Internet of Things (IoT) is an important technology employed in a variety of different applications, such as transportation, healthcare, and manufacturing. In recent years, the number of IoT devices deployed globally has been increasing at a rapid pace and is estimated to reach 20 billion by the end of 2025. In modern industry, IoT plays a pivotal role by monitoring the condition of industrial machines and, consequently, improving the efficiency of industrial processes. To optimize the efficiency of industrial IoT applications, various Artificial Intelligence (AI) techniques have been adopted, leading to a new computing paradigm, namely, Industrial Artificial Intelligence of Things (i.e. Industrial AIoT). In this paper, we describe the challenges to tackle and the opportunities to explore in Industrial AIoT. Specifically, we first review the use of state-of-the-art AI methods in Industrial AIoT applications, with a focus on Deep Learning (DL) and Machine Learning (ML) techniques. Thereafter, we present a series of important applications of Industrial AIoT. The key challenges associated with the implementation of Industrial AIoT applications are also discussed. In addition, the societal and economic impacts of Industrial AIoT are briefly described. Finally, we outline the future research directions in Industrial AIoT, which should be further investigated to fully utilize the potential of this innovative technology.

Over the past years, computationally-intensive mobile applications, such as interactive games and augmented reality, have gained enormous popularity. This phenomenon has placed a serious burden on mobile devices with limited computation resources and constrained battery capacity. Multi-access Edge Computing (MEC) is proposed to solve the problem by offloading part of the computation tasks from mobile devices to edge servers. The fundamental challenge in MEC is how to effectively select a subset of computation tasks to be offloaded so that the application requirements are satisfied and the total energy consumption is minimized. The existing Deep Reinforcement Learning (DRL) based offloading schemes focus on either non-real-time tasks or real-time tasks with soft deadlines. In addition, the existing schemes do not work well when the information of the system environment is not complete. In this paper, we propose an innovative DRL-based task offloading method, PDMO, which guarantees that the deadlines of real-time tasks are met even when the system environment is only partially observable. Our experimental results indicate that PDMO works well in partially observable environments. In addition, it outperforms the existing offloading schemes in terms of energy consumption, deadline miss number and completion rate of non-real-time tasks.

In this article, we consider a parking lot that manages the charging processes of its parked electric vehicles (EVs). Upon arrival, each EV requests a certain amount of energy. This request should be fulfilled before the EV’s departure. It is of critical importance to coordinate the EVs’ charging rates to smooth out the load profile of the parking lot because inappropriate charging rates can lead to sharp spikes and fluctuations on the load profile, imposing negative effects on the power grid. Meanwhile, empirical studies show that many parking lots exhibit statistical patterns on EV dynamics. For example, the bulk of EVs arrives during rush hours. Therefore, in this article, we incorporate such patterns into charging rate coordination. Although the statistical patterns can be summarized from historical data, they are difficult to be analytically modeled. As a result, we adopt a model-free deep reinforcement learning approach. We also take the latest continuous charging rate control technology into consideration. The decision variables are thus continuous and a policy gradient algorithm is needed to perform reinforcement learning. Technically, we first formulate the problem as a Markov decision process (MDP) with unknown state transition probabilities. To further derive a deep policy gradient algorithm, the challenge lies in the inconsistent and state-dependent action space of the MDP model, due to the constraint to satisfy EVs’ energy demands before their scheduled departure. To tackle the challenge, we design a customized model for neural network training by extending the action space to be consistent and state independent, and revise the reward function to penalize the neural network output if it is beyond the action space of the original MDP model. With this customized model, we then develop a deep policy gradient algorithm based on the proximal policy gradient framework. Numerical results show that our algorithm outperforms the benchmarks.

In this paper, we investigate a non-orthogonal multiple access (NOMA) assisted multi-antenna two-way relay system, where multi-antenna users apply NOMA to support bidirectional superposition transmission via multiple multi-antenna relays. Specifically, we propose two cooperative strategies, namely multiple-access broadcast NOMA and time division broadcast NOMA. For each of the two cooperative strategies, we devise a joint antenna-and-relay selection scheme to enhance the transmission reliability. Analytical expressions for the outage probability and diversity order are derived to evaluate the system performance achieved by the proposed cooperative strategies with the corresponding joint antenna-and-relay selection schemes. To further reduce the outage probability, we use the derived analytical results as objective functions to optimize the transmit power allocation under both cooperative strategies. Finally, extensive simulations are carried out to validate the accuracy of the derived analytical results. Our simulation results indicate that the proposed strategies significantly outperform existing benchmark strategies in terms of outage probability and diversity order.

Multi-access Edge Computing (MEC) is a promising solution to the resource shortage problem on mobile devices. With MEC, a fraction of the computational tasks on mobile devices could be offloaded to edge servers. Over the past years, a series of machine learning based offloading methods for MEC have been proposed to reduce the completion time of computational tasks. However, most of the existing methods do not work well for realtime applications, which involve tasks with rigorous deadline constraints. In addition, offloading data-intensive tasks via the latest wireless networks, such as LTE and 5G, could lead to unsatisfactory transmission delays. Furthermore, with the state-of-the-art learning-based methods, both the training and inference operation of the learning algorithm are carried out on mobile devices, undesirably leaving less computation resources for computational tasks on mobile devices. In this paper, we propose a 6G-empowered learning-based offloading scheme, MELO, which can be used to make appropriate offloading decisions for realtime tasks. Specifically, the task offloading problem is first formulated as a Markov Decision Process. Thereafter, the problem is solved with a Reinforcement Learning (RL) algorithm, TD3. In addition, 6G is adopted as the communication infrastructure to sufficiently support the data transfer between mobile devices and edge servers. Furthermore, to leave more resources on mobile devices, we devise a novel learning architecture, EALA. With EALA, the training and inference operation of a learning algorithm are decoupled. The training operation is carried out on edge servers while the inference operation is performed on mobile devices. Our experimental results indicate that MELO outperforms the existing offloading methods in terms of task completion time.

Cell-free massive multiple-input--multiple-output (MIMO) (CF-mMIMO) is a distributed massive MIMO network architecture, which leads to a better ability to resist shadow fading than centralized massive MIMO. When network slicing is applied to CF-mMIMO networks in order to serve multiple tenants, the deployment flexibility of CF-mMIMO networks could be further improved. Nevertheless, how to effectively allocate resources in network slicing based CF-mMIMO networks is still a challenging problem to be tackled. In this article, we present a novel resource allocation scheme for network slicing based multi-tenant CF-mMIMO networks to maximize the infrastructure operator’s revenue while maintaining the quality of network services. Specifically, the resource allocation scheme is first formulated as an optimization problem, which is NP-hard and nonconvex. Then, the problem is decomposed into two subproblems. For the first subproblem, we propose a joint user clustering and power control coefficient allocation algorithm, which not only maintains fairness among users but also improves the access rate of CF-mMIMO networks. For the second subproblem, the interior-point method is adopted. In our article, extensive simulations are carried out to prove the effectiveness of the proposed resource allocation scheme. Our experimental results indicate that the proposed scheme leads to a lower outage probability and a higher infrastructure provider’s revenue than existing approaches.