Core Concept of the Paper

Vehicular networks are entering a new phase with the rise of sixth-generation (6G) technologies. They demand not only faster communication but also smarter resource management. However, transmitting massive amounts of raw data leads to long delays and heavy energy consumption. Moreover, these networks must operate in real time, where even a small delay of milliseconds can affect safety and efficiency. To address these challenges, emerging technologies such as digital twins, semantic-based communication, and quantum computing are essential. Digital twins can create a virtual replica of the network and predict changes before they happen. Thus, in a vehicular environment, vehicles and servers can prepare resources in advance. Moreover, semantic communication focuses on the meaning of data rather than its size. This approach reduces the need to send every raw bit, while keeping only the useful information. Furthermore, quantum computing offers new power for decision making. It provides unique features such as superposition and entanglement, which allow quantum models to represent richer states and capture correlations more effectively. 

Therefore, this paper presents a vehicular network framework that comprises vehicles and roadside units equipped with edge computing servers, designed to offload semantic-aware tasks for processing. The framework further integrates digital twin prediction, semantic-based communication, and quantum computing-based decision making. By leveraging the above concepts, a cost-based task management optimization problem is formulated to jointly minimize latency and energy consumption, while maintaining communication quality. The optimization problem is modeled as a mixed-integer nonlinear programming (MINLP) problem, which is complex and intractable to solve directly. To address this challenge, a hybrid quantum-enhanced deep deterministic policy gradient (Q-DDPG) reinforcement learning algorithm is proposed. Simulation results demonstrate that Q-DDPG significantly outperforms the baseline DDPG, achieving lower system cost, reduced delay, and improved energy efficiency. Furthermore, the framework adapts well under varying task complexities and bandwidth conditions.

Overall, this work highlights the value of integrating semantic communication, digital twin prediction, and quantum-enhanced reinforcement learning. It offers a scalable and intelligent solution for task management in next-generation vehicular networks, where efficiency and meaning are equally critical.

View this article on IEEE Xplore

Cooperative multi-UAV networks have emerged as a transformative technology, revolutionizing critical sectors such as disaster response, environmental monitoring, and intelligent logistics. By harnessing the cooperation of multiple UAVs, these networks exhibit unparalleled operational agility, efficiency and resilience in dynamic environments, where seamless cooperation relies on high-accuracy UAV formation. However, the interdependent information flow between localization and motion planning complicates system design, often leading to degraded formation accuracy due to error accumulation. To address this issue, this paper presented an integrated localization and motion planning framework, enabling UAV formations to maintain high precision under noisy conditions.

By deriving the bounds for formation errors, we revealed the impact of measurement and motion noises on the formation accuracy, i.e., only a subset of the noises affects relative formation accuracy, which is invariant to translation and rotation. Such finding provided criticial insights into optimizing formation accuracy by minimizing the propagation of localization errors. Then, we developed tightly coupled formation algorithms to address measurement and motion noises, combining forward motion planning and backward resource allocation. For the forward process, we proposed a near-optimal motion planning algorithm that exploits relative formation equivalence to mitigate localization errors; while for the backward process, we designed wireless resource allocation algorithms that optimize bandwidth allocation and UAV activation to maximize formation accuracy under resource constraints.

Overall, this work introduces a new framework for integrated multi-agent system design, bridging localization and motion planning to enhance formation accuracy. The simulation results demonstrated its effectiveness: the proposed motion planning algorithm nearly approaches the performance of brute-force search, reducing the formation error by around 16%, while our bandwidth allocation and UAV activation strategies achieves and 25% error reduction, outperforming traditional approaches.

View this article on IEEE Xplore

Non-Intrusive Load Monitoring (NILM) seeks to disaggregate aggregate household electricity consumption, as recorded by smart meters, into individual appliance usage. Providing real-time feedback on each appliance’s consumption enables users to optimize energy use and reduce costs. Conventional NILM methods depend on dedicated sensors for each appliance to obtain detailed training data, incurring high costs, intrusiveness, and potential circuit damage. To address these issues, we propose a novel framework called State-based Supervised NILM (SS-NILM).

Rather than depending on precise power measurements for each device, SS-NILM leverages binary on-off state information (e.g., active or idle) to train its models. This state data can be collected non-intrusively via user-reported activity logs or equipment-generated timestamps. The core innovation lies in a dual neural network architecture that jointly predicts appliance-level power consumption and their corresponding on-off states. Model training is guided by enforcing consistency between the aggregated predicted appliance consumption and the observed total household load. Additional constraints ensure prediction feasibility by disallowing unrealistic outputs such as negative power values or overestimation of aggregate usage.

Empirical evaluations on both residential and industrial datasets (e.g., REDD, UK-DALE, HIPE) demonstrate that SS-NILM achieves performance comparable to conventional sensor-dependent methods, as measured by metrics such as Mean Absolute Error (MAE) and correlation scores. Crucially, it eliminates the need for intrusive sensor deployment, significantly enhancing system scalability by allowing new appliances to be integrated without hardware adjustments.

The implications are substantial: SS-NILM reduces adoption barriers for energy-monitoring systems, making them more viable for both households and industrial facilities. Future research may focus on improving computational efficiency and extending the model to support multi-state appliances (e.g., low-power modes) for enhanced prediction fidelity. Overall, this work offers a practical advance toward smarter, safer, and more sustainable energy management.

View this article on IEEE Xplore

The COVID-19 pandemic has underscored the critical need for sustainable mobility interventions that balance epidemic containment with economic stability. However, developing practical, cost-effective capacity limitation strategies for Points of Interest (POIs) in complex urban environments remains a challenging open problem. To address this, this study investigates a novel network immunity problem: optimizing POI capacity limitation measure within urban mobility networks. The aim is to maximize epidemic control under fixed resource constraints. To achieve this, we develop a dynamic metapopulation SEIR ordinary differential equation model that integrates real-world inter-POI mobility networks. The model demonstrates strong adaptability to significant changes in human mobility patterns before and after the epidemic. It outperforms baseline approaches in fitting the daily new case data of COVID-19 in Beijing. Leveraging this model, we derive the generalized basic reproduction number  for urban networks, which quantifies outbreak risk based on POI connectivity, visitor density, and transmission dynamics. Consequently, we reformulate the optimization problem as minimizing  under budget constraints. We propose a Greedy Capacity Reduction Algorithm (GCRA) to approximately solve these problems. The algorithm relies on the visitor volume of each POI and directly utilizes  as the optimization objective. Employing a greedy strategy, it iteratively identifies and reduces the capacity thresholds of key POIs to attain maximum epidemic control efficacy within a fixed budget. Subsequently, extensive experimental validation is conducted on large-scale urban networks, which connect 4,335 residential communities to 14,936 POIs with 5.7 million daily edges. The results demonstrate that, in comparison to the baseline algorithms, the proposed algorithm shows significant superiority and accuracy in reducing  and containing the epidemic. Specifically, GCRA effectively postpones the epidemic peak by  days and curtails the total number of cases by . We also validate the scalability of the GCRA on inter-county networks in Ontario, Canada, and inter-province networks in Italy, demonstrating its broad applicability. In general, the proposed approach enables cities to mitigate epidemic spread without imposing widespread mobility restrictions, offering policymakers a data-driven tool to prioritize targeted interventions while preserving economic activity. By bridging theoretical epidemiology with practical urban mobility data, this study offers robust technical and empirical support for future epidemic response strategies.

View this article on IEEE Xplore