Schools and educational systems are permeable to change and emerging social needs. For this reason, there has been an increase in the use of robotics and ML in primary school classrooms internationally in recent years. So much so that Sperling et al. (2022), following actor-network theory, point out that complex relationships develop in schools through the interaction between human and non-human actors. In fact, they warn that the latter play a crucial role in teacher performance, curriculum development, and routines, among other aspects.

ML has been used in various ways at this educational level. One trend in academic literature is the use of this technology to predict academic outcomes (Lillo et al., 2024), early school leaving (Musso et al., 2020), and student repetition rates (Constante Amores et al., 2024). Other authors (Alonso-Secades et al., 2022) have sought to improve the effectiveness of education systems in developing countries to provide evidence that could support decision-making from the policy level down to the classroom. Sperling et al. (2022) also used ML to reinforce mathematics learning in Swedish schools. Meanwhile, Villegas-Ch et al. (2024) developed an ML system that provided students with real-time visual and verbal feedback on their writing, which allowed early identification of difficulties (e.g., pressure, consistency, alignment) in writing in students aged 7–11 years.

From another perspective, educational robotics is characterized as a manipulative (Schina et al., 2025), accessible, and effective resource (Moreno-Palma et al., 2025). Its use is usually associated primarily with STEM subjects, as it has proven effective for learning mathematics and fostering students’ predisposition toward this and similar disciplines (Ruiz-Ortiz, 2023; Wu and Su, 2021). However, it is important to point out that, alongside an appropriate approach from the teacher, it facilitates interdisciplinarity (Wong, 2024), as demonstrated in an educational experience developed in the Azores (Portugal) by Santos et al. (2023). Another reason for its widespread use is its wide range of advantages, among which three stand out:

Despite its positive effects, Vera-Sagredo et al. (2024) highlight the ongoing debate within the educational community regarding how to integrate robotics as part of the official curriculum or through extracurricular activities. Franco Hidalgo-Chacón et al. (2022) state that robotics is one of the most popular cognitive extracurricular activities among students, and these findings align with those of Purković et al. (2022), who indicate that Croatian students support the inclusion of robotics in their country’s curriculum.

It is also worth highlighting some of the educational robotics initiatives that have been developed internationally in primary education. According to Moreno-Palma et al. (2025), the most widely used robotics kits are Bee-Bot, LEGO, and mBot. For example, Bee-Bot was used by Caballero González and García-Valcárcel Muñoz-Repiso (2020) in Spain, while Demetroulis et al. (2023) used the LEGO WeDo 2 kit. Experiences with other materials can also be identified, such as the application of CreaCube with primary education students in France (Schina et al., 2025). Finally, Santos et al. (2023) made various materials available to students and teachers, and they concluded that Botley was more widely used than Bubble due to its usability without requiring programming knowledge.

Training appears to be the main difficulty teachers face in implementing educational robotics in their classrooms. There is broad consensus in the academic literature regarding the lack of initial and ongoing teacher training in robotics (Moreno-Palma et al., 2025; Santos et al., 2023; Vera-Sagredo et al., 2024), which contrasts with the fact that teachers demonstrate a positive attitude, interest, and high motivation to integrate robotics into their classrooms. It is important to note that there is also a correlation between the predisposition to use robotics and teachers’ digital self-esteem (Gavrilas et al., 2024).

There are also other limitations that hinder the application of robotics in educational centers, such as the lack of resources and the excessive preparation time required from teachers (Gavrilas et al., 2025). Beyond these barriers, Santos et al. (2023) offer four general guidelines for implementing the use of educational robotics: (a) ensure continuity in these experiences—that is, ensure that they are not an anecdotal and sporadic activity; (b) their application should be gradual and progressive; (c) it is important to acquire and use diverse materials; and, finally, (d) primarily acquire materials that require a lower degree of programming knowledge to introduce students and teachers to this field.

The literature on higher education emphasizes that initial teacher training plays a crucial role in developing computational thinking and preparing future teachers to integrate programming, robotics, and, progressively, ML into their educational practices. Various studies focused on pre-service teachers have shown that experiences with programming and robotics provide suitable environments for fostering digital skills, positive attitudes, and a pedagogical understanding linked to computational thinking. Along these lines, Esteve-Mon et al. (2019) analyzed an intervention using educational robots that was designed using a design-based research methodology to promote computational thinking in student teachers. Their proposal combines unplugged, playing, making, and remixing activities, ultimately demonstrating the potential of incorporating hands-on robotics experiences into initial teacher training as a way to strengthen their digital teaching competence. Similarly, Villalustre and Cueli (2023) assessed the computational thinking of 164 prospective teachers using the computational thinking test (CTT); they identified significant differences based on gender and prior experience in robotics programming. Their results showed that men scored higher than women, and that prior experience in robotics programming is a determining factor in the level of development of computational thinking, thus providing relevant information for tailoring training programs to the diverse profiles of university students.

In addition to technical skills, the literature highlights the importance of motivational, affective, and cognitive factors in the disposition of pre-service teachers toward programming, AI, and ML. From an expectancy-value theory perspective, Weber et al. (2022) show that expectations of success and emotional costs influence the intention to teach computational thinking; university-level interventions can strengthen these expectations and values. Similarly, de Vink et al. (2023) demonstrate that peer-teaching experiences in higher education can improve students’ motivation, perceptions, and confidence in contexts related to computational thinking.

The specific field of ML has also begun to receive attention in the context of initial teacher training. Laru et al. (2025) provide relevant empirical evidence by analysing the background of AI literacy in prospective teachers. They found that elements such as knowledge of the ML process, attitudes toward AI, and perceived ability to use AI-powered applications are central components of this literacy. They emphasize that understanding the ML workflow (including data preparation, model training, and interpretation of algorithmic outputs) predicts prospective teachers’ ability to critically evaluate AI-based technologies and use them in an informed manner. Positive attitudes toward AI are also associated with higher levels of ML literacy, thus underscoring the importance of addressing both conceptual and affective/attitudinal aspects in university teacher training.

Along these same lines, Daher (2025) points out that initial teacher training must explicitly incorporate AI literacy; future teachers need a critical and ethical understanding of how AI-based technologies work in addition to technical skills. Daher warns that developing the ability to interpret and evaluate systems such as ML models is essential to avoid inequalities and ensure responsible educational integration, thus broadening the training focus beyond programming and robotics toward a comprehensive view of AI in education.

In parallel, training programs have been developed explicitly aimed at developing computational thinking in pre-service teachers. For example, Uzumcu and Bay (2020) propose a program based on interest-driven creator theory, composed of unplugged activities and computerized and robotic tasks, with the objective of improving the computational skills of pre-service teachers. In a broader context, Tsai et al. (2023) show that a programming course based on the GAME model (which integrates gamification, assessment, modeling, and inquiry) improves self-efficacy and performance in basic programming concepts among university students without a prior technological background, providing useful evidence for designing accessible and progressive learning experiences. On the other hand, studies focused on compulsory education, such as those by AUTHOR, demonstrate the potential of visual block programming and robotics to foster conceptual understanding, active student participation, and the development of computational thinking. Although situated in school contexts, their conclusions reinforce the need for future teachers to know and understand these methodologies to apply them appropriately in their professional practice.

These studies converge on the idea that initial teacher training in higher education should consider three interrelated axes. The first is the incorporation of practical proposals with educational programming and robotics, allowing future teachers to experience computational thinking and understand its formative potential (Esteve-Mon et al., 2019; Uzumcu and Bay, 2020; Villalustre and Cueli, 2023). The second is designing interventions that address motivational, affective, and cognitive-attitudinal dimensions, while considering expectations, values, perceptions, self-efficacy, and emotions associated with computational thinking and ML (de Vink et al., 2023; Laru et al., 2025; Weber et al., 2022). Finally, the third is the articulation of these experiences with curricular frameworks and empirical results in compulsory education, such that future teachers can establish links between programming and robotics practices and the learning of specific content in various areas of knowledge (Sáez-López et al., 2021; Wong, 2024).

The overall objective of this study is to analyze the impact of an intervention focused on the acquisition of computational concepts, robotics, and ML on prospective and practicing primary school teachers. The specific objectives are:

As randomization in the selection and assignment of participants to the groups could not be ensured, the study adopts a quasi-experimental design with a control group and an experimental group in both dimensions (Pérez Juste, 2006). Although such designs involve lower internal validity than true experimental designs, their applicability to naturally occurring groups makes them widely used in educational research (Tejedor, 2000).

The study was conducted in seven primary schools and at a Spanish public university within the primary education degree program. The study population consisted of undergraduate pre-service teachers enrolled in the first year of a Primary Education degree program at a public university and in-service primary school teachers working in public schools.

A non-probabilistic, purposive sample of 170 participants was used, including 83 first-year undergraduate pre-service teachers and 87 in-service primary school teachers from the city of Madrid. Participants were assigned based on the availability of educational centers and existing class groups rather than through random selection.

The undergraduate pre-service teacher group was composed of 92.8% female and 7.2% male participants, reflecting the gender imbalance typically observed in primary teacher education programs. Their mean age was 19.35 years, and the group was highly homogeneous in terms of age and educational background, as all participants were in their first year of university studies. The group of in-service primary school teachers consisted of 48.3% female and 51.7% male teachers, representing a more balanced gender distribution.

Data were collected using a coding, robotics, and machine learning knowledge test, along with several validated Likert-type scales to assess attitudes toward the curricular integration of these technologies. The instrument was structured into two dimensions, addressing cognitive and attitudinal aspects of teacher training.

Dimension 1 encompasses knowledge of ML, computational concepts, and robotics. The content validity of the CRMT instrument was evaluated considering the criteria of relevance and appropriateness by 11 expert judges, obtaining Aiken’s V coefficient values (V = S/[n(c − 1)]) greater than 0.75 for all items. Likewise, Cronbach’s alpha reliability coefficients exceeded 0.80, thus indicating an acceptable level of internal consistency (Hair et al., 1998).

Dimension 2, ML and educational robotics in teacher training, focused on the knowledge and understanding of practicing teachers and university students in the primary education degree program regarding computational concepts, educational robotics, and ML. The attitudes and perceptions of these groups toward the use of educational robotics and block programming in teacher training were analyzed using three Likert-type scales. Content validity was assessed qualitatively by 11 expert judges, considering the criteria of relevance and suitability of the instrument. Aiken’s V coefficient (V = S/[n(c − 1)]) values greater than 0.75 were obtained for all items. Cronbach’s alpha reliability coefficients were 0.918 for teachers and 0.911 for university students, exceeding 0.80, which indicates an acceptable level of internal consistency (Cronbach, 1951).

The project was carried out during the 2024–2025 academic year in seven schools, training practicing primary school teachers. The four sessions (see Appendix) focused on building a series of projects related to the micro:bit using the Maqueen kit. Temperature and light sensors were used with the micro:bit, along with the Visual Basic block-based programming environment using “micro:bit MakeCode” (see Figure 1). Participants programmed their projects and worked with basic computational concepts (sequences, loops, conditionals) using devices and robots.

Given the quasi-experimental nature of the study and the absence of random assignment, several strategies were implemented to control for potential extraneous variables that could threaten internal validity. First, both groups (undergraduate pre-service teachers and in-service primary school teachers) followed the same instructional intervention, with identical content, materials, duration, sequencing of activities, and learning objectives, thus minimizing instructional and procedural variability. Second, data collection instruments were standardized, validated, and administered under comparable conditions to reduce measurement bias. Although causal inference remains limited due to the use of naturally occurring groups, the combination of design consistency and statistical control contributes to reducing the influence of extraneous variables and enhances the interpretability of the results.

The same implementation and activities were carried out at the University of La Laguna with undergraduate students in the primary education teaching degree program. In the sessions, participants created basic programs using the micro:bit (dice, temperature sensor, light sensor), understanding block programming elements, basic computational concepts, sensors, and actuators. The Maqueen kit was also used with the micro:bit to build a robot that moves (actuators/motors) when light levels are high, simulating a solar car (Figures 1, 2). As seen in the conditional statement, if light levels are low, the motors stop.

Data analysis was conducted using IBM SPSS Statistics. Prior to inferential analyses, data were screened for missing values, outliers, and violations of statistical assumptions. Descriptive statistics (means, standard deviations, frequencies, and percentages) were calculated to summarize participants’ demographic characteristics, test scores, and questionnaire responses (Table 1).

To analyze differences in conceptual knowledge between undergraduate pre-service teachers and in-service primary school teachers, an independent-samples Student’s t-test was applied to the scores of the Coding, Robotics and Machine Learning Test (CRMT). Levene’s test was used to assess the homogeneity of variances; when p > 0.05, equality of variances was assumed. Statistical significance was set at p < 0.05, and Cohen’s d was calculated to estimate effect size, interpreted as small (0.20), medium (0.50), or large (0.80).

To examine predictors of CRMT performance, a multiple linear regression analysis was conducted, including group membership (pre-service vs. in-service teachers), gender, and teaching experience as independent variables. The model was evaluated using the F-test, R² and adjusted R², and standardized beta coefficients. Assumptions of normality, linearity, multicollinearity (tolerance and VIF), and independence of errors (Durbin–Watson statistic) were examined. Regression coefficients were considered statistically significant at p < 0.05.

For the analysis of Likert-scale questionnaire data, non-parametric Mann–Whitney U tests were used to compare groups due to the ordinal nature of the data. A conservative significance level of p < 0.01 was adopted, and effect sizes were calculated using r, interpreted as small (0.10), medium (0.30), or large (0.50). Relationships between variables were explored using Spearman’s rank-order correlation coefficients, with only correlations of ρ ≥ 0.50 and p < 0.01 reported.

The CRMT was administered to assess knowledge of ML, coding, and robotics; the average scores of university students and practicing teachers who had worked with the described training implementation were then compared. The average score for university students was 6.31 out of 10, while the teachers obtained an average of 5.59 out of 10. Statistical comparison allowed us to verify whether there were significant improvements between the two groups after the intervention. The scores on the 10-item test emphasize the importance of an educational design that includes visual programming languages, as well as the implementation of educational robotics and ML to foster the acquisition of computational concepts. Given that the significance value of Levene’s test is p = 0.053 (>0.05), equality of variances is assumed. The Student’s t-test revealed statistically significant differences between the groups (p = 0.002), with higher scores in the university student group. The effect size was small to medium, with a Cohen’s d value of 0.441.

Multiple linear regression was performed to predict test_CRMT scores based on teaching experience (exp_docente), gender (Genero), and group membership (1 = Teachers; 2 = Students). After excluding cases with missing data, the analysis was performed on N = 170. The pooled model was found to be significant, F (3,166) = 3.19, p = 0.025, and explained 5.4% of the variance in test_CRMT (R² = 0.054; adjusted R² = 0.037). Although the overall effect size was small, the model achieved statistical significance (see Table 2). Of the included predictors, only the group (teachers vs. students) variable was significant (B = 1.132, β = 0.337, t = 2.58, p = 0.011), which indicates that, controlling for experience and gender, students scored on average 1.13 points higher on the CRMT than teachers. The coefficients for teaching experience (B = 0.025, p = 0.345) and gender (B = 0.199, p = 0.530) did not reach statistical significance in the model (see Table 2).

Standardized residuals remained within the acceptable range (−2.82 to 2.29), and no extreme outliers were detected that would compromise the assumptions. Collinearity tests (tolerance and VIF) showed no serious problems: maximum VIF ~ 2.99. The Durbin–Watson statistic = 0.82 suggests some autocorrelation in residuals that should be considered in future models (e.g., hierarchical modeling or inclusion of covariates that capture temporal or group structure). Thus, belonging to the student group (vs. teachers) predicted a higher score on test_CRMT with a moderate standardized effect (β = 0.337), while neither teaching experience nor gender explained relevant added variance in the presence of the group effect.

On average, university students scored higher than teachers on most items, particularly in Sections 1 and 2 of the questionnaire, which relate to knowledge of ML and programming, and computational concepts (Table 3). In Section 1, these differences were observed in programming knowledge (1.1), knowledge of ML (1.2), and perceived importance of ML (1.3). Furthermore, students achieved higher scores in block programming (1.5), with a high standard deviation in both groups, which reflects the significant heterogeneity in experience with this type of tool, especially among practicing teachers. In basic computational concepts, this trend is maintained in knowledge of sequences (2.1), loops (2.2), and conditionals (2.3); we can highlight, here, the case of loops, where, in addition to a higher mean, students showed less dispersion (SD = 0.75) than the teachers (SD = 1.02), which indicates greater homogeneity in the mastery of this concept among future teachers.

Regarding curriculum application, students achieved higher scores in robotics learning (3.1), block-based programming content (3.2), and art and music (3.6), while teachers obtained a slightly higher average in natural sciences (3.4). Finally, the assessment of the inclusive potential of programming (3.9) was high in both groups, although students showed a higher average and less dispersion, which demonstrates greater agreement in the perception of programming as a resource for educational inclusion.

Tables 4, 5 show the strongest correlations (ρ ≥ 0.50; p < 0.01) between the questionnaire variables for university students and practicing teachers. In both groups, a high degree of consistency was observed in basic programming knowledge, with very strong associations between sequences, loops, and conditionals (e.g., in teachers: ρ = 0.781, ρ = 0.697, and ρ = 0.819; in students: ρ = 0.760, ρ = 0.696, and ρ = 0.628). Likewise, in both samples, strong associations were identified between educational innovation based on programming and robotics and learning with robotics, as well as with its application in curricular areas, especially in the natural sciences and social studies.

Among teachers, these relationships show high values between educational innovation and mathematical and scientific learning (e.g., ρ = 0.663 with mathematics and ρ = 0.575 with natural sciences), as well as with social studies (ρ = 0.567). Among students, block-based programming learning is strongly associated with content work in primary education (ρ = 0.558). In the motivational and inclusive sphere, strong relationships are observed between interest in coding, active student participation, and attention to students with special educational needs, both among teachers (ρ = 0.604 between interest and attention to special educational needs; ρ = 0.518 between interest and participation) and students (ρ = 0.579 between interest and participation; ρ = 0.533 between participation and attention to special educational needs). The pattern of correlations among practicing teachers is more consistent between educational innovation based on programming and robotics and the various curricular areas, while among university students, the associations between block-based programming, content teaching in primary education, and student engagement are particularly prominent.

Regarding curricular areas, both groups demonstrate a significant association between social studies and art education, which is explainable by their cultural, historical, and humanistic connections. Likewise, the relationship between the natural sciences and social studies is evident, highlighting an interrelationship between the different areas of knowledge and allowing for the interpretation of an interdisciplinary and interconnected educational approach. Methodological approaches such as holistic approaches, project-based learning, and integrated subject areas could thus facilitate the incorporation of technological elements from a cross-curricular perspective. Finally, it is observed that there is a relationship between educational innovation and the use of game engines among university students, a relationship not found in the group of practicing teachers.

The Mann–Whitney U test showed statistically significant differences between teachers and university students in 10 of the variables analyzed (Table 6). From a rigorous and conservative perspective, a significance level of 0.01 (p < 0.01) was adopted. In all cases, the average ranks were higher in the university student group, which reflects a greater self-perception of technological competence and the educational potential of programming and robotics in this group. The greatest differences were observed in programming game engines for educational purposes (U = 2386.50), with a medium effect size (r = 0.31). Significant differences were also found in the perceived need for programming instruction, knowledge of ML, the importance of ML in primary education, and the use of block-based programming in practice, all with small to medium effect sizes (r ≈ 0.23–0.25).

In the curricular field, university students obtained higher scores in the teaching of content through block programming, in the learning of art and music through programming and robotics, as well as in the perception of providing help to students with special educational needs, with effect sizes between r = 0.22 and r = 0.26. Differences were also observed in the understanding of computational concepts, in the knowledge of the concept of ML and in its importance for learning in primary education (Items 1.1, 1.2, and 1.3), as well as in the greater use of visual block programming compared to practicing teachers (Item 1.5).

Finally, university students assigned greater value to the innovative implementation of game engines, block-based programming in primary education, support for students with special educational needs, and the contributions of programming to art education (Items 2.5, 3.2, 3.6, and 3.9). In the other curricular areas analyzed, no statistically significant differences were observed between the two groups, which indicates a similar assessment of the technological implementation.

A central finding is that undergraduate pre-service teachers achieved significantly higher scores than in-service teachers on the Coding, Robotics and Machine Learning Test (CRMT), particularly in basic computational concepts (sequences, loops, and conditionals), programming logic, educational robotics, and ML. This result suggests that recent exposure to formal learning environments and structured training in digital technologies may play a key role in conceptual understanding.

A plausible explanation lies in the proximity of university students to assessment-oriented learning contexts and their recent engagement with formal instruction, which may facilitate test performance. This interpretation aligns with research on “teaching to the test” effects (Zakharov, 2021) and with the protégé effect, whereby learning with the intention of teaching enhances conceptual encoding and retention (Fiorella and Mayer, 2013). The absence of a significant effect of teaching experience is consistent with evidence indicating that professional experience does not necessarily translate into higher performance on theoretical or conceptual assessments (Kini and Podolsky, 2016).

These findings are consistent with previous studies showing that early and systematic exposure to visual programming and robotics positively influences computational thinking development (Wei et al., 2021; Wu and Su, 2021; Noh and Lee, 2020).

The multiple linear regression analysis indicated that group membership (pre-service vs. in-service teachers) was the only significant predictor of CRMT performance, while gender and teaching experience were not significant. Although the effect size was moderate, the explained variance was low, suggesting that performance in computational concepts and ML is influenced by multiple interacting factors, such as motivation, learning strategies, prior training, and cognitive styles (Brydges, 2019).

This finding reinforces the idea that formal training opportunities, rather than accumulated teaching experience alone, are critical for developing emerging digital competencies. It also highlights the need for continuous professional development programs that explicitly address robotics, programming, and ML for in-service teachers, as emphasized in recent reports on educational innovation and digital transformation (Robert et al., 2025).

Another relevant finding concerns the higher self-perception of technological competence among undergraduate pre-service teachers, particularly regarding block-based programming and the use of game engines for educational purposes. University students reported greater confidence and motivation to integrate game-based digital tools into teaching, which may reflect their greater familiarity with contemporary digital environments and interactive media.

These results are consistent with prior research showing that game-based learning environments enhance computational thinking, motivation, and active participation (Giannakoulas and Xinogalos, 2024). Studies combining Scratch, robotics, and game engines in teacher education have similarly reported improvements in both technical skills and innovative pedagogical design (Rich et al., 2022; Sáez-López et al., 2023).

In contrast, in-service teachers demonstrated a more consolidated pattern of curriculum-oriented integration but lower confidence in advanced digital tools, reflecting a persistent training gap identified in the literature (Moreno-Palma et al., 2025; Robert et al., 2025).

The results also revealed strong associations between basic programming knowledge, educational robotics, and pedagogical innovation in both groups, as well as between robotics use and its application across curricular areas, particularly mathematics, natural sciences, and social studies. These relationships support an interdisciplinary view of computational thinking, as proposed by Wing (2006).

Notably, undergraduate pre-service teachers assigned higher value to the use of programming and robotics in artistic and musical learning and in supporting students with special educational needs. This finding reinforces the inclusive and cross-curricular potential of these tools, in line with studies highlighting their effectiveness in promoting engagement, creativity, and participation when appropriate guidance and feedback are provided (Chevalier et al., 2022; Wu and Su, 2021).

From a theoretical perspective, these results align with Papert’s constructionism (Papert, 1980), which emphasizes learning through active creation, problem-solving, and interaction with meaningful technological artifacts.

The findings of this study have several educational implications. First, they highlight the importance of strengthening initial teacher education through systematic, hands-on experiences with block-based programming, educational robotics, and introductory ML. Early exposure appears to foster both conceptual understanding and positive attitudes toward technological integration.

Second, the results underscore the need for targeted professional development programs for in-service teachers that address emerging technologies beyond basic digital literacy, particularly ML and game-based programming environments. Such programs should adopt progressive, structured, and pedagogically grounded approaches to reduce the identified training gap.

Finally, the strong links observed between programming, student engagement, interdisciplinarity, and inclusion suggest that robotics and ML can serve as powerful tools for promoting active, inclusive, and innovative teaching practices across the primary education curriculum.

This study has several limitations that should be considered. First, the use of non-probabilistic purposive sampling, as well as the assignment of groups based on pre-existing groupings and availability in educational centers, may restrict the generalizability of the results and generate potential selection bias. Although equivalence between groups was verified, the absence of a randomization process limits the possibility of establishing robust causal inferences. Future research should consider random or stratified sampling designs to strengthen the external validity of the findings.

Second, the marked gender imbalance in the sample, characteristic of teacher training programs, could influence the interpretation of the results, especially those related to attitudes toward programming and robotics. While this distribution accurately reflects the current student body composition, it hinders the analysis of potential gender-related differences in response to the intervention. Future studies could benefit from more balanced samples or from a specific analysis of the interaction between gender and learning outcomes in educational robotics and ML contexts.

Finally, the research was conducted in a limited number of institutions and during a relatively short intervention period, which restricts the evaluation of both the long-term retention of learning and its transfer to actual teaching practice. Longitudinal and multicenter studies are recommended to validate and expand upon the results obtained in this research.

Based on the findings of this study, several recommendations are proposed for teacher education, educational practice, and future research.

First, initial teacher education programs should systematically incorporate visual block programming, educational robotics, and introductory machine learning as core components of the curriculum. These experiences should be progressive, starting with basic computational concepts and gradually advancing toward more complex applications, including game-based learning and ML-supported activities. Early and structured exposure can strengthen both conceptual understanding and positive attitudes toward technological integration.

Second, professional development programs for in-service teachers should be expanded and updated to address emerging technologies, particularly machine learning and educational robotics. Training initiatives should emphasize hands-on, practice-oriented approaches aligned with classroom realities, enabling teachers to translate technological knowledge into meaningful pedagogical applications across different curricular areas.

Third, educational institutions and policymakers should support interdisciplinary and inclusive uses of programming and robotics, promoting their integration beyond STEM subjects into areas such as the arts, social sciences, and special education. Providing adequate resources, time, and institutional support is essential to ensure sustainable implementation.

Finally, future research should adopt longitudinal and experimental designs to examine the long-term effects of robotics and ML interventions on teaching practice and student learning. Studies with larger and more diverse samples are recommended to enhance generalizability and to explore additional factors influencing teachers’ technological competence, such as motivation, self-efficacy, and prior experience.