Key points to remember
Educational robots range from programmable kits to social humanoids, telepresence units and AI teaching assistants, and the role a robot plays matters more than its hardware.
A 2026 meta-analysis of 78 controlled studies found that robots co-teaching with a human reached an effect of d = 0.88, while robots replacing the teacher showed no benefit (d = -0.06).
Educational robotics improves STEM learning performance and attitudes, but the evidence for computational thinking gains is mixed.
Robots do best on narrow, structured tasks such as practice drills, language work and social skills support.
Cost, privacy, technical support and teacher training decide whether a classroom robot succeeds.
Summary
Robots in education have moved from science fiction headlines to a modest but real classroom tool. This article explains what educational robots are, sorts them into four practical types, and walks through their main uses, from STEM lessons to support for children with autism and for students who cannot attend school. It then reviews what recent meta-analyses actually measured: robots can improve learning compared with no training, they work best alongside a teacher, and they show no advantage when they try to replace one. The debate over whether robots will replace teachers is addressed directly, followed by the practical limits: price, privacy, technical support and the guidance UNESCO has issued for AI in schools. A short list of common mistakes and a six-question FAQ close the guide, so you can judge any classroom robot claim with a clear head.
Table of contents
1. What are educational robots?
Educational robots are physical machines that support learning, either as something students build and program or as something they interact with as a tutor, peer or assistant. The educational robots definition is broad on purpose, because the label covers very different devices. Sorting them by the role they play is more useful than sorting them by hardware, and it gives four types of educational robots.
Programmable kits and construction robots
These are the classic classroom robots: wheeled floor robots for young children, brick-based kits and microcontroller builds. The student is the programmer, and the robot is a concrete way to see code, sensors and mechanics produce motion. Most STEM robotics programs belong in this group, and it is where robots used in education are most common.
Social and humanoid robots
Social robots in education talk, gesture and respond to children, which lets them act as tutors or learning companions. Humanoid robots in education, such as the NAO robot made by SoftBank Robotics, are used in schools to teach programming and communication skills, according to an industry guide published in 2024. The same humanoid platforms sit at the center of a fast-moving hardware industry, which we map in our analysis of the humanoid robotics supply chain in 2026.
Telepresence robots
Telepresence robots in education stand in for a student who cannot be in the room. A screen, camera and microphone sit on a small body, and the absent student watches and speaks through it. The robot is not teaching anything: it keeps a child connected to a class.
AI teaching assistants
So what is an AI teacher? Usually it is software, such as a tutoring system or a chatbot, rather than a machine with a body. When that software runs inside a physical robot the two ideas merge, and the robot can listen, watch and react using several signals at once, the same multi-modal approach we describe in our article on multi-modal AI for elderly monitoring. The distinction still matters for cost and for privacy, as discussed later in this guide.
2. Uses of robots in education
The question of what is the use of robots in education has several answers, and the application of robots in education is easier to judge when you look at who benefits. So how can robots be used in education in practice? Examples of robots in education cluster in five areas.
STEM teaching is the largest. A 2024 multilevel meta-analysis in the International Journal of STEM Education pooled 21 studies and 2,433 participants and found a moderate positive effect of educational robotics on STEM learning (Hedges' g = 0.488), with g = 0.665 for learning performance and g = 0.497 for attitudes. Robots work well here because abstract ideas such as loops, sensors and feedback become visible. The sensing concepts students meet on a classroom robot also scale up to aircraft, as our guide to fixed wing drones and UAV sensors shows.
Language and maths tutoring comes second. In their widely cited review, Social robots for education, in Science Robotics, Tony Belpaeme and colleagues report that social robots acting as tutors or peer learners have achieved results similar to human tutoring on restricted tasks, and they link much of the effect to the robot's physical presence.
Special education is third. A systematic review of twelve studies on social humanoid robots, covering children aged 2 to 15 and mostly children with autism, associated robots such as NAO and KASPAR with improvements in eye contact and joint attention and with more frequent initiation of interaction. Gains in imitation skills were limited, and the authors flag small samples and a lack of control groups, so the results are promising rather than settled.
Remote attendance is fourth. In a qualitative OsloMet study of the AV1 telepresence robot for children with long-term illness, almost half of 37 participants described themselves as very satisfied, mainly because the robot kept them in contact with friends and reduced loneliness.
Higher education is fifth. Robotics in higher education is mostly about studying robots rather than being taught by them: engineering and computing courses use robots as lab platforms and as subjects of research on how people and machines interact. At university level, robotics in STEM education therefore blends the content with the tool.
3. Teaching with robots in the classroom
Teaching with robots works best when the teacher stays in charge of the lesson and the robot handles a narrow job. Robots for teaching take four common roles. As a tool, the robot is what students program. As a tutor, it delivers practice and feedback on a defined task. As a peer, it plays a learner that the child can teach, which pushes children to explain ideas in their own words. As a co-teacher, it supports the human teacher with repetitive or attention-grabbing parts of the lesson. The last role matters most, as the evidence in the next section shows.
A few practical principles follow. Start from the learning goal, not from the robot. Choose tasks that are restricted and measurable, such as vocabulary practice or a single programming concept, because that is where tutoring results are strongest. Plan the teacher's role explicitly: who introduces the task, who handles errors, who leads the debrief. Keep sessions short and structured so that novelty does not do all the work. And plan the basics early, because robots in schools depend on infrastructure. In the OsloMet telepresence study, many families struggled with unstable internet, inconsistent audio and video, school objections over privacy and GDPR, and a lack of support.
Robots also make good teaching material for the science of learning itself. A robot that reacts to a clap, a light or a gesture is a concrete demonstration of the stimulus and response logic explained in our guide to the definition and types of stimuli in psychology, and our 50 examples of stimuli give a teacher plenty of ideas for a short classroom activity.
Finally, teaching with robots does not have to start with the most advanced machine. A simple programmable kit with a clear lesson plan and a teacher who has practiced beforehand will usually teach more than an impressive humanoid used as a demonstration. Novelty fades quickly in a classroom, while a well designed task keeps its value.
4. What research says about robots and education
Robots and education have now been studied enough to support meta-analyses, which pool many studies into one estimate. The most recent, published in Frontiers in Robotics and AI by Joost de Winter and colleagues, retrieved 146 studies. Of these, 78 had control groups and covered 4,819 participants. Robots improved learning most clearly against no training (d = 0.75). Against human teachers the overall effect was small and positive (d = 0.31) but varied widely, and against virtual interfaces robots held a modest edge (d = 0.20). The striking result concerned roles: robots co-teaching with a human reached d = 0.88, while robots replacing the teacher showed no benefit (d = -0.06).
Earlier work gives context. The 2024 STEM meta-analysis found a significant moderate effect overall, but its effect on computational thinking was small and not significant (g = 0.079) and rested on only four effect sizes. A separate 2024 meta-analysis in Heliyon on early childhood education reported that robotics activities had a significant impact on computational thinking, so the evidence on that particular skill is currently mixed. A 2023 meta-analysis in Sustainability found that educational level significantly moderated results, with larger standardized mean differences for higher education (1.42) and secondary school (1.69, from only three effect sizes) than for primary (0.78) or pre-school (0.55). The type of robot did not significantly change outcomes, which supports the idea that pedagogy matters more than hardware.
Two cautions apply. First, these are averages across very different robots, tasks and ages, and the authors of the 2026 review note that they weighted studies equally, included non-randomized designs and could not rule out publication bias. Second, that review counted only cognitive outcomes, so emotional or social benefits are not captured. Reading any of these papers well also means checking how students were sampled, and our guide to stratified vs cluster sampling explains why the design behind a sample limits what can be generalized.
5. Can robots teach, and will they replace teachers?
Can robots teach? Yes, in a narrow sense. Belpaeme and colleagues found that on restricted tasks, tutoring robots can reach results similar to human tutoring. The word that matters is restricted. A robot can present a vocabulary drill, adjust the difficulty and give immediate feedback. Teaching a full classroom also means judging who is confused, managing behavior, adapting to a bad day, building trust and deciding what matters most. The research reviewed above does not show robots doing these things reliably.
Are robots replacing teachers? Not according to the data. In the 2026 meta-analysis, robots that replaced the teacher showed no benefit, while robots that worked with a teacher showed the largest effect. When you meet a headline about teacher robots in Japan or India, check what the robot actually did: delivering a scripted lesson is very different from running a class.
The related debate, should robots replace teachers, and what AI means for the future of education, is as much political as technical. UNESCO's 2023 guidance on generative AI takes a human-centred position. It examines how generative AI could threaten the agency of learners and teachers, and it calls on governments to regulate its use, protect data privacy and train teachers, according to a summary of its release.
The realistic picture is a robot as an assistant. If a robot takes over drills, repetition and routine feedback, the teacher gains time for the parts that only a human can do: explaining, encouraging, noticing and deciding. That outcome is a design choice, not a given, and it depends on schools asking what the robot is for before deciding that it is needed.
6. Costs, privacy and limits of robots in schools
Cost is the first limit. Humanoid platforms sit at the expensive end of the market: the 2024 industry guide quoted above puts NAO at roughly $7,500 to $13,000, although it gives inconsistent ranges and does not date its sources, so treat the figure as indicative. Programmable kits and software-only AI assistants cost far less, which is one reason the hardware is rarely the first thing a school should buy. Training, maintenance, replacement parts and connectivity add to the total.
Privacy is the second limit. A robot with a microphone and a camera raises the same questions as any classroom AI tool: where recordings are stored, who can access them and what consent applies. UNESCO's guidance sets an age limit of 13 for the use of AI tools in the classroom, according to the same summary, and a UNESCO survey of over 450 schools and universities found that less than 10% had institutional policies or formal guidance on generative AI. In the OsloMet study, schools raised privacy and GDPR objections to the telepresence robot.
Support and fit are the third limit. Some children in that study were too ill to use the robot at all, and parents often handled the bureaucracy themselves, which shows that even a well designed device fails without institutional support. Access also matters: if only well-funded schools can afford robots, they can widen existing gaps. The Heliyon authors suggest that supportive legislation and educational contexts that prioritize these skills appear more important than the digital access gap, which is a reminder that policy shapes outcomes as much as devices do.
7. Common mistakes to avoid with robots in education
Buying the robot before the learning goal. Schools do this because a demonstration is persuasive. Start with a measurable objective, then choose the simplest device that serves it. Write the objective in one sentence first. Example: a vocabulary drill needs a tutor interface, not a walking humanoid.
Treating one study as proof. A single positive result is easy to find, and it is a data point, not a verdict. Look for meta-analyses, and read their limits. The 2026 review weighted studies equally and could not rule out publication bias.
Expecting computational thinking gains automatically. Teachers assume that robot activities teach it by default. The evidence is mixed: one 2024 meta-analysis found a small, non-significant effect (g = 0.079), another found a significant one. Design tasks that target the skill and assess it.
Letting the robot replace the teacher. It looks efficient on paper. The data point the other way: robots replacing teachers showed no benefit (d = -0.06), while co-teaching reached d = 0.88.
Ignoring privacy and support. Both are invisible until something breaks. Check data storage, consent and maintenance first, and ask the vendor for its data policy in writing, as the OsloMet families learned with unstable connections and GDPR objections.
8. Frequently asked questions about robots in education
Can robots teach children effectively?
On narrow, structured tasks, yes. Belpaeme and colleagues report that social robots acting as tutors have reached results similar to human tutoring on restricted tasks, and the 2026 meta-analysis found that robots improved learning against no training (d = 0.75). Effectiveness depends heavily on the role: robots worked best as co-teachers (d = 0.88). Treat the robot as a practice partner inside a lesson led by a teacher, and set a clear goal for each session so that you can measure whether it worked. Results are strongest where the task is narrow, the instructions are clear and a teacher checks the outcome. They are weaker when the robot has to cope with open conversation or a whole class at once. Before adopting one, ask what the robot will teach, how you will measure it and who steps in when it fails.
Are robots going to replace teachers?
The evidence does not support it. Robots that replaced the teacher showed no benefit (d = -0.06) in the 2026 meta-analysis, and UNESCO's guidance stresses human agency and teacher training. The likelier future is a robot that takes over repetitive practice and routine feedback so that the teacher can spend more time on explanation, encouragement and judgment. Whether that happens depends on how schools design the role. Headlines can suggest otherwise, so check what a robot actually did in a pilot and for how long. Teachers also carry duties such as safeguarding, classroom culture and pastoral care that the studies reviewed here do not measure.
What is the use of robots in education?
Robots serve five main purposes: teaching STEM through programming and building, tutoring in languages and maths, supporting children with special needs such as autism, keeping ill or absent students connected to class through telepresence, and serving as lab platforms in higher education. Each use has different evidence behind it, with the strongest support for STEM learning performance and for co-teaching, and the weakest for general replacement of teachers or for computational thinking. For a school choosing where to begin, STEM and co-teaching are the safest starting points because they have the most evidence. Remote attendance and special education are valuable but depend on individual needs and specialist support, so pilot them carefully and evaluate the results with students and families.
Do robots help children with autism?
Early evidence is encouraging but limited. A systematic review of twelve studies, mostly on children with autism aged 2 to 15, linked social humanoid robots such as NAO and KASPAR to improvements in eye contact, joint attention and the initiation of interaction, with limited gains in imitation. The authors point to small samples and a lack of control groups. A robot should be used under the guidance of a specialist, as one tool among several. Look for studies with control groups before generalizing, and involve parents and therapists in any decision. The research reviewed here does not show a robot replacing human therapy or specialist teaching, and results for one robot or one child do not automatically transfer to another.
How much do educational robots cost?
It depends on the type. An industry guide from 2024 quotes the NAO humanoid at roughly $7,500 to $13,000, with inconsistent ranges. Programmable kits and software-only AI tutors cost much less. The sticker price is only part of the total: training, maintenance, replacement parts, connectivity and data protection all need a budget, and the OsloMet study shows how missing support can undermine a robot that works well on paper. Ask vendors for a total cost over three years, including software licenses, training and repairs, and compare it with a simpler kit that serves the same learning goal. Because the 2023 meta-analysis found that robot type did not significantly change outcomes, a more expensive machine is not automatically a better teaching tool.
What is the difference between an AI teacher and a robot teacher?
An AI teacher is usually software, such as a tutoring system or chatbot, that explains, asks questions and gives feedback. A robot teacher adds a physical body, and the research suggests that physical presence contributes to the learning effect of social robots. In practice the two overlap, since many classroom robots run AI software. The distinction matters because a physical robot adds hardware costs, sensors that can collect more data, and a different set of privacy questions. When a vendor says its product is an AI teacher, ask whether the intelligence runs in a cloud service, what data it sends there and whether it works offline. You can also test your science knowledge afterwards with our quiz collection.
9. Conclusion: robots work best with teachers
Robots in education are neither a revolution nor a gimmick. The research shows real gains for STEM learning and for narrow tutoring tasks, mixed evidence for computational thinking, and no benefit when a robot tries to replace the teacher. The best results come from clear learning goals, a teacher who stays in charge, and attention to cost, privacy and support. Ask what the robot is for before asking which robot to buy. Used that way, a robot becomes one more tool in a teacher's hands, and the question for every school is whether it earns its place against simpler options.
Sources
Social robots: a meta-analysis of learning outcomes (de Winter, Dodou, Moorlag, Broekens) - Frontiers in Robotics and AI - https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2025.1735198/full
The effects of educational robotics in STEM education: a multilevel meta-analysis (Ouyang, Xu) - International Journal of STEM Education, vol. 11, article 7, 2024 - https://link.springer.com/article/10.1186/s40594-024-00469-4
Enhancing computational thinking in early childhood education with educational robotics: A meta-analysis (Alonso-García et al.) - Heliyon (Elsevier), 2024 - https://doi.org/10.1016/j.heliyon.2024.e33249
The Effectiveness of Educational Robots in Improving Learning Outcomes: A Meta-Analysis (Wang et al.) - Sustainability (MDPI), vol. 15, no. 5, article 4637, 2023 - https://www.doi.org/10.3390/su15054637
The Use and Future of Social Humanoid Robots in Special Education: A Systematic Review (İnan, Güldenoğlu) - International Journal of Sciences: Basic and Applied Research, vol. 75, no. 1, 2025 (university repository record) - https://avesis.bozok.edu.tr/yayin/50b05b9b-57a0-4ecf-9543-4df3b591c89a/the-use-and-future-of-social-humanoid-robots-in-special-education-a-systematic-review