Education in the AI era: a long-term classroom technology based on intelligent robotics

The paper addresses how to implement AI literacy across education levels by grounding instruction in the intelligent agent paradigm and using educational robotics to make abstract AI concepts concrete. It highlights consensus in the AIEd community on five core ideas (perception, natural interaction, representation and reasoning, learning, and societal impact) and notes the challenge of transferring AI literacy into real classrooms given heterogeneous systems, teacher preparation, and resource gaps. While machine learning—especially supervised learning—has numerous resources, areas such as perception, natural interaction, and reinforcement learning lack formal classroom-ready materials due to requirements for real-time, uncertain, environment-coupled interactions. The authors propose robotics as a natural solution, as robots embody intelligent agents interacting with real environments and humans. The research question is whether a robotics-based, long-term, level-adaptable platform (the Robobo Project) can feasibly support formal AI literacy aligned to these core ideas from secondary school through higher education.

Educational Robotics (ER) has expanded from university to general education, supporting STEM and project-based learning as well as 21st-century skills. Existing platforms include Bee/Blue-bot for early primary, KIBO and mBot, and LEGO MINDSTORMS for later primary/early secondary; at higher levels, iRobot Create, NAO, Turtlebot, and Khepera IV are used. However, most were not designed for AI literacy and lack key features: natural-interaction sensors (camera, microphone, touch), diverse actuators (mobility, speech, screens), computational power for vision/ML/reasoning, rich communications, multi-language programming with AI libraries, simulation, and aligned teaching materials. Two ER platforms aimed at AI education—Thymio and Fable—cover several needs (block-based and text programming, some wireless control, teaching materials), but still fall short of a complete roadmap for AI literacy: Thymio lacks a camera and advanced actuation/computation; Fable lacks a dedicated simulator and comprehensive AI-literacy-aligned resources. Thus, there remains a gap for a platform and curriculum that comprehensively support formal AI literacy across levels.

The study combines engineering and educational design research within the Robobo Project, a long-term initiative comprising: (1) a mobile robot that couples a simple motorized base with a standard smartphone (control, sensing, and communication via Bluetooth and WiFi/Internet), (2) modular software and development tools (Scratch-based blocks; Python libraries for perception, actuation, control; real-time streaming; and a Unity-based simulator, RoboboSim), and (3) aligned teaching units (TUs), documentation, and tutorials. Hardware followed a classical cycle (concept, design, prototyping, testing, refinement) initiated in 2016 with iterative validation sessions. Software followed the Unified Process with iterative analysis–design–implementation–testing from 2016 to 2019 and subsequent improvements. Educational resources were developed via educational design research with iterative classroom testing (2017–2023), analyzing learning outcomes and teacher acceptance via surveys and feedback to refine materials. The theoretical foundation adopts the intelligent agent perspective and structures AI literacy into seven topics: Sensing, Acting, Representation, Reasoning, Learning (including reinforcement learning), Collective AI, and Ethical/legal aspects. Robobo’s hardware integrates smartphone sensors (camera, microphone, touch, light, gyro, accelerometer, GPS) and base sensors (IR distance, encoders, battery), plus actuators (wheel and pan–tilt motors, LEDs, screen, speaker, torch). Software exposes functionality uniformly across levels: Scratch blocks for K12 and Python for K16/higher education, supporting advanced vision (color blobs, object recognition, face detection, lane/traffic sign, ArUco/QR), hearing (sound detection, offline speech recognition), touch, ambient sensing, actuation, and control (reactive to advanced, single- and multi-robot). RoboboSim supports two realism modes (deterministic simplified and physics-based) and an optional randomization mode to encourage robust solutions. Teaching methodology uses project-based learning: students complete challenges in teams or individually through phases of problem analysis, planning, solution design, validation in simulator and/or real robot, and presentation. Data for validation were collected via online questionnaires from secondary, undergraduate, and master-level cohorts over multiple years.

Across six years and multiple levels, results indicate Robobo effectively supports AI literacy aligned to the intelligent agent framework. Secondary school (AI+ project, final activity with 30 students and 12 teachers): teachers rated machine learning and social impact highest, with perception and actuation next—supporting the agent-based approach and robotics as an AI context. Students highlighted novel capabilities: computer vision (24 responses; 58.5%), human-robot interaction (22; 53.7%), speech (17; 41.5%), and Scratch programming (8; 19.5%). Teachers on Python: 50% support learning Python within AI units; 50% support Python but as separate preparatory training; 0% deem it too complex for secondary. On smartphone use: 58.3% favor using students’ own devices; 41.7% prefer school devices; 0% opposed. Undergraduate (n=56, first-year Engineering): solving a practical Robobo task was perceived as closer to engineering work (agreement skewed to 4–5 on a 1–5 scale; only 14.5% rated 1–2). Preference for traditional standalone coding exercises was low (1–2 on a 1–5 disagreement scale totaled 64.3%). The RoboboSim+Python activity received only 12.7% negative feedback, and documentation was rated helpful by 78.2% (ratings 4–5). Master level (n≈33): 94.1% found Robobo useful for intelligent robotics; 85.3% reported improved understanding of the role of basic sensors/actuators; 79.4% valued having a robot supporting vision algorithms; 94.1% agreed the Python library is adequate. Additionally, advanced students successfully implemented reactive/hybrid architectures, mapping and planning, and transitioned solutions from simulation to the real robot (with some tuning needed). Overall, both research questions were supported: RQ1—Robobo is perceived as a suitable, long-term tool for AI literacy across levels; RQ2—methodology, simulator/real integration, smartphone-based hardware, and materials are adequate and well received.

Findings affirm that a robotics-centered, intelligent-agent approach meaningfully operationalizes AI literacy across secondary to master levels. For RQ1, teachers’ prioritization of ML and social impact, with strong recognition of perception/actuation, shows that learners and instructors engage with the full agent loop, not just data-centric topics. Students valued computer vision and HRI capabilities, indicating that embodied interaction bridges abstract AI concepts and practical understanding. At university levels, hands-on, goal-driven tasks improved perceived relevance and motivation, aligning programming and systems thinking with real-world constraints. For RQ2, the PBL methodology, progressive tooling (Scratch to Python), robust documentation, and a simulator with adjustable realism reduced technical barriers while preserving authenticity. The smartphone-centric design provided high computational power and modern sensors for perception, natural interaction, and learning tasks, enabling coverage of under-resourced AI-literacy areas (e.g., perception, RL, collective AI). Master-level outcomes demonstrate that the platform scales to advanced architectural design and integration of external AI libraries. Compared to Thymio and Fable, Robobo uniquely satisfies the full set of required features (sensors/actuators for AI, computation, communication, multi-language programming, simulation, and AI-literacy-aligned teaching resources), enabling a coherent roadmap for long-term AI education.

The Robobo Project delivers a validated, classroom-ready ecosystem—hardware, software, simulator, and aligned teaching units—that supports AI literacy grounded in the intelligent agent paradigm from secondary school to higher education. It enables instruction across core topics (sensing, acting, representation, reasoning, learning, collective AI, and ethics) using project-based learning with smooth transitions from simulation to real robots. Over six years, students and teachers reported positive perceptions of relevance, feasibility, and learning support, with strong acceptance of Python, documentation, and the smartphone-based architecture. Compared to prominent ER platforms, Robobo uniquely fulfills all critical capabilities for formal AI literacy, including comprehensive teaching resources aligned with current AI-literacy frameworks. Future work should expand teaching units (e.g., reinforcement learning and multi-agent coordination), deepen assessment of learning gains across contexts, refine transfer from simulation to real robot, and explore equity and policy aspects of device provisioning.

Several practical limitations emerged: (1) Transfer from simulation to the real robot sometimes required additional tuning and time, which was not always available within course schedules. (2) Secondary-level activities often occurred mainly in simulation to manage time and technical complexity, potentially reducing exposure to real-world uncertainty. (3) Variability in students’ programming/AI backgrounds (multi-country cohorts) may confound outcome generalizability. (4) Smartphone heterogeneity can introduce performance disparities; some teachers preferred school-provided devices for equity and management. (5) Sample sizes per cohort were modest, and some survey responses were missing. (6) Teachers may require prior AI upskilling; while materials assist, professional development remains a dependency.