The advent of Large Multimodal Models (LMMs) offers a promising technology to tackle the limitations of modular design in autonomous driving, which often falters in open-world scenarios requiring sustained environmental understanding and logical reasoning. Besides, embodied artificial intelligence facilitates policy optimization through closed-loop interactions to achieve the continuous learning capability, thereby advancing autonomous driving toward embodied intelligent (El) driving. However, such capability will be constrained by relying solely on LMMs to enhance EI driving without joint decision-making. This article introduces a novel semantics and policy dual-driven hybrid decision framework to tackle this challenge, ensuring continuous learning and joint decision. The framework merges LMMs for semantic understanding and cognitive representation, and deep reinforcement learning (DRL) for real-time policy optimization. We starts by introducing the foundational principles of EI driving and LMMs. Moreover, we examine the emerging opportunities this framework enables, encompassing potential benefits and representative use cases. A case study is conducted experimentally to validate the performance superiority of our framework in completing lane-change planning task. Finally, several future research directions to empower EI driving are identified to guide subsequent work.

Emerging virtual reality (VR) applications enable users to immerse themselves in an artificial world. However, the stringent requirements for ultra-high bandwidth and ultra-low latency streaming pose significant challenges to current radio frequency (RF)-based wireless networks. The convergence of visible light communication (VLC) and RF networks synergizes their complementary advantages in terms of enhanced channel capacity and ubiquitous coverage, thereby holding substantial potential for 360$^{\circ }$ VR video streaming. Moreover, the holistic integration of users' multidimensional perceptual needs of latency and video quality is critical to enhancing immersive experiences, which necessitates the human-centric design. Therefore, this paper presents a latency- and quality-efficient human-centric VR video streaming scheme over an indoor hybrid VLC-RF network. The VR video streaming issue is addressed to maximize the overall human-centric streaming utility by jointly designing the user's resolution adaptation, user association, and transmit power allocation of the VLC and RF access points. The formulated problem is a binary constrained and multi-variable coupled optimization problem, and is then decomposed into two subproblems. An overall iterative algorithm is proposed to solve these subproblems jointly. Simulation results illustrated that the proposed scheme achieves significant improvement in overall performance compared to the baselines.

The Networked Blue Economy (NetBE), including maritime activities from emergency response and infrastructure inspection to smart aquaculture, depends on pervasive underwater sensing and communications. However, the energy constraint of battery-powered underwater nodes continues to be a performance bottleneck, limiting the scale and sustainability of deployment. To address this limitation, underwater acoustic backscatter (UAB) has emerging as an ultra-low-power communication paradigm that enables a scalable and energy-sustainable Internet of Underwater Things, as a sensing platform of the NetBE. Nevertheless, the shift from theoretical investigation to practical application remains a daunting task. Therefore, this article gives a thorough overview to explore how UAB can empower the NetBE. First, we present specific energy problems of underwater systems. Then, we detail the fundamental operational principles of UAB, covering its system components, communication process, and configuration mode. On this basis, we propose a concrete technological pathway for the deployment of UAB into the NetBE, which discusses network architectures and key enabling techniques, supported by an analysis of its associated use cases. A case study confirms a significant improvement in system energy efficiency with an acceptable per-backscatter node average rate, demonstrating the potential of UAB for sustainable future of the NetBE. Finally, we conclude by some research challenges and prospective directions for future explorations.

Integrating multimodal semantic communications with lane-change (LC) planning has been considered as a promising framework, enabling the vehicle-road collaborative autonomous driving with higher degree of efficiency and flexibility. However, it is crucial to dynamically adjust resource allocation based on multimodal semantic representations for adapting to the dynamic and complex traffic situations, which requires the co-design of motion planning and resource allocation. In this paper, we target a joint optimization problem of the LC planning as well as the bandwidth and transmit power allocation at the modality transmitters of automated vehicles to minimize the overall latency of LC task completion. To tackle the high-dimensional, dynamic, and uncertain challenges of this problem, a deep reinforcement learning-based co-design scheme which employs Dueling Double Deep Q-Network (D3QN) algorithm is proposed. Simulation results illustrate the superiority of our proposed scheme, achieving the overall latency reduction by at least 4.13 ms and the percentage of latency improvement by up to 98.21% over the benchmarks.

Integrating Generative Artificial Intelligence (GAI) into Industry 5.0 presents transformative potential for Cyber-Physical Manufacturing (CPM) systems, enabling the human-machine interactive operations with higher degree of efficiency and productivity. However, deploying Artificial Intelligence Generated Content (AIGC) services in the CPM systems is fraught with challenges, such as resource constraints, extensive model sizes, and stochastic task generation. Therefore, it is essential to provision AIGC services through a co-design of fine-tuning service association, fine-tuned model transmission, and resource allocation. Notably, Industry 5.0’s emphasis on human-centricity allows humans to experience a more smooth operational interaction with smart machines (SMs), which necessitates shortening the completion latency of AIGC tasks. In this paper, we propose a human-centered AIGC service provision scheme for jointly designing the AIGC service associations of fine-tuning and model transmission at the SMs and edge servers (ESs), as well as the bandwidth and transmit power allocations at the ESs. We target the complete workflow particularly with model finetuning, model transmission, and inference execution. On this basis, we formulate the AIGC service provision problem for average task completion latency minimization, and develop a hierarchical deep reinforcement learning algorithm to solve it. Specifically, the discrete action policy of AIGC service association is learned via Dueling Double Deep Q-Network (D3QN) in the first layer, to guide the training of the second layer that learns the continuous action policy of resource allocation by soft actor-critic (SAC). Moreover, to improve the training stability and accelerate convergence, curriculum learning is adopted during the training phase to progressively update the neural networks of both D3QN and SAC. Simulation results show that the proposed scheme outperforms benchmark schemes, achieving a significant reduction in average task completion latency.

Integrated Sensing and Communication (ISAC) is driving the evolution of edge intelligence. In ISAC-enabled wireless edge networks, federated learning (FL) is crucial for realizing edge intelligence by supporting the networks with privacy protection, efficient data management, and dynamic adaptability. Specifically, FL allows distributed computing nodes (e.g., sensor devices) to first train local models by using data collected or sensed via ISAC and subsequently send them to one or multiple aggregation nodes for global model collaboration. However, traditional FL frameworks face significant challenges in the ISAC scenarios. For example, the privacy sensitivity of heterogeneous sensor data and the lack of transparency in model parameter exchange make it difficult to ensure the credibility of local and global models. Sharding distributed ledger technology (DLT), which divides the ledger into smaller and manageable shards, offers a potential solution to address these challenges by utilizing multi-node trust capabilities to facilitate distributed consensus during FL training. In this paper, we propose a trusted FL framework that incorporates sharding DLT within ISAC-enabled wireless edge networks to enhance both model training and consensus performance. Specifically, we develop a theoretical model to examine the interactions between model training performance and network capacities of sensing nodes (e.g., storage, computing, and communication capabilities) based on ISAC’s real-time channel state information. Based on this theoretical model, we design a trusted clustering scheme for aggregating local models. Numerical results demonstrate that in ISAC-enabled wireless edge networks, our proposed scheme significantly increases network throughput for model transmission while ensuring optimal model learning performance compared to some classical baselines.

With the proliferation of data-intensive industrial applications, the collaboration of computing powers among standalone edge servers is vital to provision such services for smart devices. In this paper, we propose an edge-driven industrial computing power network (CPN) by orchestrating the computing and network resources of edge servers through the centralized resource scheduling and decentralized task computing. However, efficient task offloading and collaborative processing is challenging, which requires higher degrees of network automation and intelligence. Therefore, we incorporate digital twins (DTs) into the edge-driven CPN architecture, where the DTs are created as the digital replicas to assist both the computation offloading and collaborative processing. A joint optimization problem of the computing power assignment, service association, task partition, and transmit power control is formulated for maximizing the system average weighted utility. Due to the temporal-spatial variability of tasks and the resulted dynamic environment, we transform the original problem as a Markov decision process aiming at maximizing the long-term average weighted utility. To efficiently handle the high-dimensional discrete-continuous action space, a hybrid soft actor-critic based deep reinforcement learning algorithm is developed for optimizing the joint design. Simulation results validate the superiority of our proposed algorithm over the benchmarks, showing the significant gains obtained by integrating DTs into the edge-driven CPN.

With the rapid development of vehicular networks, the computational capabilities and application scenarios of vehicles are becoming increasingly diverse, leading to a continuous emergence of complex computational tasks. Facing these tasks, a single vehicle node often struggles to handle them effectively; thus, it is necessary to offload tasks to other vehicles with computational resources through Vehicle-to-Vehicle (V2V) communication. However, due to the mobility of vehicles and the limitations of computing and communication resources, efficiently completing these complex computational tasks presents a significant challenge. To address this, this paper proposes an innovative optimization scheme that combines Digital Twin (DT) technology with vehicular edge computing. It constructs digital twins of vehicles through Roadside Units (RSUs) and utilizes these digital twins to optimize task offloading strategies. The scheme aims to jointly optimize transmission power, task offloading ratios, and computational resource allocation to minimize the impact of communication constraints and vehicle mobility on task completion delay. The paper models the wireless communication channel between vehicles using the Nakagami-m fading model, taking into account both transmission delay and computation delay in the overall task completion time. To solve this non-convex optimization problem, we introduce a joiSACnt optimization framework based on the Soft Actor-Critic (SAC) algorithm for efficient task allocation and dynamic transmission power adjustment. The simulation results show that the proposed scheme significantly reduces the maximum task delay and improves overall communication efficiency, particularly when compared with baseline schemes without power optimization and digital twin modules, as well as the DQN and DDPG algorithms. It demonstrates better task processing efficiency and communication performance, providing an effective solution for task handling in vehicular networks.

The collaboration of computing powers (CPs) among uncrewed aerial vehicles (UAVs)-mounted edge servers is essential to handle data-intensive tasks of user equipments (UEs). This paper presents a multi-UAV computing power network (CPN) that orchestrates the CP resources of edge servers to cooperatively process tasks through collaborative computing. Due to the frequent interactions among edge servers and the increased complexities for large-scale task executions, we propose a digital twin (DT)-driven multi-UAV CPN framework, where the aerial DT network is created as a digital replica of the multi-UAV network to assist collaborative computing. A joint optimization problem of the service association, offloading power control, task partitions, CP allocation, and UAV trajectory design is formulated to minimize the long-term average delay of UEs. We use the Lyapunov technique to tackle the long-term energy constraint, forming a tractable optimization problem, which is modeled as a Markov decision process to balance the UAV's deficit queue stability and the UE's task completion delay. A hybrid soft actor-critic based deep reinforcement learning algorithm is developed to optimize the joint design by sampling the hybrid action from the Gumbel and Gaussian distributions. Simulation results verify that the multi-UAV CPN contributes to significant performance gains when supported by DTs.

Data-intensive smart applications are currently driving the emergence of edge-enabled computing power network (Edge-CPN) by orchestrating the computing powers (CPs) of edge servers, enabling the converged computing and networking at the edge. Besides, the proliferation of these applications and various smart devices (SDs) is arousing great interest in the joint training of a shared global model by massive SDs via federated learning (FL). Due to the heterogeneity and constrained resources of SDs, the FL performance in the Edge-CPN suffers from the straggler effect, reducing the efficiency of global model aggregation. To tackle this challenge and upgrade the training mode into higher degrees of efficiency and intelligence, in this article, we propose the TwinFed, a novel digital twin (DT)-driven FL framework, which fully leverages ubiquitous CPs to configure the DTs for assisting model training of stragglers. An interplay between the end and edge layers is captured into the architecture design via the hierarchical model aggregation. We develop a unified twinning pipeline to achieve the high-fidelity DTs and efficient model training. A two-stage workflow is also introduced to implement TwinFed by flexibly integrating the computing resource orchestration and training process. Finally, we conduct a case study for anomaly detection in smart factory to validate the superiority of TwinFed in testing accuracy and training loss.

In this letter, we study the pilot assignment scheme in the uplink cell-free massive multiple-input multiple-output system. To accurately depict the potential co-pilot interference (CPI) among the user terminals (UTs), we introduce the normalized contribution value of AP to the received signal strength at the UT. A weighted conflict clustering graph is constructed by using the potential CPI to define the weight of the edge that connects the co-pilot UTs. Then, the pilot assignment optimization is transformed into a Min-Cut problem, which is further converted as a more effective normalized cut to find a partition of the UTs into disjoint clusters for minimizing the overall CPI. We develop a two-stage algorithm via spectral clustering to solve the relaxed problem efficiently. Numerical results are provided to validate the superiority of the proposed algorithm as compared to the existing methods in terms of the uplink average sum-rate and the quality of channel estimation.

This paper considers a two-hop wireless powered relay network consisting of multiple sources, multiple destinations, and one relay. The relay can receive energy from the sources and forward data to the destinations. We focus on the source selection problem during the energy transfer process and the resource allocation problem during the data transmission process. Firstly, the relay can choose among all sources based on the transferred energy from the sources. A credit mechanism is introduced for the relay to achieve optimal selection. Secondly, a Stackelberg differential game-based model is adopted for the resource allocation problem in the data transmission process, using the differential equation to describe the dynamic variation of energy, and the Stackelberg game to describe the relationships between the sources and the relay. In the proposed approach, both sources and relays consider energy consumption and energy revenue. To find the optimal solutions, an adaptive dynamic programming-based algorithm is utilized. The Lyapunov-based stability analysis shows that the system has uniform ultimate boundedness and convergence. Finally, the trained neural networks can achieve optimal resource allocation strategies. Through extensive simulation experiments, the effectiveness of the proposed algorithm is verified.

This letter considers a reconfigurable intelligent surface (RIS)-aided indoor visible light communication system, where a mirror array-based RIS is deployed to assist the communication from a light-emitting diode (LED) to multiple user terminals (UTs). We aim to maximize the sum-rate in an entire serving period by jointly optimizing the orientation of the RIS reflecting unit, the time fraction for the UT, and the transmit power at the LED, subject to the communication and illumination intensity requirements. To solve this high-dimensional non-convex problem, we transform it as a constrained Markov decision process. Then, a soft actor-critic (SAC)-based deep reinforcement learning algorithm is proposed with the goal of maximizing both the average reward and the expected policy entropy. Simulation results prove the effectiveness of the proposed SAC-based joint optimization design in improving the sum-rate and long-term average reward.

Underwater acoustic backscatter communication, which allows an underwater sensor node (USN) to communicate with the surface sink node (SN) by reflecting the acoustic signal from the SN, is a promising solution for enabling the Internet of Underwater Things. However, underwater backscattering is vulnerable to jamming attacks, especially from a gliding autonomous underwater vehicle (AUV) that can inject jamming signal to the underwater acoustic channel accordingly. This paper considers the attack-defense interactions between the AUV jammer and the SN defender in their transmit power allocations of jamming signals and acoustic signals over multiple USNs. We formulate the strategic power allocation problem for the SN and AUV as an asymmetric anti-jamming Colonel Blotto (CB) game model with a finite budget of total transmit powers. In particular, the competitive interactions between the SN and the AUV is modeled as the competition of two players in the game for limited resource budgets over the battlefield set. We analyze the mixedstrategy Nash equilibrium to the game, and obtain the closedform performance bound of the defense power allocation strategy. Numerical results show the superiority of the proposed CB game based multi-USN secure transmission scheme as compared to the benchmarks in terms of the expected sum-utility.

A combination of Industrial Internet of Things (IIoT) and federated learning (FL) is deemed as a promising solution to realize Industry 4.0 and beyond. However, scheduling more IIoT devices engaged in FL contributes to accelerated learning speed, but resulting in increased learning cost in terms of energy consumption and model accuracy reduction. In this article, we investigate the tradeoff between learning speed and cost in a three-layer FL-enabled IIoT system. Particularly, a weighted learning utility function is designed by capturing such a tradeoff. We aim to maximize the weighted learning utility in an FL training round by jointly optimizing the edge association as well as the allocations of resource block, computation capacity, and transmit power of the IIoT device. The resulting problem is a nonconvex and mixed-integer optimization problem, and consequently, it is difficult to solve. We thereby decompose the original problem into three subproblems, and then propose an overall alternating optimization algorithm to solve the subproblems iteratively until convergence. Via experimental results, it is demonstrated that the proposed scheme significantly improves the system-wide learning utility as compared to other baseline schemes. It is also shown that the proposed scheme can achieve the optimized tradeoff between learning speed and learning cost.

Mobility generally may lead to the continual loss of data and link interruptions during the communications between secondary users (SUs) in distributed underlay cognitive radio ad hoc networks (CRANETs), which raises important concerns about the stability for routing design and maintenance. In this paper, we investigate the effect of path stability on multi-path routing in distributed underlay CRANET scenario. We firstly model the stability factor of the initial state of each path as the product of the stability coupling factor of the sorted link on this path. Then, we obtain the predicted stability of each path by using the risk level of link based on a second-order Markov predictor by taking into account the relative movement direction of SUs in the proposed Markov model. Particularly, we characterize the overall stability of each path on the basis of the predicted stability of this considered path. Moreover, an effective algorithm is presented to calculate the route stability of the path with the help of the Markov predictor. We also investigate the selection problem of the optimal path through the proposed optimal path selection algorithm. Numerical results are presented to demonstrate the effectiveness and practicality of our proposed algorithm, which achieves the evaluation of the overall stability for multi-path routing.

Sixth generation (6G) wireless networks require very low latency and an ultra-high data rate, which have become the main challenges for future wireless communications. To effectively balance the requirements of 6G and the extreme shortage of capacity within the existing wireless networks, sensing-assisted communications in the terahertz (THz) band with unmanned aerial vehicles (UAVs) is proposed. In this scenario, the THz-UAV acts as an aerial base station to provide information on users and sensing signals and detect the THz channel to assist UAV communication. However, communication and sensing signals that use the same resources can cause interference with each other. Therefore, we research a cooperative method of co-existence between sensing and communication signals in the same frequency and time allocation to reduce the interference. We then formulate an optimization problem to minimize the total delay by jointly optimizing the UAV trajectory, frequency association, and transmission power of each user. The resulting problem is a non-convex and mixed integer optimization problem, which is challenging to solve. By resorting to the Lagrange multiplier and proximal policy optimization (PPO) method, we propose an overall alternating optimization algorithm to solve this problem in an iterative way. Specifically, given the UAV location and frequency, the sub-problem of the sensing and communication transmission powers is transformed into a convex problem, which is solved by the Lagrange multiplier method. Second, in each iteration, for given sensing and communication transmission powers, we relax the discrete variable to a continuous variable and use the PPO algorithm to tackle the sub-problem of joint optimization of the UAV location and frequency. The results show that the proposed algorithm reduces the delay and improves the transmission rate when compared with the conventional greedy algorithm.

In this paper, a laser-powered aerial mobile edge computing (MEC) architecture is proposed, where a high-altitude platform (HAP) integrated with an MEC server transfers laser energy to charge aerial user equipments (AUEs) for offloading their computation tasks to the HAP. Particularly, we identify a new privacy vulnerability caused by the transmission of wireless power transfer (WPT) signaling in the presence of a malicious smart attacker (SA). To address this vulnerability, the interaction between the HAP and the SA in their allocation of tile grids as charging points to the AUEs in laser-enabled WPT is formulated as a Colonel Blotto game (CBG), which models the competition of two players for limited resources over multiple battlefields for a finite time horizon. Moreover, the utility function that each player receives over a battlefield is developed by identifying the tradeoff between privacy protection level and energy consumption of each AUE. We further obtain the mixed-strategy Nash equilibrium for the modified CBG with asymmetric players. Simulation results are presented to show the effectiveness of this game framework.

Integrating digital twins (DTs) and multi-access edge computing (MEC) is a promising technology that realizes edge intelligence in 6 G, which has been recognized as the key enabler for Industrial Internet of Things (IIoT). In this paper, we explore a DT-assisted MEC system for the IIoT scenario where a DT server is created as a virtual representation of the physical MEC server, via estimating the computation state of the MEC server within the DT modelling cycle. To achieve spectrally efficient offloading, we consider that IIoT devices communicate with industrial gateways (IGWs) through a non-orthogonal multiple access (NOMA) protocol. Each IIoT device has an industrial computation task that can be executed locally or fully offloaded to IGW. We aim to minimize the total task completion delay of all IIoT devices by jointly optimizing the IGW's subchannel assignment as well as the computation capacity allocation, edge association, and transmit power allocation of IIoT device. The resulting problem is shown to be a mixed integer non-convex optimization problem, which is NP-hard and challenging to solve. We decompose the original problem into four solvable sub-problems, and then propose an overall alternating optimization algorithm to solve the sub-problems iteratively until convergence. Validated via simulations, the proposed scheme shows superiority to the benchmarks in reducing the total task completion delay and increasing the percentage of offloading IIoT devices.

Federated learning (FL) is a key solution to realizing a cost-efficient and intelligent Industrial Internet of Things (IIoT). To improve training efficiency and mitigate the straggler effect of FL, this paper investigates an edge-assisted FL framework over an IIoT system by combining it with a mobile edge computing (MEC) technique. In the proposed edge-assisted FL framework, each IIoT device with weak computation capacity can offload partial local data to an edge server with strong computing power for edge training. In order to obtain the optimal offloading strategy, we formulate an FL loss function minimization problem under the latency constraint in the proposed edge-assisted FL framework by optimizing the offloading data size of each device. An optimal offloading strategy is first derived in a perfect channel state information (CSI) scenario. Then, we extend the strategy into an imperfect CSI scenario and accordingly propose a Q-learning-aided offloading strategy. Finally, our simulation results show that our proposed Q-learning-based offloading strategy can improve FL test accuracy by about 4.7% compared to the conventional FL scheme. Furthermore, the proposed Q-learning-based offloading strategy can achieve similar performance to the optimal offloading strategy and always outperforms the conventional FL scheme in different system parameters, which validates the effectiveness of the proposed edge-assisted framework and Q-learning-based offloading strategy.