Against the backdrop of advancements in artificial intelligence, edge computing, the Internet of Things (IoT), and mobile embedded hardware, this paper systematically introduces the emerging field of intelligent IoT. It empowers IoT sensing, communication, computing, and applications with AI technology, exhibiting new characteristics such as ubiquitous intelligent sensing, cloud-edge-device collaborative computing, distributed machine learning, and human-machine-thing integration, resulting in greater flexibility, self-organization, and adaptability. This paper first introduces the basic concepts and characteristics of intelligent IoT; secondly, it elucidates the architecture of intelligent IoT; further, it details the research challenges and key technologies in intelligent IoT, including ubiquitous intelligent sensing, crowdsourced intelligent sensing computing, intelligent IoT communication, terminal-adaptive deep computing, distributed learning for IoT, cloud-edge-device collaborative computing, and security and privacy protection; finally, based on the latest research dynamics, it looks forward to highly promising future research directions, including hardware-software collaborative terminal intelligence, intelligent evolution for AIoT, next-generation intelligent IoT networks, continuous evolution of dynamic scene models, human-machine-thing integrated crowdsourced intelligent computing, and general-purpose AIoT system platforms.
The Internet of Things (IoT), or "the Internet of Everything," is considered the next wave of the information industry after computers and the Internet, and a crucial component of the new generation of information technology. It's a network that extends and expands upon the Internet, combining various information sensing devices with the network to form a vast network that enables interconnection, information exchange, and intelligent services between people, machines, and things anytime, anywhere. The Internet of Everything represents another major revolution in human technological history, having a profound and far-reaching impact on social production and life. Since its inception, the rapid development of IoT technology has continuously led to industrial upgrading, while simultaneously placing higher demands on its technological evolution. Specifically, there are five important development trends.
First, the widespread adoption of IoT terminal devices has led to an explosive growth in terminal data and connections. According to Huawei's GIV (Global Industry Vision)① and Cisco②, the number of connected devices globally will reach 100 billion by 2025, and more than 500 billion IoT devices will be connected to the internet by 2030. At that time, the total amount of data generated globally each year will reach 1 YB, a 23-fold increase compared to 2020. This massive data connectivity necessitates IoT architectures with higher computing power to enable timely data analysis and processing.
Secondly, the requirements for real-time data processing and privacy are becoming more urgent. New IoT businesses are constantly emerging, and the data deluge brought about by ubiquitous sensing and interconnection will deeply integrate with various industries, giving rise to the industrial IoT. Many specialized application scenarios, such as security monitoring, autonomous driving, and online healthcare, on the one hand, have high requirements for real-time data processing and low data transmission latency; on the other hand, because they are gradually becoming deeply integrated into people's daily lives, the requirements for privacy protection are also extremely urgent.
Thirdly, the rise of artificial intelligence technologies such as deep learning. In recent years, a new generation of artificial intelligence technologies, represented by deep learning, has developed rapidly. Compared with traditional machine learning models, deep learning has achieved better performance results in many domain tasks. However, at the same time, as the number of network layers increases, the scale of its model parameters continues to grow, and the computational cost continues to increase, which brings great challenges to its deployment and execution in the Internet of Things environment.
Fourth, the computing power of IoT terminals is constantly improving . Traditional IoT terminals are mainly responsible for data collection and transmission. However, with the continuous development and miniaturization of smart chips, embedded processors, sensing devices, etc., terminal devices are increasingly endowed with intelligent data processing capabilities. They can complete some data processing and intelligent reasoning tasks under cost constraints, which can provide support for improving the real-time performance of computing and protecting data privacy.
Fifth, the rise of edge computing and edge intelligence. Edge computing refers to computing performed at or near the physical location of the user or data source, providing edge intelligent data processing services locally, thus reducing latency and saving bandwidth. The rise of edge computing has further enhanced local data processing capabilities. Gartner listed edge computing as one of the top ten strategic technology trends in 2020③, and its emergence has solved the bottleneck problem in the development of the smart Internet of Things.
In summary, the processing and computing capabilities of traditional IoT architectures are insufficient to support the demands of deep coverage, massive connectivity, real-time processing, and intelligent computing in IoT networks. Against the backdrop of the development of terminal intelligence and edge computing, Artificial Intelligence of Things (AIoT, also generally referred to as AI+IoT or AI IoT) has gained widespread attention in recent years as a new trend in IoT development. AIoT, a concept that emerged in 2017①, is a product of the integration of artificial intelligence and IoT technologies, and is growing into a promising new frontier field, evolving from "interconnection of everything" to "intelligent interconnection of everything." According to Gartner's prediction, in the future, more than 75% of data will need to be analyzed, processed, and stored at the network edge. AIoT first collects various types of data (environmental data, operational data, business data, monitoring data, etc.) in real time through networked sensors, and then performs intelligent processing and understanding on terminal devices, edge devices, or in the cloud through data mining and machine learning methods. In recent years, AIoT applications have gradually integrated into various fields of major national needs and people's livelihoods, such as smart cities, intelligent manufacturing, and social governance.
The intelligent Internet of Things (IoT) brings new challenges such as ubiquitous intelligent sensing, context-adaptive communication, distributed swarm intelligence, and cloud-edge-device collaborative computing. Researchers from MIT, Stanford, Yale, UC Berkeley, Cambridge, and China have conducted systematic research in this cutting-edge field. For example, MIT researchers have systematically studied techniques such as deep model compression on resource-constrained IoT terminals. Yale researchers have proposed an efficient deep inference model for edge-device collaboration. A Stanford research team has studied the distributed collaborative learning capabilities among agents based on multi-agent deep reinforcement learning. Cambridge researchers have proposed a new method for lightweight automatic search of deep learning models in resource-constrained environments. Researchers at Hong Kong Polytechnic University have conducted in-depth analysis and exploration of the application of edge intelligent computing in the context of connected vehicles.
With the rapid development of AIoT, renowned IT companies both domestically and internationally have accelerated their deployments, achieving significant foundational results in areas such as edge intelligence, smart chips, and smart IoT software platforms. Microsoft officially released Azure IoT Suite in 2015. In 2021, it further released the new Azure Edge Zone edge computing platform to support real-time data processing. Amazon also pioneered the AWS IoT platform in 2015 and launched the FreeRTOS operating system in 2017, suitable for programming, deploying, connecting, and managing small, low-power edge devices. In 2018, Alibaba launched AliOSThings, an IoT operating system providing services such as IoT connectivity, intelligent processing, and cloud-edge-device collaborative computing. That same year, JD.com released its "City Computing Platform," combining deep learning and other technologies to build spatiotemporal correlation models and learning algorithms to solve intelligent application problems in various urban scenarios such as traffic planning, thermal power generation, and environmental protection. In 2019, Huawei launched HarmonyOS, a microkernel-based operating system designed for the Internet of Things. 5G's full-scenario distributed operating system, building upon the capabilities of traditional single-device systems, proposes a distributed concept based on a single system capability, adaptable to various terminal forms. In summary, the intelligent Internet of Things (IoT) has become a new development trend in both academia and industry. Therefore, this paper will focus on the interdisciplinary frontier of ubiquitous computing, artificial intelligence, and the IoT, elucidating its basic concepts, architecture, key technologies, and typical applications, and exploring its future scientific challenges and opportunities.
Intelligent Internet of Things Architecture
The core of the Internet of Things (IoT) is the information interaction between things and between people and things. The traditional IoT architecture consists of three layers: the perception layer, like the various sensory organs of a person, is composed of various sensor devices used to sense information such as temperature/humidity, pressure, light, air pressure, and stress conditions in the external environment; the network layer, equivalent to the human nervous system, is composed of various heterogeneous networks that transmit information from the perception layer to the application layer; and the application layer acts as a bridge between users and the IoT, providing application solutions for different industries through technologies such as cloud computing, big data, and middleware. The intelligent IoT, centered on data processing, faces new opportunities and challenges, and will form a new architecture and system software platform, which will be described in detail below.
The intelligent Internet of Things (AIoT) centers on efficient intelligent information and real-time processing. With the introduction of edge computing and edge intelligence, it will form a cloud-edge-device collaborative AIoT architecture. As shown in Figure 1, the system is divided into three layers, including the intelligent terminal layer, the edge intelligence layer, and the cloud computing layer.
The intelligent Internet of Things (IoT) is an intelligent system that combines hardware and software. On top of the cloud-edge-device collaborative intelligent IoT architecture, the software platform is a core component. The software platform provides interoperability between devices and applications, integrates heterogeneous computing and communication devices, simplifies application development, and provides interoperability between various applications and services running on heterogeneous devices. Generally, this manifests as middleware, such as a microservice framework.
The new characteristics of the Intelligent Internet of Things (AIoT), such as human-machine-object integration, ubiquitous computing, distributed intelligence, and cloud-edge-device collaboration, along with its system and software architecture that differ from traditional IoT, bring many new challenges. The following section briefly describes these challenges and related technologies. This section introduces key AIoT technologies from four levels: intelligent sensing, network communication, collaborative computing, and privacy protection, as shown in Figure 3.
Summarize
The Intelligent Internet of Things (AIoT) expands upon the three-layer architecture of IoT (sensing, network, and application), leveraging artificial intelligence (AI) technology and the sensing, storage, computing, and learning capabilities of ubiquitous IoT device platforms. Its goal is to achieve efficient, real-time, and intelligent processing of intelligent information, and it realizes intelligent enhancements in sensing, communication, computing, and applications based on a cloud-edge-device collaborative AIoT architecture. This paper elucidates the basic concept of the cloud-edge-device collaborative AIoT architecture and AIoT system software platform, and introduces key technologies and cutting-edge explorations in several areas, including ubiquitous intelligent sensing, swarm intelligence computing, swarm intelligence IoT communication, terminal-adaptive deep computing, IoT distributed learning, cloud-edge-device collaborative computing, and security and privacy protection. In the future, intelligent IoT research requires greater participation from researchers to delve into IoT system application problems, overcome key technological bottlenecks, and develop and refine general-purpose platforms. On the one hand, continuous breakthroughs are needed in key technologies such as hardware-software collaborative terminal intelligence, intelligent evolution for AIoT, next-generation intelligent IoT networks, continuous evolution of dynamic scene models, and human-machine-thing swarm intelligence computing. On the other hand, facing multimodal sensing, ubiquitous interconnection, dynamic scenes, and... Technical challenges such as resource constraints, real-time processing, and ubiquitous services necessitate the development of general-purpose AIoT operating systems, middleware, and other system platforms with characteristics such as "self-organization, configurability, and abstraction" to promote ecosystem development.
