The Use of Augmented Reality in Latin-American Engineering Education: A Scoping Review

As part of a recent change, Augmented Reality (AR) has filled engineering classrooms, being employed for various pedagogical purposes around the world. However, little is known about the different features and uses of this technology in Latin America. This Scoping Review asks how are educational AR systems designed, used and evaluated in the region, comparing this to the international literature. To address this question, we charted 36 conference papers and scientific articles, taking care of the quality gaps and methodological diversity within our sample. Our results show that, even though most converge on conventional research, design and pedagogical practices, engineering educators working at institutes are taking the lead of design, pedagogical and research innovation. Furthermore, we show that Latin-American literature distinctively reveals how teachers adapt to the particular contexts of teaching, and the special importance of the usually overlooked conference papers literature.


INTRODUCTION
We tend to assume that, for pedagogical purposes, STEM faculties should be the first to adopt current teaching technologies before any other educational institution, as show various summaries of innovations Mkrttchian et al., 2019). But specialists seem to overlook the fact that this progressive change depends on a sum of external factors apart from the sole diffusion of innovations, as happens to be the case with AI, e-learning, and data mining (Aljawarneh, 2020;Alyahyan & Dustegor, 2020;Greenan, 2021;Zawacki-Richter et al., 2019). A case could be made for the inclusion, among these, of the global skill inequalities, which affect higher education in the developing world .
At least since Ivan Sutherland's The Ultimate Display (1965), Augmented Reality (AR), referred as well as "advanced", "improved" and "enriched reality", has been understood as a set of applications that complement or combine real and digital environments, ideally blurring the difference between the two (Billinghurst, 2021;Iatsyshyn et al., 2020). The use of AR in education was first introduced by the aviation industry at the end of the twentieth century, transforming higher education ever since (Akcayır & Akcayır, 2017;Wang et al., 2018). However, the use of AR in higher education became a protagonist in devoted conferences and publications only during the last five years (Altinpulluk, 2019). As a consequence, numerous recent literature reviews report that the AR literature is filled with evidence-based pedagogical practices and innovative design procedures, benefiting students' motivation and learning (Altinpulluk, 2019;Garzón et al., 2019;Nesenbergs et al., 2021;Sommerauer & Müller, 2018). AR seems to be especially useful for STEM education (Ibáñez & Delgado-Kloos, 2018;Sirakaya & Sirakaya, 2 / 20 2018. And since they are usually considered CTML and mobile-learning related technologies, the benefits of AR's environment enhancement for engineering education are well known (Diao & Shih, 2019;Hernandez-de-Menendez et al., 2020;Singh et al., 2019). But the adoption of AR in STEM teaching still faces some challenges, like the lack of knowledge and skills in teachers, as well as institutional barriers (Barroso Osuna et al., 2019). Additionally, the evidence about the use of AR in the very diverse field of engineering education is still quite unknown in contrast to its applications in other educational disciplines and levels. Considering the existing STEM education disparities in the world (Drew, 2020), we hypothesize that these issues may worsen in Latin-American universities, but there's scant evidence of this.
There are remarkable gaps in the extant literature. Diao and Shih (2019) and Singh et al. (2019) reviewed the research designs, educational outcomes and technological features of AR technologies in journal papers, focusing specifically on Architectural, Civil Engineering and Electronics education. Sirakaya and Sirakaya (2018) performed a similar systematic review including science education and medical training. Other reviews include technologies such as VR (Wang et al., 2020). However, we note a scarcity of variables measured, quality appraisal reports and a general lack of interest in this area of research. Moreover, there is a complete relegation of other literature types as part of these needed technology evaluation synthesis, even though most innovation reports are not published through journal articles. A scoping review under the PRISMA-ScR guidelines seem to be the best choice for an exploratory path.
Hence, this paper addresses the following questions: how are AR systems designed, used and evaluated in engineering education in Latin America, and how does this compares with the rest of the world? To address this question, we present a scoping review of papers and conference articles published by Latin-American authors. To do this, we chart publications from four international databases and perform a threefold quality appraisal according to the different literature types found. We draw inspiration from a wide diversity of contributions: among these, reviews about the use of dynamic and static contents, pedagogical affordances, evaluation types and outcomes of education-oriented AR.

METHOD
Scoping reviews are comprehensive literature reviews that bring provisional answers to general questions, not requiring the precision of a systematic review (Munn et al., 2018). Previous recent international literature reviews were normally systematic reviews, a few of them being meta-analyses (Garzón et al., 2019) or less systematic methods (Altinpulluk, 2019). This includes a previous scoping review published in this journal (Saltan & Arslan, 2016), which inspired this work. But, in contrast with the latter, we focus on one particular geographical region and follow the 20 PRISMA-ScR criteria for scoping reviews, proposed originally for literature reviews of medical journals and articles (Tricco et al., 2018).
Scoping reviews under the PRISMA framework proceed by defining research questions, inclusion criteria, search strategies and sources, literature screening, selection, extraction and analysis processes, and result reporting along with discussions (Peters et al., 2020). The protocol for this review was registered in OSF (Bellido García & Paucar Villacorta, 2021). The complete process is reported in Figure 1. Our research questions were the following: • What are the main bibliometric patterns of the Latin-American literature reviewed?
• What types of software and hardware systems prefer Latin-American engineering educators employing AR?
• What pedagogical perspectives and practices guide the educational applications of this technology?
• What are the stated advantages and disadvantages of using AR systems in Latin-American engineering education?
• What are the research designs in Latin-American tests and evaluations of the said technology?
• Are there significant differences between our results and similar international reviews?
Earlier reviews typically focus on English-written academic papers gathered from sources like SSCI, Scopus, and Google Scholar. We chose to depart from this trend in three ways. First, we selected the databases considering their importance for Latin-American authors: Scielo, the Red Iberoamericana de Información y Conocimiento Científico (REDIB), Web of Science (WOS), and SCOPUS. Secondly, inspired on the recent appraisal

Contribution to the literature
• We focus on Latin-American publications, quite neglected on earlier reviews. • Our PRISMA informed Scoping Review notably includes conference papers and scientific articles.
• We build a composite quality index for the quality appraisal of IT case presentations, observational and quasi-experimental studies.
3 / 20 of grey literature to conduct literature reviews (Adams et al., 2017;Garousi et al., 2019;Hartling et al., 2017), we decided to include conference proceedings and scientific journals in my search.  Table 2. Scopus and WOS allowed me to be much more specific with my search, and hence produced larger search strings. Abstract screening lasted for one month after the original search of databases, and the quality assessment of the collected evidence lasted for two more months. While this process was made by only one author, eligibility and quality criteria were chosen by debate and consensus after parallel readings of the extant literature.
Hence, we iteratively designed and tested a weighted quality index for each report based on three components: an indicator of the quality of technology design presentations (based on the principles set by Isaksson et al. (2020), Petersen (2020) and Schön et al. (2017)), an indicator of the quality of the empirical testing or evaluation of the technology earlier presented, excluding design-only papers (Liu et al., 2016;Mårtensson et al., 2019), and an independent indicator of the quality of quasi-experimental designs (drawn mostly from Cochrane criteria). The final criteria list with weights and requirements is shown in Table 3. The indicators where defined as the division between the sum of weights and the weights of all applicable criteria for the current paper. The second component extended to three additional criteria when the papers where comparative or regression based-designs. All papers below the 40% threshold in all three indicators at the same time were excluded. The composite index was defined by ∑ max( ) − ∑( ), where are the existing indicators for the three types of literature, and quartiles Does the report answer the research question defined or presents the results in a clear way? 1 All 4 Is the report based on research? 1 All 5 Is the report well-written? 1 All 6 Is there any intention to be a technological innovation? 1 All 7 Is the technological design based on recent innovations? 1.5 All 8 Does the author succeed in developing a legitimate innovation (e. g., is the software more useful than already existing software?) 1.5 All 9 Does the author add additional relevant information? (Code, operation steps, common problems and their resolution) 1.5 All 10 Does the report include images representing steps of operation? 1.5 All 11 Is there an explicit relationship between a pedagogical perspective and the technology described? where calculated as an additional variable for exploratory data analysis.
Following the selection of a final sample of documents (n=36), we automatically extracted bibliographic data using Zotero (database name, author, year, country, publication, item type, accessibility and URL/DOI). We defined thirty-five variables for the chart, divided in four big groups: bibliographic details, research design, AR design features, and pedagogical traits of the AR systems, along with a quality index variable and a final reviewer commentary. Many variables were inspired in earlier reviews; buy the variable "Application type", drawn from Altinpulluk (2019), was simplified to indicate exclusive categories. Only three variables of the second and third group weren't taken from the literature, including the presence of coding tasks, and the origin of 3D models in 3D-based AR applications. Furthermore, we grouped many of these variables, including: journal name, engineering specialization, type of educational institution, type of device, software name, pedagogical perspectives, pros and cons. The full list of variables along with their sources in the literature and examples are shown in Tables 4-9. Weak Aplicación de realidad aumentada para la enseñanza de la robótica REDIB Mexico [8] Weak Aplicación móvil conrealidad aumentada para la asignatura de metodología de la investigación REDIB Mexico [9] Good Aplicación móvil de realidad aumentada, utilizando la metodología mobile-d, para el entrenamient de técnicos de mantenimiento de maquinaria pesada en la empresa zamine service peru sac

REDIB Peru
[10] Regular Arquitectura interactiva como soporte al aprendizaje situ ado en la enseñanza de la ingeniería      The results described below where obtained after exploratory data analysis and visualization during the last month of this research. We decided to add to this analysis the quality variable to minimize our bias against the supposed bad quality of Latin-American research, as stated in PRISMA guidelines. We carefully chose the most telling results, given the space limitations; however, we specifically compare our results with those in other reviews on the subject. We later summarize and interpret these findings within the bigger framework of education technology in the discussion.

Bibliometric Patterns
It is usually thought that Brazil is the Latin-American country with the largest scientific productivity in the region, given the prominence of Brazilian authors and journals in Scopus (UNESCO, 2021). However, the largest part of the documents reviewed were written by Mexican (n=12, 33%), Colombian (n=9, 25%) and Brazilian (n=7, 19%) authors. Ecuadorian, Chilean and Peruvian authors only authored 8 of the 36 reports (22%). Furthermore, we noticed that Brazil was the country with the least percentage of documents in Scopus, while the opposite happened with Mexico. This outstanding fact was also found in international reviews that give importance to less science-productive countries than the US or the UK in the pedagogical AR usage-related literature, like Taiwan and Spain (Altinpulluk, 2019;Diao & Shih, 2019 (Bacca et al., 2014;Iatsyshyn et al., 2020). In contrast, 50% of our sample was found in journals or proceedings primarily published in Spanish or Portuguese. Among the rest, just one paper was published in the third of the before listed journals.
Our sample seems to have been progressively accumulating in the span between 2015-2020, following the international trend (Altinpulluk, 2019;Diao & Shih, 2019;Ibáñez & Delgado-Kloos, 2018). However, we notice a delay in the productivity peak: Even though Altinpulluk (2019) shows an increasing rate of production during and Diao and Shih (2019 between 2017-2018, we only found a notorious increase in the number of Latin-American documents between 2018-2019. Interestingly, this was driven by a numerical increase of documents from subscription-based journals indexed in Scopus, whereas open-access documents stagnated within the five-year period (except for the REDIB documents, that are decreasing in number versus new Scopus open-access documents). This pattern seems important, given that the extant literature usually rely on WOS or Scopus only. Figure 2 shows the number of documents by type and data-source per year.
Nonetheless, we believe that this change was rather related with an increase in the number of Scopusindexed international conference papers. In fact, the number of papers published in peer-reviewed journals stagnated since 2018 at a rate of three papers per year. It is usually thought that the first are texts of lesser quality than the latter. Overall, we found six documents located in the first ("optimal") quartile of our quality index, and remaining quartiles contained ten documents each. The number of documents in the third quartile seems to be increasing with time, and in parallel, the fourth and first quartiles shrink. However, we report no relevant differences between the quality of documents screened from different databases or publication types. This finding contrasts with the current selection practices in other reviews, which seem to be guided by an exclusion bias.

Research Designs
Though some of the works reviewed are simple fulltext descriptions of the design process of technologies or classroom activities (n=9, 25%), most include some form of empirical evaluation or testing, either by observational (n=5, 14%), pre/post (n=15, 42%) or quasiexperimental (n=7, 19%) designs. Two documents ([5], [19]) are only evaluations. Questionnaires (n=13, 36%), academic grades (n=4, 11%) and a mix of both (n=4, 11%) compose the largest part of data collection techniques used, although some documents also mention qualitative techniques as interviews and observational forms (n=4, 11%), two mention object recognition data ([4], [17]), one mentions Emotiv Insight cognitive sensory data ([22]) and another one system development outputs ([33]). In contrast, Diao and Shih (2019) find predominantly experimental designs in their engineering-themed review; the further importance of mixed methods and questionnaires for data collection was revealed by the wider reviews of Bacca et al. (2014) and Altinpulluk (2019). Figure 3 depicts the number and percentage of research quality quartiles by research design and time.
The authors in our review mainly engaged with engineering Students (n=26, 72%), a mix of Students and Teachers (n=5, 14%) and Employees (n=2, 6%). The subjects of these studies where systems (n=7), civil (n=4), mechatronics (n=2), and cartographic (n=2) engineering university students, as well as electric (n=1) and industrial (n=1) engineering institute students, and mechanical (n=3), electronic (n=4), and mixed (n=8) engineering specializations students from various institutions. Confirming a wider pattern in the secondary literature, sample sizes in evaluations ranged from 5 [22] to 312 ([19]) subjects, but 55% of the evaluations fell between 30-60 subjects (e. g., Bacca et al. (2014) found most of the samples in their population to be between 30-200 subjects, while Sirakaya and Sirakaya (2018) placed the sample mean between 31-100 subjects). Besides this, we registered the educational year corresponding to subjects or programs as described in the literature, when possible (n=16). Figure 4, most authors worked with first-year students, but older students were also part of bigger sample sizes. Figure 5 shows that different  To explore this pattern, we observed the structure of research design quality and the main outcomes studied per design. Interestingly, almost 67% of the case presentations were placed either in the first or second quality quartiles (which can be interpreted as "optimal" and "good"), although this is true for just 36% of the texts that contain evaluations. No quasi-experimental design comes from a document deemed as "optimal" (considering that all the quasi-experimental studies included control groups, but only one was explicitly randomized, [18]). In parallel, 20%-33% of the literature was classified in the last quartile, irrespective of the research design followed. Secondly, we found that most quasi-experimental designs measured academic performance or less popular variables (spatial abilities in [19] and KPIs in [9]), compared with pre/post designs, that mainly focused on satisfaction measures, and crosssectional research, primarily interested in satisfaction and system performance measurement. This does not mean that these were the only exclusive possibilities, as shown in Table 10.

Design Features
On the following lines, we will describe the hardware and software listed in the literature. Most of the devices used by the literature were only Smartphones (n=23, 64%), or both Smartphone and PC/Tablets (n=7, 21%). The second most used device was the PC (n=3, 8%), followed by Tablets (n=1, 3%) and VR/AR mixes (n=1, 3%). Apparently, the found dominance of Smartphones in higher education is supported by the literature on STEM education-focused AR (Shirazi & Behzadan, 2015) as opposed to reviews that include other education levels. On the other hand, earlier reviews state that teachers lean towards Junaio, ARMedia, and ARToolkit for designing their AR-based activities (Diao & Shih, 2019;Sirakaya & Sirakaya, 2018). It seems that Latin-American AR-based educational programs rather depend on Vuforia (n=14, 39%), Aumentary (n=3, 8%), Unity (n=3, 8%), and ARToolkit-based (n=2, 6%) applications. A small group (n=5, 14%) even favored native applications, despite being a percentage fewer than the 43% reported by Ibáñez and Delgado-Kloos (2018); nonetheless, 63% (n=23) report or included some form of coding, including all applications based on Vuforia.
Diao and Shih (2019) stablish a difference between "general" and "specific purpose" AR software. Half of the applications reviewed by them where of "general" use (displaying text or graphics, or allowing 3D-object manipulation, for example), and the other half were of "specific use" (for object or architecture design, for example). On the other hand, drawing from literature about different education levels, Altinpulluk (2019) typified AR applications and found that most of them where 3D-Image based, Location-based, Video-based, games, or simulations and text based (from 17 overlapping types). In opposition to this literature, 71% (n=25) of the applications in our review were of "general purpose" and mainly 3D-Image (n=12), Text (n=7), Simulation (n=6), and Video-based (n=3) software. Most of the general purpose software where largely 3D-Image  (n=12), Text (n=4) and even Video-based (n=3), while the other group was composed by mainly Text-based (n=3), Simulation (n=3) and Robot mediated (n=2) software. Finally, no relevant differences were found between purpose and the use of native/non-native software. Figure 6 displays the distribution of documents according to software used while Figure 7 depicts the number of documents by type of content and app type.
The following paragraph describe additional AR software features in our engineering education literature. AR software based on marker or label recognition is predominant in the extant literature. We confirm this after finding 26 (72%) marker-based, 3 (8%) layer-based, and 2 (6%) object-recognition software. In the same vein, drawing from de Belen et al. (2019), we delimited a three-step interaction continuum for AR technology. Our results show that a big part of our AR technology in our sample only allowed Perception (n=16, 44%), some endorsed Annotation (n=6, 17%) and the rest where based on interaction by direct Manipulation (n=14, 39%). In addition to this, following the findings of Montoya et al. (2016), we coded the presence of dynamic content (n=15, 42%), as opposed to static content. Though all observed application types had some dynamic content-focused examples, dynamic contents were only predominant among all Location and Video-based as well as most Simulation apps (n=8, 22%). Finally, out of the 20 documents reporting both considerable and secondary use of 3D-Objects or images, most were created by the teacher (n=12), followed by those created by the class (n=5) and those downloaded or already part of the employed software (n=3).

Pedagogy
We coded the AR affordances and the main pedagogical perspectives linked with this technology. Saltan and Arslan (2016) suggested a seemingly useful categorization of three main AR pedagogical affordances. On the same line, AR in the reviewed literature afforded knowledge comprehension (n=17, 47%), concept development (n=14, 39%), and learning retention (n=5, 14%). Secondly, perhaps the pedagogical perspectives that frame educational practices linked with AR are more difficult to define. Despite the lack of consensus, we identified two favored cognitivist frameworks, CTML (n=13, 36%) and Mobile Learning (n=9, 25%), and three constructivist frameworks, Situated Learning (n=5, 14%), Experiential Learning (n=7, 19%), and Collaborative Learning (n=1, 3%) (Sommerauer & Müller, 2018). Examining the data, its easily seen that constructivist approaches favor AR concept development affordances in contrast with the other two. Interestingly, we also found a relationship between affordances and dynamic/static contents.
While most research engaged with engineering students, our literature populations pertained to a diversity of institutions: most of them to universities (n=25, 69%), some to technical schools (usually known as institutes, n=8, 22%), and a few to businesses (n=2, 6%). The latter were more prone to engage with a cognitivist framework, but half of the AR-related practices in institutes were constructivist. Besides, we analyzed teaching and academic evaluation practices related with AR, finding out that 78% were task-based and 69% (n=25) were problem-solving-focused activities (Diao & Shih, 2019;Wu et al., 2013). Following our analysis, we correspondingly saw the importance of technical schools for experimenting with more collaborative approaches to teaching (whether role or location-based) and evaluation activities (e. g., group or pair projects, peerbased work, etc.): most of the synchronic task-based activities ([30], [35], [34]) and the only group projectbased course ([21]) were done in these institutions. Even if this trend contrasts with the project-based pedagogy prevalent in other AR education contexts (Diao & Shih, 2019), the relationship between constructivism and collaborative learning became apparent when we saw that the only remote-based collaborative course found ([21]) was supported by an institute (de Belen et al., 2019). Figure 8 shows the number of documents by pedagogical perspective and post-secodary education institution.
Another way to look at this is to understand the kind of pedagogical experiences that students undergo when using AR. Following Chubukova and Ponomarenko (2018), these can be: modeling situations (n=14, 39%), acquiring skills (n=12, 33%), learning with textbooks or manuals (n=4, 11%), game-like experiences (n=3, 8%), and 3D object modelling (n=2, 6%). We saw that skill training and game-alike experiences are the only ones that partly support knowledge retention, however content acquisition is helped by all experiences except for game-alike, and concept development is only entirely absent of textbook/manual-based experiences. On the other hand, it is interesting to note that dynamic contents are a minority in all experiences, except for modelling. Figure 9 depicts the number and percentage of documents according to their educational affordances by pedagogical perspective and content type.
What are the main advantages and disadvantages of the use of AR in engineering education? In our review, most authors (n=15) agreed that AR motivated students (n=15), followed by those who valued an increase in academic achievement (n=11), the ease of use (n=9), innovativeness (n=6) and collaboration (n=2). Interestingly, more authors with ideas closer to experiential and situated learning report motivation benefits; whereas, among those reporting increases in academic achievement, the mobile learning framework is more common. In spite of a common consensus of AR being beneficial for academic achievement among other advantages (Akcayir & Akcayir, 2017;Bacca et al., 2014;Singh et al., 2019), a recent meta-analysis point towards the more nuanced conclusion that AR actually helps student engagements and abstract concept understanding (Garzón et al., 2019;Liono et al., 2021).
To conclude, virtually all authors mentioned an advantage, but less than half (n=16) mentioned disadvantages, namely heterogeneous benefits for different types of users (n=6), demanding technical requirements (n=6), accessibility issues related with skill gaps (especially among teachers and older professionals, n=5), the complexity of the setups used (n=2) and pedagogical insufficiencies (n=1). Both the lack of limitations and the complexity and technical problems have been found before in the AR literature (Akcayir & Akcayir, 2017;Bacca et al., 2014).

DISCUSSION
AR is a nowadays considered a mainstream tool for engineering education in Latin America (Hidrogo et al., 2020). Although this technology enhances important research, social and work-related skills in higher education (Klimova et al., 2018), questions about humanbased design, display technology, pedagogy and collaboration remain open (Billinghurst, 2021). In this work we reviewed conference papers and scientific articles published by Latin-American authors, focusing on AR uses in engineering education. Even though others reviewed experiences from different educational levels and disciplines, we tried to tackle many of the still open themes while only focusing on higher education.
One of the reasons to do this was to rethink the role of innovation to address the current knowledge gaps in the world. We found an increasing number of quality indexed conferences and a stagnant number of articles written by mainly Mexican, Colombian, and Brazilian authors. Even though most of the literature presented medium quality evaluations, different research designs seem to relate with corresponding sample sizes, variables measured and data collection techniques. At the same time, Latin-American engineering educators prefer conventional open-source AR software and Smartphone devices, incorporating some basic coding and 3D object modelling; however, we reported a big interest for manipulation and annotation based applications, as well as important object recognition software applications. Pedagogically, most university AR-related engineering programs and activities engage with cognitivist frameworks, but institutes seem to be embracing the emergence of constructivist and collaborative innovations. In general, authors highlight motivation academic achievement advantages, but overlook the disadvantages; when acknowledged, they emphasize accessibility and technical issues.
These findings integrate with the literature in two important ways. First, we can support the view that this literature leaves aside a needed focus on accessibility and longitudinal approaches (Bacca et al., 2014). Nonetheless, Latin-American authors, especially those affiliated with institutes, tackle, at least partly, collaboration, interaction issues and other largely overlooked UX design issues, as well as vocational learning, in a very intermingled way (Bacca et al., 2018;Ibáñez & Delgado-Kloos, 2018;Phon et al., 2014;Shirazi & Behzadan, 2015). These innovative authors seem likely interested in the motivational benefits of game and simulation-based learning (Ayer et al., 2016). Yet, contradicting earlier reviews, this trend is far from the mainstream. Our review also revealed a delay in evidence-based pedagogical practices, especially within universities: few authors seem interested in randomized controlled trials or mixed methods, and task-based evaluation practices within cognitivist pedagogies are still preferred over newer approaches.
We further believe to have shown the value of reviewing conference papers along with scientific articles. This helped us to learn about the importance of contextual factors before making assumptions about the advancement of Industry 4.0 technologies through ARbased engineering education . We think that Latin-American university educators, which are the greatest part of our sample, prefer to report conventional AR uses under cognitivist approaches, in contrast with other technologies and pedagogies, given the cost of Smartphones for their students, the limitations of their university budgets, the accessibility of open-source 3D object modeling and AR software, and the greater simplicity of conference formats in contrast with the demanding formats of international journals.
The limitations of the following review include proceeding without a pairwise quality assessment, applying a largely experimental quality assessment tool (and including some low-quality documents, due to the nature of scoping reviews), unavertedly or intentionally over-simplifying non-exclusive categories of certain variables, and having worked against time with an extensive number of research questions and variables. Future reviews should a) attend to relevant or influential pedagogical and/or technological innovations in engineering education in different global regions, b) discover the barriers for the adoption of such innovations by more precise literature review questions and informative methods (ranging from meta-analyses to multivocal reviews), and c) develop recommendations to better manage the knowledge production in different higher education institutions. Finally, we confirm the lack of longitudinal studies, the small quantity of correlational and experimental research, and very few direct references to qualitative methodologies, which justifies future additional research.

CONCLUSIONS
This scoping review shows that the accumulating Latin-American literature regarding the use of AR in engineering education is mostly pedagogically and technologically conservative, and that the research designs behind this literature are diverse but still limited. Nevertheless, we believe to have found a positive and emerging trend among institute-based engineering education. Moreover, using a literaturebased categorization, we found a diversity of application types and contents, contradicting the international trends in certain aspects, and even finding various direct mentions of software coding in all the literature. We also find that most advances are reported as mostly Scopusindexed conference papers, which is the only literature type in expansion.
We believe that these results inform the management of STEM education policies in the region. Knowledge gaps around the world, including those in research quality, are relevant to the diffusion of innovations in engineering education. Universities and teachers might consider accessibility and performance issues when trying out AR-based courses, but also should experience more with other pedagogies and forms of evaluation. Finally, future literature reviews might consider our solutions to the lack of representation of developing regions, as well as the differences between international patterns and locally-based phenomena.