Introduction

Robot-based education has gained significant support in the educational landscape for a multitude of compelling reasons. The outbreak of COVID-19 has forced millions of learners to transition to remote education, during which robot-based education has been widely used. After the COVID-19 pandemic, many countries are seeking digitalized pedagogical approaches to keeping social distance and enhancing educational quality. Digitalization of teaching and learning requires robot-based education to support the teaching and learning processes (Mishra et al., 2013). Robot-based education has enjoyed broad acceptance in both informal and formal educational contexts, encompassing various educational levels ranging from kindergartens to universities (Tselegkaridis & Sapounidis, 2022). Moreover, the widespread acceptance of robot-based education within educational spheres can be attributed in part to the constructivism philosophy.

The philosophy of robot-based education is shaped by the theory of constructivism. Knowledge was accumulated through interactions with peers, teachers, learning resources, and educational environments (Huitt & Hummel, 2003). The learning effectiveness could be improved if computer technologies and real practice were integrated into the learning process (Papert, 1980). The interactions with robots could leave impressions on human brains, which enhance the organic structures and systematize the obtained knowledge. In this way, the connections between the real world and the internal knowledge structures could be established, paving a way for the successive acquisition of knowledge. While a substantial amount of research has investigated the effectiveness of robot-based education, relatively few studies have compared the impact of robot-based education on educational outcomes with traditional methods.

In the aftermath of COVID-19, despite the widespread adoption of robotic platforms in education, only a handful of studies have conducted systematic cross-comparisons of their effectiveness against traditional methods across different global contexts. This study aims to partially bridge the research gap in the literature by pooling the effects of robot-based education on various dimensions. Specifically, the study attempts to identify the effectiveness of robot-based education, compared with non-robot-based education, through a meta-analysis and systematic review. Through inclusion and exclusion criteria, a certain number of studies will be included for the meta-analysis. This study will then pool and compare the effects of the robot-based education on academic achievements, computational knowledge, motivation, and learning performance.

Literature review

Overall robot-based educational outcomes

Robot-based education has the potential to effectively enhance students’ learning outcomes in multiple ways. For instance, robot-assisted coding programs could effectively enhance preschoolers’ reasoning skills in mathematics (Somuncu & Aslan, 2022). Telepresence robots were able to train patients remotely and safely and elevate professional skills (Ooms et al., 2021). Robot-based education could boost the number of correct responses, improve problem-solving skills, and enhance students’ interest in mathematics learning (Alfieri et al., 2015). Students’ confidence, attitude, and perception towards education improved significantly when assisted with robots. Teachers found robots could help connect classroom learning in Science, Technology, Engineering, and Mathematics (STEM) to real-world use (Knop et al., 2017).

Robot-based education could enhance educational outcomes such as learners’ thinking abilities, spatial visualization abilities, and programming skills (Wang et al., 2023). A game-based robot-based education simulator could significantly improve negotiation skills, facilitate inquiry learning outcomes, and enhance reflective thinking abilities compared with a conventional robot simulator among university students (Lee et al., 2013). The LEGO-based education could significantly improve elementary school students’ reflective thinking skills of problem-solving and spatial visualization abilities compared with those without the assistance of LEGO, an educational robot (Koca & Çakir, 2021). Programmable robots could significantly improve spatial relations and mental rotation skills, although they could not significantly improve visual memory among children (Brainin et al., 2021). Robot-based education could increase students’ knowledge of programming, which was not related to their previous learning experiences (Fernández-Llamas et al., 2017). Robot-based education could improve the basic programming knowledge of students (Titon et al., 2018). In general, robot-based education has the potential to exert a positive influence on learners’ thinking abilities, spatial visualization abilities, and programming skills.

Robot-based education could improve learners’ attitudes, satisfaction, problem-solving skills, and mood. For instance, robot-based coding teaching could effectively boost the problem-solving skills of pre-schoolers (Çakır et al., 2021). The robot-based camp could improve students’ attitudes and perceptions towards STEM education (Üçgül & Altıok, 2021). Virtual reality human-robot interaction technology could increase the satisfaction of secondary school students (Lo et al., 2022). Students held positive attitudes towards the use of a Robot Rabbit in English vocabulary learning, and the interaction with the robot rabbit could create a favorable mood for learners (Eimler et al., 2010). Therefore, robot-based education appears to exert a positive influence on learners’ attitudes, satisfaction, problem-solving skills, and mood.

Academic achievements

The robot-based educational approach can outperform traditional methods in improving academic achievements. The robot-based education program “TangibleK” could substantially boost the academic achievements of secondary school students (Caballero-Gonzalez & Garcia-Valcarcel, 2020). Robot-based education could significantly outperform traditional methods in improving secondary school students’ academic achievements (Kert et al., 2020). Robot-based coding education could improve the academic achievements of preschoolers compared to traditional methods (Turan & Aydoğdu, 2020). However, inconsistent findings were revealed in terms of the effect of robot-based education on academic achievements. A humanoid robot could not significantly increase the academic achievements of nutritional knowledge acquisition among children (Rosi et al., 2016). It is thus necessary to further explore the effect of robot-based education on academic achievements.

Computational knowledge

Robot-based education is particularly effective in improving computer-related skills. For example, unplugged programming could greatly improve the computational knowledge of junior high school students regardless of their backgrounds (Sun et al., 2021). Robot-based education without computers could significantly improve the computational knowledge of children (Chou & Shih, 2021). Robot-based education based on the 6E model could significantly improve students’ computational knowledge and practical skills compared with the traditional methods (Hsiao et al., 2019). The programming robot-based education could significantly improve the computational knowledge of elementary school students (Chiazzese et al., 2019). Robot-based education could lead to a significantly higher level of computational knowledge than the non-robot-based method among middle school students (Kert et al., 2020). The robot-based education program “TangibleK” could greatly improve secondary school students’ computational knowledge (Caballero-Gonzalez & Garcia-Valcarcel, 2020). Thus, robot-based education generally improves computational knowledge.

Motivation and performance

The robot-based method tends to outperform traditional methods in improving motivation and performance. Robot patients could significantly improve the training skills of nursing students and nursing teachers who held positive attitudes toward robot patient-assisted education (Huang et al., 2016). The interactive robots could greatly improve students’ learning performance, enhance their interest, and help them concentrate on learning tasks (Hakim et al., 2020). Robot-based education could increase high school students’ troubleshooting abilities and improve their academic achievements and learning performance (Zhong & Li, 2019). Robot-based education could significantly outperform the non-robot-assisted approach in improving elementary school students’ motivation and learning performance of English as a foreign language (Hong et al., 2016). Robot-based education based on the 6E model could significantly improve students’ motivation and performance (Hsiao et al., 2019).

A wealth of research suggests that robot-based education holds remarkable potential for significantly enhancing both student motivation and academic performance. Children with programming interest held significantly higher interest, motivation, and self-efficacy in robot-based education (Master et al., 2017), leading to favorable learning performance. The robot-based discovery approach could successfully measure and improve the academic achievements, learning performance, attitudes, and motivation in math learning of elementary school students (Casad & Jawaharlal, 2012). Robot-based education could significantly improve the learning performance of children (Benvenuti & Mazzoni, 2019). Robot-based teaching could achieve significantly higher academic achievements, motivation, collaboration, interactions, and teaching performance than the non-robot-based method among higher school students (Marin-Marin et al., 2020). Robot-based education could improve spatial abilities and the learning performance of elementary students (Julia & Antoli, 2016). Thus, robot-based education is associated with positive predictive power for learning performance and motivation.

Research questions and hypotheses

Based on the review of the literature, this study seeks to determine the effects of robot-based education on educational outcomes compared with traditional methods. Researchers proposed the alternative hypotheses as follows:

H1. Robot-based education is associated with moderate-to-large improvements in overall educational outcomes compared to traditional methods.

H2. Robot-based education is associated with moderate-to-large improvements in academic achievement compared to traditional methods.

H3. Robot-based education is associated with moderate-to-large improvements in computational knowledge compared to traditional methods.

H4. Robot-based education is associated with moderate-to-large improvements in learning motivation compared to traditional methods.

H5. Robot-based education is associated with moderate-to-large improvements in learner performance compared to traditional methods.

Methods

This study conducts a meta-analysis and systematic review in accordance with the protocol of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Moher et al., 2009).

Literature search

Researchers extracted data from multiple databases on June 12, 2022, and March 5, 2025. They obtained a total of 400 results from the Core Collection of Web of Science (CCWOS) by entering “robot* AND education (topic)” and “control group* (topic)” in the search command. CCWOS covers Science Citation Index Expanded–1900-March 5, 2025, Science Citation Index Expanded–1900-March 5, 2025, Social Sciences Citation Index –1900-March 5, 2025, Social Sciences Citation Index–1900-March 5, 2025, Arts & Humanities Citation Index–1975-March 5, 2025, Arts & Humanities Citation Index –1975-March 5, 2025, Conference Proceedings Citation Index – Science–1990-March 5, 2025, Conference Proceedings Citation Index – Science–1990-March 5, 2025, Conference Proceedings Citation Index – Social Science & Humanities–1990-March 5, 2025, Conference Proceedings Citation Index – Social Science & Humanities –1990-March 5, 2025, Book Citation Index – Science–2005-March 5, 2025, Book Citation Index – Science–2005-March 5, 2025, Book Citation Index – Social Sciences & Humanities –2005-March 5, 2025, Book Citation Index – Social Sciences & Humanities –2005-March 5, 2025, Emerging Sources Citation Index–2017-March 5, 2025, Emerging Sources Citation Index–2017-March 5, 2025, Current Chemical Reactions –1985-March 5, 2025, Current Chemical Reactions–1985-March 5, 2025, and Index Chemicus–1993-March 5, 2025.

They retrieved 64 results from ScienceDirect (Elsevier) by typing “control group” in “Find articles with these terms” and “robot AND education” in “Title” for the period ranging from 1983 to 2025 on March 5, 2025. The obtained article types included research articles (n = 58), book chapters (n = 2), conference abstracts (n = 3), and other (n = 1). Major publication titles included IFAC Proceedings Volumes (n = 13), Computers & Education (n = 5), IFAC-PapersOnLine (n = 5), Procedia Computer Science (n = 3), Journal of the American College of Surgeons (n = 2), and Computers in Human Behavior (n = 2). They obtained 9 results from EBSCOhost by keying in “TI (robot AND education) AND AB control group” on March 5, 2025, ranging from 2010 to 2025. They obtained 29 results from Scopus by keying in “(TITLE (robot AND education) AND TITLE-ABS-KEY (control AND group))” on March 5, 2025, for the time span ranging from 2010 to 2025.

A total of 62 records were discovered in EI Compendex for the search (((robot AND education) WN TI) AND (((control group)) WN KY)) on March 5, 2025, ranging from 1987 to 2025. They obtained 5 results by entering [All “control group”] and [Title robot education] into Sage on March 5, 2025, for the years ranging from 1990 to 2025. The major subjects included Engineering & Computer (n = 2) and Education (n = 3). They retrieved 35 results from Springer on March 5, 2025, by entering “control AND group” in “keywords” and “robot education” in “title”. They retrieved 5 results from Taylor & Francis Online on March 5, 2025, using the search: [Publication Title: robot education] AND [Abstract: control group]. Finally, a total of 609 results were gathered from multiple databases to experience both inclusion and exclusion processes for the meta-analysis.

The inclusion and exclusion criteria

Researchers formulated both inclusion and exclusion criteria to filter the obtained studies. They included the studies if they (1) fell within the scope of robot-based education, (2) focused on the effects of the robot-based education on academic achievements, computational knowledge, motivation, and learning performance, (3) were randomized controlled trials and divided participants into control and experimental groups, and (4) could provide sufficient data (e.g., means, SDs, sample sizes) for the meta-analysis. They were excluded if they (1) were duplicated records, (2) were withdrawn publications or corrections, (3) were classified as editorials, news, or data, (4) fell outside the scope of robot-based education, (5) were of inferior quality, (6) did not provide abstracts, (7) recruited a small sample size, (8) failed to provide a clear control condition or randomized procedure, (9) arrived at unconvincing conclusions, or (10) failed to provide enough data for the meta-analysis. They ultimately selected 36 results for the meta-analysis (Fig. 1).

Fig. 1: The inclusion and exclusion processes.
figure 1

This is a flow chat showing the specific steps to include and exclude the literature for the meta-analysis.

Full size image

The exclusion of studies with small sample sizes or poor design aims to guarantee scientific and reliable results. Scientific research requires rigorous methods, reasonable sample sizes, and transparent designs to boost reproducibility. We defined the small sample sizes based on statistical calculation formulas in different fields. For instance, the research on genes might call for 10 thousand participants, while a qualitative interview might need several interviewees. We prefer randomized-controlled studies to transactional research and case studies. We also assess the quality based on the framework established by the American Educational Research Association, encompassing theoretical framework, research design, data quality, ethics, and academic contribution.

The rationales for pooling diverse studies include increasing the sample sizes to enhance the representativeness of the included studies and reduce sample bias, boosting stability and generalizability, and comparing diverse studies to obtain convincing pooled effect sizes. Individual studies may not reveal the potential information, which can be uncovered by a meta-analysis where diverse studies are collected, analyzed, and compared. In this way, the meta-analysis can thoroughly explore the previous literature to pool the effect sizes and draw a generalized conclusion.

Data coding

Researchers extracted data from the included studies to gather related information for the meta-analysis. Two independent raters extracted the information using the content analysis method (Hsu et al., 2013). Researchers employed Cohen’s kappa statistics to measure the inter-rater reliability (Cohen, 1968). The inter-rater reliability reached a satisfactory level (k = 0.95). If both reviewers could not reach any agreement on any selection, a third examiner would be invited to make a final decision. The extracted information comprised author names, publication years, means, standard deviations, and numbers of participants in both control and experimental groups, subgroup, country, treatment, journal titles, and sample sizes (Table 1).

Table 1 Details of included studies.
Full size table

Before the calculation of effect sizes

Before computing effect sizes, researchers assessed the heterogeneity of estimated results via I2, Heterogeneity statistics, and values of z and p. The heterogeneity was considered negligible in case of 0 ≦ I2 ≦ 40%, moderate in case of 30% ≦ I2 ≦ 60%, substantial in case of 50% ≦ I2 ≦ 90%, and considerable in case of 75% ≦ I2 ≦ 100% (Higgins & Green, 2021). If the value of I2 exceeded 50%, researchers would adopt a random-effect model to conduct the meta-analysis, and a fixed-effect model would be utilized if the value of I2 was below 50%. Researchers also tested the stability or robustness via a sensitivity analysis and examined publication bias using both Begg’s (Begg & Mazumdar, 1994) and Egger’s tests (Egger et al., 1997) and the trim-and-fill analysis.

Calculation of the effect size

The meta-analysis seeks to quantitatively combine the effects of robot-based education on overall educational outcomes, academic achievements, computational knowledge, motivation, and learning performance compared with non-robot-based education. The effect sizes were computed through Hedge’s g based on the guidelines proposed by Taylor and Alanazi (2023). The effect sizes were computed by dividing the differences between means in experimental and control groups by the pooled standard deviations across both groups (Sedgwick & Marston, 2013). Hedge’s g justifies the effect sizes from small sample sizes and is thus deemed more reliable and stable than Cohen’s d. The calculation formula is:

$${S}_{p}=\sqrt{\frac{({n}_{1}-1){s}_{1}^{2}+({n}_{2}-1){s}_{2}^{2}}{{n}_{1}+{n}_{2}-2}}$$

where n1 and n2 refer to the sample sizes of experimental and control groups, respectively, while s1 and s2 refer to their standard deviations. If the sample is too small, there may exist bias regarding the value of Hedge’s g, which will be adjusted through the formula: \(g\ast =g\times (1-\frac{3}{4d\,f-1})\), where df = n1 + n2 − 2.

The effect size was very small in case of around 0.1, small in case of around 0.2, medium in case of around 0.5, large in case of around 0.8, very large in case of around 1.2, and huge in case of around 2 (Sawilowsky, 2009).

There were different research designs in the included studies. The first design entailed dividing participants into both control and experimental groups, each of whom was evaluated before and after the experiment. This design allowed for more precise calculation of effect sizes and improved internal reliability (Morris, 2008). The second design did not include a pretest. Instead, participants were assigned to both experimental and control groups in a post-test. The third design involved assessing the participants before and after the experiment in a controlled condition. For pretest-posttest and posttest-only designs, we calculated the difference between pre- and post-tests, which was then divided by the standard deviation for the effect size (Hedge’s g). The last two designs yielded similar values since researchers combined both small and large sample sizes to obtain pooled effect sizes. Results of different research designs could be combined in case that the effect size estimated the same experiment, that all effect sizes could be pooled into a general value, and that the meta-analysis could accurately estimate the effect sizes (Morris & DeShon, 2002).

Methodological limitations

There are several limitations in the methodology. The method of meta-analysis is vulnerable to publication bias since the studies with positive results are more likely to be published than those with negative results. The included studies may exhibit heterogeneity in designs, treatments, participants, and sample sizes. The diverse research designs and inconsistent robot implementations may inflate effect sizes, limiting generalizability. Languages not written in Chinese and English may be inaccessible to the authors since they can only read English and Chinese. We find it difficult to specify the subgroup studies in terms of age group, subject areas, and cultural contexts because the number of studies falling into the same variables is limited. We thus focus on the comparison between robot-based education and traditional methods using a meta-analysis and systematic review.

Results

Testing publication bias

Publication bias tends to occur in meta-analytical analyses since positive results are more likely to be accepted and published than negative ones. Researchers tested publication bias using Begg’s and Egger’s tests, non-parametric trim-and-fill analysis of publication bias, and the funnel plot (Egger et al., 1997; Begg & Mazumdar, 1994). We obtained the results of testing publication bias using both Begg’s and Egger’s Tests. Begg’s test indicated the absence of publication bias (z = 0.99, Pr > |z | = 0.3239). Egger’s test also showed the absence of publication bias (z = 1.18, Pr > |z | = 0.2372).

The funnel plot also showed no presence of publication bias. In the funnel plot, the x-axis represents the estimates of standardized mean differences, while the y-axis represents the standard errors of standardized mean differences. A dot represents an individual study. The nearly symmetric distribution of the dots along both sides of the middle suggests that there is no presence of publication bias in the estimates of standardized mean differences (Fig. 2). For the non-parametric trim-and-fill analysis of publication bias, the number of studies is 62, where the number of observed studies is 48, while the number of imputed studies is 14. Both observed studies (g = 0.711) and observed + imputed studies (g = 0.985) have reached a level of large effect sizes. Therefore, it was concluded that there was no presence of publication bias in the included studies in this meta-analysis.

Fig. 2: A funnel plot to test publication bias.
figure 2

This is a funnel plot to show publication bias of the obtained literature, where one dot indicates the effect size of an individual study.

Full size image

A sensitivity analysis

To evaluate the robustness and stability of the estimates, researchers carried out a sensitivity analysis. They performed a leave-one-out meta-analysis using Stata 18.0, and the results were presented in Fig. 3. As depicted in Fig. 3, each dot corresponds to an individual study. The left line means the lower bound of the confidence interval, while the right one represents the upper bound of the confidence interval. The middle line indicates the pooled effect sizes, i.e., standardized mean differences. The analysis revealed that all the pooled effect sizes fell within the 95% confidence intervals (0.4–1.0). This indicates that the estimates are stable and robust, given that any study is omitted.

Fig. 3: A sensitivity analysis to test the robustness and stability.
figure 3

This is a diagram to show the sensitivity analysis results where effect sizes and 95% confidence intervals are displayed given one study is omitted.

Full size image

Researchers utilized I-squared values and the Heterogeneity statistics to measure the degree of heterogeneity. To determine whether to apply random or fixed-effect models to the meta-analysis, researchers identified the heterogeneity of the estimates by calculating the values of I-squared, Tau-squared, and the Heterogeneity statistic (Fig. 4). As shown in Fig. 4, the I-squared values of the estimates of achievement (z = 3.38, p < 0.01), computational knowledge (z = 5.53, p <0.01), motivation (z = 3.26, p = 0.002), performance (z = 2.70, p = 0.01), and overall (z = 6.69, p < 0.01) all exceeded 50%, indicating that they are significantly heterogeneous at the 0.05 level. Consequently, researchers opted for a random-effect model to conduct the meta-analysis.

Fig. 4: A forest plot of meta-analyses of robot-based educational outcomes.
figure 4

This is a forest plot to show the meta-analysis results where effect sizes, weights, means, standard deviations, and sample sizes are presented and clarified.

Full size image

To conduct a meta-analytical examination of the results related to robot-based education, researchers performed a meta-analysis using Stata 18.0, and the findings were visualized in a forest plot (Fig. 4). Figure 4 presents the results of meta-analyses in the form of a forest plot, where the left column indicates authors and publication years, the middle forest indicates the effect sizes and confidence intervals, and the right column numerically indicates the effect sizes, confidence intervals, and weights. The middle line, referred to as a no-effect line, signifies no effect if a given line crosses it. A dot indicates an effect size, while the line crossing the dot indicates the confidence interval. The diamond at the bottom of the forest represents the pooled effect size.

Figure 4 clearly illustrates differences between the robot-based and traditional methods. As shown in Fig. 4, the diamonds for the subgroups “achievement, computational knowledge, motivation, performance, and overall educational outcomes” are all located to the right of the no-effect line. Meta-analysis results indicate that robot-based education is associated with higher academic achievement (g = 0.72; 95%CI = 0.30~1.13, medium-to-large effect), computational knowledge (g = 0.85; 95%CI = 0.55~1.15, large effect), motivation (g = 0.47; 95%CI = 0.19~0.75, medium effect), performance (g = 0.81; 95%CI = 0.22~1.40, large effect), and overall educational outcomes (g = 0.71; 95%CI = 0.50~0.92, medium-to-large effect) compared to traditional methods, though the observed effects vary across studies due to differences in implementation, cultural contexts, and outcome measures. Thus, researchers accepted all the five alternative hypotheses. To enhance readability, the results were summarized in Table 2.

Table 2 Summarized results of hypothesis testing.
Full size table

Discussion

The proven effectiveness of robots in education has led to the development of numerous robotic platforms for educational purposes (Zheng et al., 2024). This study made valuable contributions to the field of robot-based education through a meta-analysis and systematic review. It synthesizes the effects of the robot-based education on academic achievements, computational knowledge, motivation, and learning performance. The conclusions of this study align with previous studies (e.g., Kert et al., 2020; Chiazzese et al., 2019; Hsiao et al., 2019; Ooms et al., 2021; Hu et al., 2023), indicating that robot-based education is associated with higher academic achievement, computational knowledge, motivation, performance, and overall educational outcomes compared to traditional methods.

It is reasonable to conclude that robot-based education appears to offer advantages over traditional methods in terms of overall educational outcomes. Robots stimulate students’ curiosity about learning and encourage interaction, though the extent of this effect likely varies based on individual learner characteristics and instructional design. Students could form their organized structures and acquire knowledge systematically based on the guidelines of robots. However, these benefits may depend on contextual factors such as teacher training, student readiness, and technological infrastructure. Featuring multidisciplinary accessibility, AI technologies (Samala et al., 2024a), interactions, easiness, sharing, affordability, and availability, educational robots could exert a great influence on overall educational outcomes (Haddadin et al., 2022), including academic achievements, computational knowledge, motivation, and performance. However, it should be recognized that factors such as cultural differences, variations in pedagogical implementation, and disparities in technological infrastructure may moderate the effect of robot-based education on educational outcomes. Limitations, such as accessibility challenges, teacher training requirements, and potential negative effects on student engagement in certain situations, require further exploration.

The conclusion that robot-based education can improve academic achievements is well-founded. Students can improve their academic achievements by understanding the knowledge and actively constructing knowledge when solving problems using robotics. Robots can also integrate modern cultural-historical technologies into students’ preferences, improving their academic achievements. The pooled effect of robot-based education on academic achievement is substantial. Robot-based education may strengthen the interactions between the robot and learners (Ao & Yu, 2022), and may enhance knowledge acquisition and practices in human brains and learning behaviors. The established operation procedure in robots could also enhance the self-regulation in learning behaviors and improve academic achievements. The highly systematic output knowledge through the robot-based education could systematize and organize the pieces of knowledge, thereby enhancing academic achievements.

Robot-based education tends to improve computational knowledge. Practices and error checking of programming robotics assist students in constructing computational thinking and enhancing their computational abilities. Students have close contact with robotics, participate in the cultural-historical development of robotics, and deepen their understanding of computation. The frequent access to robots with computational components could facilitate the operation of robots, coupled with computational knowledge acquired in the learning process. The computational thinking embedded in the robots could accumulate students’ computational knowledge, which could, in turn, improve students’ engagement in robot-based education. In this way, a positive cycle was established from robot-based learning to computational knowledge gain. Students could conveniently and frequently complete exercises with the help of robots, eliminating the need to carry heavy printed books and bags for learning. Instead, a robot with multiple functions could provide abundant learning resources and facilitate the learning process (Ao & Yu, 2022).

The robot-based education is linked to higher motivation. The robotic interactions can stimulate students’ learning interest and improve their learning motivation by helping students construct knowledge and skills. As symbols of modern technology, robots attract students’ attention and motivate students to engage in socially interactive learning environments. The use of robots in education could improve cost-effectiveness, helping students with various backgrounds concentrate on robot-based teaching and learning. This great convenience could provide equal access to education, especially in the countries where there were poor digital infrastructures, less developed economies, and geographical barriers to education (Sun et al., 2021). This great convenience could motivate students to learn with robots rather than with heavy printed books and laptops. The robots could also group students and promote collaborative learning, enhancing students’ learning motivation.

Robot-based education may be associated with higher performance. Students can apply what they have learned to practice due to their active construction of knowledge and skills. Robotics, as cultural-historical tools, can offer an innovative presentation of new technologies and improve students’ learning performance. Robots can establish guidelines and standards for students’ learning processes, making it easier for students to follow and perform well academically. The teamwork in robot-based education could also enhance peer interactions and improve their performance since they could mutually benefit from the interactions. The more successful performers could set an example for less successful ones to follow and perform better. In this way, excellent achievers can make progress with the help of robots, while the slow learners’ interest and motivation can be sparked, ultimately leading to improved performance in educational settings.

Several factors may moderate the varying outcomes of robot-based education. Firstly, different educational institutions may vary in their educational philosophy and goals (Fan, 2024). Some educational institutions highlight critical thinking abilities and creativity, while others may focus on knowledge input and academic achievements. Robots may exert various influences on educational outcomes due to different highlights. Secondly, educators’ professional literacy and their perception of robot-based education may exert an important influence on educational outcomes (Mei et al., 2025). Some educators, with in-depth understandings of robots, can integrate robots into course design and teaching, leading to satisfactory teaching effects. However, others can merely present their lecture notes with the help of robots mechanically. Different educational outcomes will thus be caused. Thirdly, learners’ individual differences and social backgrounds cannot be ignored (Yang & Kuo, 2022). Individual cognitive styles and learning habits may exert different influences on robot-based educational outcomes. Some learners are skilled at operating with the assistance of robots, while others tend to engage in theoretical understanding and feel uneasy when assisted by robots. Finally, the evaluation systems may exert a direct influence on robot-based educational outcomes (Gao et al., 2025). It may be hard to comprehensively evaluate the educational outcomes using standardized evaluation tests. Creative thinking abilities, practical operation, and team collaboration can be measured through multi-faceted evaluation systems.

Conclusion

Major findings

Through a meta-analysis and systematic review, this study identified the effects of robot-based education on academic achievements, computational knowledge, motivation, performance, and overall educational outcomes, filling an important gap in robot-based education.

Limitations

Despite the rigorous design based on the guidelines of PRISMA, there are still several limitations to this study. Firstly, this study could not include all the related studies due to the limited library sources. Secondly, the various research designs of the included studies might have reduced the reliability of the findings. Thirdly, significant heterogeneity in included studies has been detected in terms of types of robots, educational levels, and cultural contexts, possibly limiting the generalizability of the results. Fourthly, it is hard to completely rule out publication bias, although no presence of publication bias was detected. Fifthly, the long-term effect of robot-based education needs further exploration. Finally, other factors such as teacher expertise and socioeconomic status may exert a great influence on the effect of robot-based education. Future research could incorporate more comprehensive studies and delve into different robot-based educational designs.

Implications for future research

This study provides concrete, actionable recommendations for educators and policymakers. Educators can flexibly and reasonably design robot-based education. For instance, they can present physical phenomena using robots in science teaching, making knowledge immediately accessible to students. They can organize learning activities and guide students to operate robots through programming, promote collaboration, and cultivate computational thinking. They can also regularly hold robot competitions to increase students’ interest in robot-based education. Policymakers may invest in robotics equipment, training, and standards. Meanwhile, they can organize training programs to improve robot-based education. They can also establish course standards related to robot-based education, normalize teaching contents, ensure the implementation of robot-based education, and cultivate talents suitable for the era of science and technology. Designers and educators should provide insights into optimal robot features and teacher training strategies in the future. Educators in countries like China could prioritize robots with computational thinking modules, while those in Turkey might focus on motivation-enhancing features.

Future research could not only focus on optimizing the design of robot-based education programs to improve educational outcomes, but also address identified limitations such as variability in implementation, cultural differences, and the need for longitudinal studies. It should also explore the acceptance model of educational robots, as this model is crucial for enhancing educational effectiveness (Wang et al., 2022). According to the acceptance model, robot designers could pay enough attention to important influencing factors, such as perceived usefulness, perceived ease of use, subjective norms, and behavioral intention (Liang & Hwang, 2023). Online learning philosophies, e.g., Community of Inquiry, could also be included to facilitate robot-based education because billions of learners have received online education since the outbreak of COVID-19 (Yu & Li, 2022).

Future studies should also aim to establish a sustainable model to enhance the effectiveness of robots in education, which is essential for the long-term development of robot-based education (Yu et al., 2022). The sustainable model should incorporate key factors such as interactivity, digital literacy, social emotions, deep neural network learning, and storytelling methods (Liang & Hwang, 2023). Designers and teachers could focus on enhancing students’ engagement, motivation, curiosity, and interest (Luo et al., 2024). They could also explore aspects such as the robot interface, virtual realities, augmented realities, humanlike functions, portability, costs, learning resources, emerging technologies for global education (Samala et al., 2024b), and learning environments (Yang et al., 2024).