.2, total Mnon-AR = 2.3, ­SDnon-AR = 0.2)
Non-AR 0 16 4.2 5.15 that what they learned in the workshop was an interesting,
Performance (negative AR 0 18 5.6 5.62 useful, and important subject to learn, with the AR group
correlation) agreed more than the non-AR group. The task-anxiety cat-
Non-AR 0 16 5.3 4.04 egory results reveal that students somewhat disagreed (total
Overall task load (%) AR 25 83 47.1 14.55 MAR = 3.8, ­SDAR = 0.2, and total Mnon-AR = 4, SD non-AR = 0.2)
Non-AR 0 68 35 17.28 about how uneasy and upset they felt during the workshop,
with the non-AR group disagreed more than the AR group.
The results from the self-regulated-learning category show
that students agreed (total MAR = 1.7, ­SDAR = 0.1, total
Mnon-AR = 2, ­SDnon-AR = 0.1) that the learning materials of
non-AR groups. However, the data were significantly dif- the workshops were helpful in self-learning and answering
ferent in physical demand (Z = − 2.8, p = 0.005), effort the questions, with AR group agreed more than the non-AR
(Z = 2.27, p = 0.02). This outcome reveals that students group.
felt a substantially higher task load, specifically regarding
physical demand and effort, in the AR group compared to 4.5 Play time
their peers in the non-AR group, which may be derived
from the physical load of holding the device and playing During the workshops, students were encouraged to play
with the physical model at the same time (in the AR work- with the application for 20–40 min; however, they were free
shop) and their unfamiliarity with the AR environment. to quit when they did not feel like playing more. The results
from the iPads’ screen records showed that, on average, stu-
dents in the AR sessions were more willing to play with the
Virtual Reality (2024) 28:12 Page 15 of 18 12

Table 6  Mean values of students’ ratings to the questions of the MSLQ survey (range 1 strongly agree to 5 strongly disagree)
Category Item Mean value
AR Non-AR

Intrinsic value I like what I learned, and I think the subject is interesting 1.7 1.9
I think that what I learned in this workshop is useful for me to know 2.1 2
Understanding this subject is important to me 2.6 2.9
Mean of the intrinsic value category questions 2.1 2.3
Task anxiety I was so nervous during the workshop that I could not remember the material I had learned 3.6 3.5
I had an uneasy, upset feeling when I was participating in the workshop 4.3 4.6
When I was performing the task in the workshop, I thought about how poorly I was doing 3.6 3.9
Mean of the task-anxiety category questions 3.8 4
Self-regulated learning When I was taking the tests (math and mental rotation test), I put together what I learned in the work- 1.6 1.8
shop and lecture
When I was taking the math test, I used visual imagery to visualize the geometric transformations and 1.6 1.9
what I experienced in the workshop
Visualization of matrix representations during the workshop helped me in solving the math test 1.8 2.2
Mean of the self-regulated-learning category questions 1.7 2

the non-AR app, with a significant difference (Z = − 5.51,

p < 0.001). Based on the play time we may interpret that the
AR app was more interesting and engaging for the students
to learn through play.

5 Discussions

This paper presented BRICKxAR/T and the experiment with

test cases to evaluate the apps in the AR vs. non-AR environ-
ments. BRICKxAR/T is an AR educational tool for learning
Fig. 17  Students’ play time with the application in the AR vs. non- the geometric reasoning behind mathematical representa-
AR sessions tions of spatial transformations. In the test cases, we first
guaranteed that both groups had similar spatial visualization
skills prior to participation in the workshops through PVRT.
Table 7  Descriptive statistical Play time Then, we assessed students’ learning gains in math skills
information of students play (minute) through the math test on transformation matrices before
time with the apps
Group AR Non-AR and after participating in the workshops and compared the
results between the AR and non-AR groups. The application
Min 17 8.78
task load and participants’ motivations were also evaluated
Max 35.95 22.65
through NASA_TLX and MSLQ surveys, respectively.
Mean 25.33 15.99
Based on the between-subjects ANOVA result of the
SD 5.56 3.8
PVRT pre-test scores, students had similar spatial visuali-
zation skills before participating in the workshops. Based
on the between-subjects ANOVA result of the math pre-
application (M = 25.33 min) compared to their peers in the test scores, students of non-AR group had better knowledge
non-AR sessions (M = 15.99 min) (Fig. 17). (near statistically significant difference) than the AR group
We used the data of screen recordings to compare the before the workshops. Students’ math scores for matrix alge-
length of students’ play time in the AR vs. non-AR work- bra improved significantly after the workshops in both of
shops (Table 7). the AR and non-AR groups, with more improvements in
The result of the nonparametric Wilcoxon Rank Sum tests the AR group. Hence, the researchers conclude that the fea-
on students’ play time with the apps showed that students tures integrated in BRICKxAR//T interventions (in both AR
played more with the AR app during the workshops than and non-AR, and especially in AR) may improve students’
12 Page 16 of 18 Virtual Reality (2024) 28:12

understanding of mathematical representations of spatial • .BRICKxAR/T showed the potential to help students
transformations. conceive, connect, and compare math conceptions of
This research has not conducted a comparative study motions, mappings, and functions in AR to help over-
between the AR group and a conventional “control group” come well-documented difficulties students face when
(for example, a group that learns the same topic through con- learning spatial transformations and allied mathemati-
ventional methods, such as lectures, handbooks, or videos). cal representations.
The reason is that CAL has been practiced widely and has • The functions of BRICKxAR/T help students see rela-
already shown significant improvement compared to the con- tionships between spatial manipulations and mathe-
ventional methods due to literature. AR can be considered matical operations, bridging the spatial–mathematical
as a novel CAL method, and thus we decided to compare divide.
BRICKxAR/T with a higher bar, which is our non-AR set- • BRICKxAR/T’s motion tracking of physical object
ting, and the results are promising. In particular, students transformations and the student’s hand that controls
in the AR group agreed more than the non-AR group on the objects showed the potential to help collect fine-
that what they learned in the workshop was an interesting, grained behavior data to enhance learning analytics, for
useful, and important subject to learn, and that the learning example, what the rotation axes and angles are when a
materials of the workshops were helpful in self-learning and hand rotates a physical model.
answering the questions. More importantly, the data from
the screen recordings showed that students were willing to The results from the NASA_TLX showed that students
spend significantly more time playing with AR than non- perceived significantly more physical load and effort while
AR app. This suggests that students are more interested and playing in the AR workshop compared to their peers in
engaging in the BRICKxAR/T learning environment. This the non-AR workshop. The researchers believe that the
result is also aligned with researchers’ observations: Most outcome may be driven by the current limitations of
students of AR group were more excited and curious to play BRICKxAR/T, which require students to hold the iPad
with the AR app and spent more time to play and learn, (an affordable AR device for learning) with one hand and
while many students in the non-AR group got bored early manipulate the physical model or the screen user interface
and stopped playing and learning. with the other hand. Also, the unfamiliarity of students
with the AR environment may impact their subjective
assessment of the effort factor, meaning that they felt more
6 Conclusions and future work effort in learning the new environment. Leveraging 3D
model registration, such as Vuforia Model Target (Vuforia
BRICKxAR/T has the potential to support many students Developer Library 2020), and exceling the implementation
who struggle with spatial and math reasoning, especially through immersive devices (such as the more convenient
those from underrepresented groups in STEM. Gaining a and immersive future AR glasses) may improve the task
fundamental understanding of spatial transformations can load rating by the AR users. Addressing the limitations
uniquely contribute to students’ learning and development of the current application in future work is expected to
of spatial reasoning and allied mathematical skills, leading significantly improve the learning of spatial transforma-
to improve STEM coursework, STEM retention, and degree tions and their mathematical representations utilizing AR
attainment, and thus support students’ future development technologies.
of expertise and career success across STEM disciplines.
Supplementary Information The online version contains supplemen-
The contributions of this study include the follow- tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 10055-0​ 23-0​ 0904-x.
ing computer–human interaction technology and learning
innovations: Acknowledgements This material is based upon work supported by the
National Science Foundation under Grant No. 2119549.
• BRICKxAR/T is an innovative learning environment that
Data availability Sample data generated or analyzed during this study
enables physical and virtual interplays to engage students are included in this published article (a supplementary video demo and
in embodied learning (by transforming physical models the developed math test for matrix algebra, submitted with the manu-
by hand). script, will be available publicly upon acceptance of the manuscript
for publishing). Other datasets generated and/or analyzed during the
• BRICKxAR/T integrates spatial transformation matrices
current study are available from the corresponding author on reason-
and related math information with the physical model able request.
movement controlled by students in AR, making diffi-
cult invisible concepts visible for supporting an intuitive Declarations
and formal understanding of spatial reasoning and math-
Conflict of interest No conflicts of interest/competing interests.
ematical formulation.
Virtual Reality (2024) 28:12 Page 17 of 18 12

Ethical statement The manuscript complies to the Ethical Rules appli- Seichter H (2007) Augmented reality and tangible interfaces in col-
cable for this journal as stated in the Instructions for Authors of the laborative urban design. Comput Archit Des Futur 2007:3–16.
journal Virtual Reality. https://​doi.​org/​10.​1007/​978-1-​4020-​6528-6_1
Kendon A (1994) Do gestures communicate? A review. 1813(1994).
https://​doi.​org/​10.​1207/​s1532​7973r​lsi27​03
Figen Gül L, Halıcı SM (2015) Collaborative design with mobile
augmented reality. In: Complex. Simplicity—Proc. 34th
References eCAADe Conf., vol 1, pp 493–500
Dünser A, Steinbügl K, Kaufmann H, Glück J (2006) Virtual and
Wheatley GH (1990) Spatial sense and mathematics learning. Natl augmented reality as spatial ability training tools. ACM Int
Counc Teach Math 37(6):10–11 Conf Proc Ser 158:125–132. https://​doi.​org/​10.​1145/​11527​60.​
van Garderen D (2006) Spatial visualization, visual imager, and mathe- 11527​76
matical problem solving of students with varying abilities. J Learn Hohenwarter M (2018) GeoGebra. https://w ​ ww.g​ eogeb​ ra.o​ rg/?l​ ang=e​ n
Disabil 39(6):496–506 de Ravé EG, Jiménez-Hornero FJ, Ariza-Villaverde AB, Taguas-
Rellensmann J, Schukajlow S, Leopold C (2017) Make a draw- Ruiz J (2016) DiedricAR: a mobile augmented reality system
ing. Effects of strategic knowledge, drawing accuracy, and designed for the ubiquitous descriptive geometry learning. Mul-
type of drawing on students’ mathematical modelling perfor- timed Tools Appl 75(16):9641–9663. https://​doi.​org/​10.​1007/​
mance. Educ Stud Math 95(1):53–78. https://​doi.​org/​10.​1007/​ s11042-​016-​3384-4
s10649-​016-​9736-1 Kaur N, Pathan R, Khwaja U, Murthy S (2018) GeoSolvAR: aug-
Kim MJ, Lou Maher M (2008) The impact of tangible user interfaces mented reality based solution for visualizing 3D Solids. In: Proc.
on designers’ spatial cognition. Hum-Comput Interact. 23(2):101– IEEE 18th Int. Conf. Adv. Learn. Technol. ICALT 2018, pp 372–
137. https://​doi.​org/​10.​1080/​07370​02080​20164​15 376. https://​doi.​org/​10.​1109/​ICALT.​2018.​00093
Preece D, Williams SB, Lam R, Weller R (2013) ‘Let’s Get Physical’: Weng S-C, Wu J-P, Yu C-H, Wu C-C (2014) Using augmented real-
advantages of a physical model over 3D computer models and ity to assist learning geometry concepts. In: World conference
textbooks in learning imaging anatomy. Anat Sci Educ 6(4):216– on e-Learning in corporate, Government, healthcare, and higher
224. https://​doi.​org/​10.​1002/​ase.​1345 education 2014(1), 2027–2031. http://​www.​editl​ib.​org/p/​148964/
Hoe ZY, Lee IJ, Chen CH, Chang KP (2019) Using an augmented Rossano V, Lanzilotti R, Cazzolla A, Roselli T (2020) Augmented real-
reality-based training system to promote spatial visualization abil- ity to support geometry learning. IEEE Access 8:107772–107780.
ity for the elderly. Univers Access Inf Soc 18(2):327–342. https://​ https://​doi.​org/​10.​1109/​ACCESS.​2020.​30009​90
doi.​org/​10.​1007/​s10209-​017-​0597-x Le HQ, Kim JI (2017) An augmented reality application with hand ges-
Hollebrands KF (2003) High school student’s understanding of geomet- tures for learning 3D geometry. In: 2017 IEEE international con-
ric transformation in the context of a technological environment. J ference on big data and smart computing (BigComp), pp 34–41.
Math Behav 22(1):55–72. https://d​ oi.o​ rg/1​ 0.1​ 016/S
​ 0732-3​ 123(03)​ https://​doi.​org/​10.​1109/​BIGCO​MP.​2017.​78817​12
00004-X Kaufmann H, Schmalstieg D (2003) Mathematics and geometry
Gülkılıka H, Uğurlub HH, Yürükc N (2015) Examining students’ education with collaborative augmented reality. Comput Graph
mathematical understanding of geometric transformations using 27(3):339–345. https://​doi.​org/​10.​1016/​S0097-​8493(03)​00028-1
the pirie-kieren model. Kuram Ve Uygulamada Egit Bilim Cahyono B, Firdaus MB, Budiman E, Wati M (2018) Augmented
15(6):1531–1548. https://​doi.​org/​10.​12738/​estp.​2015.6.​0056 reality applied to geometry education. In: Proceedings—2nd east
Garmendia M, Guisasola J, Sierra E (2007) First-year engineering stu- Indonesia conference on computer and information technology:
dents’ difficulties in visualization and drawing tasks. Eur J Eng internet of things for industry, EIConCIT 2018, pp 299–303.
Educ 32(3):315–323. https://​doi.​org/​10.​1080/​03043​79070​12768​ https://​doi.​org/​10.​1109/​EICon​CIT.​2018.​88785​53
74 Nemec M, Fasuga R, Trubac J, Kratochvil J (2017) Using virtual real-
Martin J, Saorín J, Contero M, Alcaniz M, Perez-lopez DC, Mario O ity in education. In: 15th international conference on emerging
(2010) Design and validation of an augmented book for spatial eLearning technologies applications proceedings. https://​doi.​org/​
abilities development in engineering students. Comput Graph 10.​1109/​ICETA.​2017.​81025​14
34:77–91. https://​doi.​org/​10.​1016/j.​cag.​2009.​11.​003 Anifowose H, Wei Y, Dixit M (2022) Interactive virtual construction‚ a
Tang A, Owen C, Biocca F, Mou W (2003) Comparative effectiveness case study of building component assembly towards the adoption
of augmented reality in object assembly. Conf Hum Factors Com- of BIM and VR in business and training.
put Syst Proc 5:73–80. https://​doi.​org/​10.​1145/​642625.​642626 Azuma RT (1997) A survey of augmented reality. Presence Teleoper
Fuys D, Geddes D, Tischler R (1988) The van Hiele model of thinking Virtual Environ 6(4):355–385. https://​doi.​org/​10.​1162/​pres.​
in geometry among adolescents. Res. Math. Educ. Monograph 1997.6.​4.​355
i–196 Velev D, Zlateva P (2017) Virtual reality challenges in education and
Sorby SA (2009) Educational research in developing 3-D spatial skills training. Int J Learn Teach 3(1):33–37. https://​doi.​org/​10.​18178/​
for engineering students. Int J Sci Educ 31(3):459–480. https://​ ijlt.3.​1.​33-​37
doi.​org/​10.​1080/​09500​69080​25958​39 Kruijff E, Poupyrev I, LaViola JJ, Bowman DA (2017) 3D user inter-
Viswanathan VK, Linsey JS (2012) Physical models and design think- faces: theory and practice. Addison-Wesley, Boston
ing: a study of functionality, novelty and variety of ideas. J Mech Bujak KR, Radu I, Catrambone R, MacIntyre B, Zheng R, Golubski G
Des Trans ASME 134(9):1–13. https://d​ oi.o​ rg/1​ 0.1​ 115/1.4​ 00714​ 8 (2013) A psychological perspective on augmented reality in the
Chandrasekera T, Yoon SY (2015) The effect of tangible user interfaces mathematics classroom. Comput Educ 68:536–544. https://​doi.​
on cognitive load in the creative design process. In: Proc. 2015 org/​10.​1016/j.​compe​du.​2013.​02.​017
IEEE Int. Symp. Mix. Augment. Real.—Media, Art, Soc. Sci. Shaghaghian Z, Burte H, Song D, Yan W (2021) Learning geometric
Humanit. Des. ISMAR-MASH’D 2015, pp 6–8. https://​doi.​org/​ transformations for parametricdesign: an augmented reality (AR)-
10.​1109/​ISMAR-​MASHD.​2015.​18 powered approach. In: International conference on computer-aided
Sun L, Fukuda T, Tokuhara T, Yabuki N (2014) Differences in spatial architectural design futures, pp 515–527
understanding between physical and virtual models. Front Archit Shaghaghian Z, Burte H, Song D, Yan W (2022) learning spatial
Res 3(1):28–35. https://​doi.​org/​10.​1016/j.​foar.​2013.​11.​005 transformations and their math representations through embodied
12 Page 18 of 18 Virtual Reality (2024) 28:12

learning in augmented reality. In: International conference on “Khan Academy.” https://​www.​khana​cademy.​org/​math/​geome​try-​

human-computer interaction, pp 112–128 home/​trans​forma​tions
Fife JH, James K, Bauer M (2019) A learning progression for geomet- Hart S (1980) Human performance research group. NASA task load
ric transformations. ETS Res Rep Ser 2019(1):1–16. https://​doi.​ index user manual v. 1.0. https://​human​syste​ms.​arc.​nasa.​gov/​
org/​10.​1002/​ets2.​12236 groups/​TLX/
Yan W (2021) Augmented reality instructions for construction toys Pintrich PR, De Groot EV (1990) Motivational and self-regulated
enabled by accurate model registration and realistic object/ learning components of classroom academic performance. J Educ
hand occlusions. Virtual Real, pp 1–14. https://​doi.​org/​10.​1007/​ Psychol 82(1):33–40
s10055-​021-​00582-7 Vuforia developer library (2020) https://​libra​ry.​vufor ​ia.​com/​featu​res/​
Qualtrics XM The leading experience management software. https://​ objec​ts/​model-​targe​ts.​html
www.​qualt​rics.​com/ Accessed 25 Jan 2022
Bonder GM, Guay RB (1997) The purdue visualization of rotations Publisher's Note Springer Nature remains neutral with regard to
test. Chem Educ 2(4):1–17. https://​doi.​org/​10.​1007/​s0089​79701​ jurisdictional claims in published maps and institutional affiliations.
38a
Casey BM, Andrews N, Schindler H, Kersh JE, Samper A, Copley J Springer Nature or its licensor (e.g. a society or other partner) holds
(2008) The development of spatial skills through interventions exclusive rights to this article under a publishing agreement with the
involving block building activities. Cognit Instr 26(3):269–309. author(s) or other rightsholder(s); author self-archiving of the accepted
https://​doi.​org/​10.​1080/​07370​00080​21771​77 manuscript version of this article is solely governed by the terms of
Terlecki MS, Newcombe NS, Little M (2007) Durable and generalized such publishing agreement and applicable law.
effects of spatial experience on mental rotation: gender differences
in growth patterns. Appl Cognit Psychol 22:877–895. https://​doi.​
org/​10.​1002/​acp