The Best Hardware for Teaching Artificial Intelligence in Middle and Primary Schools: 9 Suitable Smart Devices (Part 1)

Source: Wenxin Yiyan

“A craftsman must sharpen his tools before doing his work.”

Although the core of deep learning is training models based on actual problems, which can be implemented on computers, smart hardware is still indispensable, as AI application development requires the support of smart hardware. Smart hardware is not only a carrier of artificial intelligence knowledge but also a key tool to stimulate students’ interest and cultivate practical innovation abilities.

When selecting smart hardware, factors such as computing resources, hardware functions, application performance, and resource services must be considered. Zhong Baichang, Yu Junzhan, and Xie Zuoru selected 16 types of smart hardware for middle and primary school AI courses from four aspects: educational value, representativeness, audience reach, and focus on deep learning, and made a detailed comparison based on computing resources, hardware functions, application performance, and resource services.

Considering the common applications of deep learning (image recognition/speech recognition/speech synthesis/text generation/image generation/audio generation, etc.), this article selects 9 types of smart hardware from these 16 that have both image input and speech input capabilities, providing a brief introduction for middle and primary school teachers and AI enthusiasts.

01 Hangkong Board

1. Hardware Configuration: Equipped with a 2.8-inch color touch screen. It integrates Wi-Fi and Bluetooth communication modules, facilitating network connection and wireless communication.

2. Sensor Integration: Onboard multiple sensors, including light sensors, accelerometers, gyroscopes, and microphones, providing direct support for learning physical interactions, data collection, and simple IoT applications.

3. Processor and Operating System: RK3308, Arm 64-bit, 4 cores, 1.2GHz main frequency, with sufficient processing power to run complex Python programs. It runs a Linux operating system, providing users with a familiar and feature-rich development environment.

4. Programming Friendliness: The Hangkong Board is designed to simplify programming education, with special emphasis on compatibility with Python, allowing users to easily write and run complete Python programs, lowering the barrier to programming learning, especially suitable for teenagers and programming beginners.

5. Expansion Interfaces: It provides a rich set of expansion interfaces, including analog/digital/I2C/UART/SPI interfaces, supporting the connection of various external sensors and actuators, enhancing the scalability and diversity of projects.

6. Educational Applications: The Hangkong Board is very suitable for school education and STEM (Science, Technology, Engineering, and Mathematics) teaching, supporting everything from basic programming concept teaching to more complex IoT and robot control project development.

7. Community and Support: DFRobot, as a well-known open-source hardware manufacturer, provides strong community support and rich online educational resources for the Hangkong Board, including tutorials, project examples, software tools (such as Mind+ software), etc., helping users quickly get started and dive deeper into learning.

02 Raspberry Pi 4B

1. Processor: Raspberry Pi 4B is equipped with Broadcom’s BCM2711 chip, which is a quad-core 64-bit ARM Cortex-A72 processor, with an operating frequency boosted to 1.5GHz. Compared to the previous Raspberry Pi 3B+, both processing and graphics capabilities have been significantly enhanced.

2. Memory: Raspberry Pi 4B offers multiple memory configurations, including 1GB, 2GB, 4GB, and later added 8GB LPDDR4 SDRAM, marking the first time the Raspberry Pi series offers more than 1GB of memory, greatly enhancing its ability to run complex applications.

3. Graphics Processing: Built-in VideoCore VI GPU, supports OpenGL ES 3.0, Vulkan, and 4K resolution video decoding at 60 frames per second, capable of handling high-definition multimedia content and light gaming experiences.

4. Interfaces: Provides two USB 3.0 interfaces and two USB 2.0 interfaces, one Gigabit Ethernet port, dual-band 802.11ac Wi-Fi, and Bluetooth 5.0, improving data transfer speeds and wireless connectivity.

5. Video Output: Supports dual display output, including two Micro HDMI ports, supporting up to 4K resolution output, enabling multi-screen display and high-definition display capabilities.

6. Storage: No longer uses SD cards as the default storage medium but adopts a microSD card slot, while also supporting SSD or HDD connections via USB to enhance storage speed and capacity.

7. Operating System: Raspberry Pi 4B supports multiple operating systems, including the officially recommended Raspberry Pi OS (a Debian-based Linux distribution), as well as other third-party systems like Ubuntu, Windows 10 IoT Core, etc.

8. Size and Power Consumption: Maintains the small size typical of Raspberry Pi, approximately 85 x 56 x 17 mm, but due to performance improvements, maximum power consumption has also increased, recommending at least a 2.5A power adapter.

Raspberry Pi 4B has become a popular choice in education, DIY projects, IoT applications, media centers, lightweight servers, and other fields due to its powerful performance, rich interfaces, and low price.

03 Sunrise X3 Pie

1. High-performance AI Processing Capability: The Sunrise Pie is equipped with Horizon’s self-developed BPU (Brain Processing Unit), a hardware-level AI accelerator designed for edge computing, capable of providing up to 5 TOPS (trillion operations per second) of edge-side inference capability, making it very suitable for running machine learning and deep learning algorithms.

2. Processor Specifications: Built-in quad-core ARM Cortex-A53 processor, running at 1.2GHz, supports overclocking, providing strong processing power for general computing tasks. Provides 5 Tops of strong computing power support.

3. Interface Compatibility: The Sunrise Pie’s interface design is compatible with Raspberry Pi, meaning it can use a large number of accessories and peripherals designed for Raspberry Pi, increasing its flexibility and extensibility.

4. Multi-sensor Support: The development board can process multiple Camera Sensor inputs simultaneously, which is particularly useful for building visual applications involving image recognition and object tracking.

5. Rich Hardware Resources: In addition to a powerful computing core, it is also equipped with Wi-Fi, Bluetooth, and other wireless communication modules, as well as rich GPIO, USB, HDMI interfaces, facilitating the connection of various peripherals and data transmission.

6. Educational and Development Resources: Horizon provides detailed documentation, tutorials, and development toolchains for the Sunrise Pie, including support for Linux systems and tutorials for beginners and advanced users, making it very suitable for educational institutions, individual developers, and corporate R&D teams for AI project development and learning.

7. Temperature Management: According to user tests, the Sunrise Pie maintains good temperature control during normal operation, with chip temperatures around 51 to 55 degrees, indicating thoughtful heat dissipation design.

8. Community and Ecosystem: The Sunrise Pie has an active developer community where users can share project experiences, obtain technical support, and participate in the growing ecosystem.

In summary, the Sunrise Pie is a comprehensive and powerful AI development platform designed to lower the barriers to AI technology development, accelerate the innovation and implementation of AI applications, and is particularly suitable for education, scientific research, and product prototype development.

04 NVIDIA Jetson Nano

1. Processor and GPU: Jetson Nano is equipped with a quad-core 64-bit ARM Cortex-A57 CPU and a 128-core Maxwell architecture GPU. Although the GPU architecture is relatively old, it achieves a balance between power consumption, size, and cost, suitable for running optimized small-scale neural networks.

2. Memory and Storage: Typically equipped with 4GB LPDDR4 memory, but note that a portion of this memory (about 1GB) is shared with video memory. In terms of storage, it usually uses a microSD card for system image writing and data storage.

3. Rich Interfaces: Provides GPIO interfaces (40-pin), supporting various peripheral connections; USB 3.0, USB 2.0 interfaces for external devices; HDMI and DisplayPort interfaces supporting dual display output; Gigabit Ethernet interface; and interfaces for camera connections supporting MIPI CSI-2 protocol.

4. Software Support: Supports NVIDIA JetPack SDK, a comprehensive development toolkit that includes Linux operating system, CUDA, cuDNN, TensorRT, etc., as well as optimized support for mainstream AI frameworks (such as TensorFlow, PyTorch, Caffe, etc.).

5. Power Consumption: Due to its design focusing on energy efficiency, Jetson Nano has low power consumption during operation, making it suitable for battery-powered or energy-sensitive scenarios.

Application Areas:

1. Computer Vision: As mentioned, Jetson Nano is widely used in scenarios requiring real-time image processing and analysis, such as people counting, object recognition, and security monitoring.

2. Robotics: Due to its powerful computing capabilities and good support for sensors, Jetson Nano has become the preferred computing platform for many robotics projects, used for autonomous navigation, obstacle avoidance, etc.

3. Healthcare: Can be used to create health monitoring devices supporting computer vision, such as remote patient monitoring systems.

4. Retail Solutions: Real-time customer behavior analysis, inventory management, etc., improving store operational efficiency.

5. Education and Research: Due to its relatively affordable price and powerful AI capabilities, Jetson Nano is also an ideal tool for educational institutions and researchers for AI education and project development.

In summary, Jetson Nano is a multifunctional edge computing platform that provides strong support for a wide range of industry applications with its outstanding performance in AI and computer vision fields.

05 Maix Bit

1. Processor Core: The K210 chip includes two 64-bit RISC-V dual-core processors, supporting vector instruction sets, making it particularly suitable for applications such as machine learning, image processing, and signal processing.

2. AI Computing Power: The K210 chip is claimed to achieve 1 TOPS (trillion operations per second) of computing power, which is a significant performance indicator for edge devices, enabling it to run complex AI models directly on the device without relying on cloud processing.

3. Onboard Resources: The Maix Bit development board is compactly designed, containing Type-C interface, USB-UART interface, GPIO pins, camera interface, etc., facilitating peripheral expansion and development debugging. The onboard K210 chip also integrates audio input/output, SPI, I2C, PWM, and other interfaces, adapting to diverse application scenarios.

4. Camera Support: Notably, Maix Bit supports connecting camera modules, suitable for visual-related project development, such as image recognition and object tracking.

5. Development Environment: Maix Bit supports two main development environments, namely Kendryte IDE and MaixPy IDE. The former is based on C/C++, suitable for low-level development and performance optimization; the latter is a MicroPython-based development environment, more concise and user-friendly, suitable for rapid prototyping and educational purposes.

6. Community and Learning Resources: An active developer community has formed around Maix Bit and the K210 chip, providing a wealth of learning materials, project cases, and tutorials, catering to both beginners and experienced developers.

7. Application Scenarios: Due to its powerful AI processing capabilities and flexibility, Maix Bit is widely used in IoT, smart home, educational robots, wearable devices, smart security, industrial automation, and other fields.

In conclusion, Maix Bit is a forward-looking, highly integrated AIoT development platform designed to lower the development threshold of AI technology and promote the popularization and application of AI technology in edge devices.

06 Wio Terminal

1. Core Hardware: Based on a high-performance SAMD51 microcontroller, running at a speed of 120MHz, with a maximum overclocking capability of 200MHz, equipped with 4MB external flash memory and 192KB RAM, providing sufficient resources for complex applications.

2. Wireless Connectivity: Built-in Realtek RTL8720DN module, supporting Wi-Fi and Bluetooth connections, facilitating IoT application development and remote control.

3. Integrated Peripherals: Features a 2.4-inch color LCD screen that supports touch input, and an onboard IMU (LIS3DHTR) for motion sensing, along with a microphone, buzzer, microSD card slot, light sensor, and infrared transmitter, providing hardware foundation for various interactive projects.

4. Expansion Interfaces: Provides two Grove ports, compatible with Grove ecosystem sensors and actuators, and a 40-pin GPIO interface compatible with Raspberry Pi, supporting the connection of more external hardware, increasing flexibility and extensibility.

5. Development Environment: Compatible with Arduino and MicroPython, allowing developers to choose their preferred programming environment, lowering the learning barrier, especially suitable for the education field.

6. Application Examples: Wio Terminal can be used to construct various projects, such as an automatically connected weather forecast instrument, audio recognition, action recognition, face recognition, infrared thermal imaging, smart home control systems, etc., fully demonstrating its application potential in IoT and AI fields.

7. Community Support: Has an active developer community and rich online resources, including tutorials, project examples, libraries, and forum support, helping users quickly get started and innovate.

In summary, Wio Terminal, with its integrated design, powerful processing capabilities, rich peripherals, and good extensibility, has become an ideal choice for middle and primary school AI education and IoT project development, helping students intuitively understand AI and IoT technologies and enhance skills through hands-on practice.

07 M2 Dock

1. Computing Resources: The V831 chip includes a neural network processor (NPU), and although its theoretical computing power is relatively low (0.2 TOPS), it is sufficient for lightweight machine learning and image processing tasks, making it suitable for education and rapid prototyping.

2. Hardware Interfaces and Functions:

– Image Processing Capability: Equipped with a camera interface, supporting 2MP cameras, suitable for image recognition, object tracking, and other visual applications.

– Screen Display: Features a 1.3-inch screen for easy viewing of processing results or user interfaces, enhancing interactivity.

– Extensibility: As shown in the analysis table, M2 Dock supports multiple expansion functions, including image input, audio input/output, user interaction, etc., and some functions are already integrated into the motherboard, eliminating the need for external expansion devices, improving convenience.

3. Development and Learning Resources: M2 Dock supports various programming methods, such as Linux shell commands and Jupyter Notebook (Python), and provides detailed getting started guides and online resources, facilitating quick entry for beginners and suitable for experienced developers to explore in depth.

4. Application Scenarios: Due to its integrated AI accelerator and relatively comprehensive hardware interfaces, M2 Dock is very suitable for projects such as automatic following cars, image recognition, object tracking, etc. In particular, in middle and primary school AI education, it can serve as a teaching tool, helping students understand the basic concepts of AI, practice programming, and hardware integration projects.

In the field of AI education in middle and primary schools, M2 Dock, with its integrated AI accelerator, rich extensibility, and easy-to-use development environment, provides students with an intuitive learning platform for AI technology. It can support basic deep learning applications such as image processing and speech recognition, and through its flexible extensibility, students can learn how to connect and control different sensors and actuators in practical operations, deepening their understanding and practical abilities regarding AI applications.

In summary, M2 Dock is an AI development board aimed at education and beginners, which lowers the barriers to learning AI by integrating hardware resources and providing comprehensive development support, promoting the popularization and development of AI education.

08 MAIX-III AXera-Pi

1. Core Processor: The MAIX-III AXera-Pi is designed based on the Aisen Smart AX620A chip, which is a high-efficiency SoC (System on Chip) integrating a dedicated AI accelerator. The AX620A chip supports high-performance image processing and AI inference, with a theoretical peak computing power of up to 3.6 TOPS (trillion operations per second), suitable for complex visual tasks.

2. Visual Processing Capability: Particularly emphasizes its “low-light full-color” technology and AI visual capabilities, meaning that this development board can provide good imaging effects even in low-light conditions and efficiently execute image recognition, object detection, and other visual algorithms based on deep learning.

3. Linux Platform: As a Linux board, MAIX-III AXera-Pi provides complete support for the Linux operating system, which is very beneficial for developing complex applications and software stacks. Developers can program in a familiar Linux environment, utilizing rich open-source libraries and frameworks for development.

4. Open-source Ecosystem: The AXera-Pi series actively participates in building an open-source ecosystem, collaborating with platforms like Baidu PaddlePaddle, promoting co-creation of hardware ecosystems, meaning developers can enjoy abundant community resources, toolchains, and development support.

5. Hardware Features: Although some people expressed disappointment with the packaging experience (only using cardboard packaging), the board itself provides various interfaces, including camera interfaces, multiple peripheral interfaces (such as USB, HDMI, network interfaces, etc.), and expansion interfaces, facilitating the connection of various sensors and actuators to meet diverse project needs.

6. Education and Applications: MAIX-III AXera-Pi is not only suitable for professional developers for product prototype development but also very suitable for the education field, especially for AI and machine vision-related course teaching, allowing students to master AI technology through practice.

In summary, MAIX-III AXera-Pi is a high-performance visual AI development board aimed at professional developers and the education market, combining powerful hardware acceleration capabilities, flexible development environments, and rich ecosystem resources, aiming to promote the widespread application and innovation of AI visual technology.

09 CocoRobo AI Module (C-AI)

The AI module serves as the main control module for AI-related applications, possessing basic computer vision recognition functions and machine learning model inference capabilities, helping to achieve beginner and advanced-level AI applications.

The module is mainly applied in AI-related computations, such as visual recognition and speech recognition. The module runs on a MicroPython system, allowing programming using Python syntax.

The main controller adopts the AI chip Sipeed M1w, equipped with a Kendryte K210 (RISC-V DualCore 64bit@400MHz), with built-in FPU, KPU, FFT, enabling offline AIoT functionality (Edge Computing). The module supports Wi-Fi 802.11b/g/n, enabling 2.4GHz wireless communication via an external antenna. The module also includes most common interfaces for Arduino and Micro:bit.

Conclusion

Reviewing the history of information technology education in middle and primary schools in our country, it is clear that the continuous advancement of tools has made it possible for visual programming, sensors, IoT, and other content to enter the classrooms of middle and primary schools.

After discussing the above nine types of smart hardware, a distinct trend emerges: the hardware devices supporting AI teaching are not only numerous but also have strong performance and functionality, opening an unprecedented technological door for learners and assisting artificial intelligence, especially the new generation of AI represented by deep learning, to enter the classrooms of middle and primary schools.

Of course, the choice of smart hardware should not only focus on single performance indicators but should comprehensively consider its functional practicality, economic cost, adaptability of the development environment, and other factors to ensure that the selected hardware can seamlessly integrate into the teaching system, maximizing learning outcomes.

After solving the hardware problem, the adaptability and application ability of teachers towards AI become key. Teachers need to master how to effectively utilize these smart hardware, design and implement AI application teaching, transforming technological potential into students’ knowledge and skills.

The professional growth of teachers and the rational selection of smart hardware together constitute a dual-wheel drive to advance the development of AI education in middle and primary schools.

References:

[1] Zhong Baichang, Yu Junzhan, Xie Zuoru. What kind of smart hardware is needed for AI courses in middle and primary schools? – Current situation analysis and development direction [J]. Journal of Distance Education, 2024

[2] Raspberry Pi 4B. https://wiki.dfrobot.com.cn/Introduction_to_Raspberry_Pi_4B_c

[3] Horizon RDK Kit User Manual. https://developer.horizon.ai/api/v1/fileData/documents_rdk/index.html

[4] Wio Terminal Getting Started Tutorial https://wiki.seeedstudio.com/cn/Wio-Terminal-Getting-Started/

[5] MAIX-III AXera-Pi. https://wiki.sipeed.com/hardware/zh/maixIII/ax-pi/axpi.html