As a novel paradigm, Vehicular Edge Computing (VEC) can effectively support computation-intensive or delay-sensitive applications in the Internet of Vehicles era. Computation offloading and resource management strategies are key technologies that directly determine the system cost in VEC networks. However, due to vehicle mobility and stochastic arrival computation tasks, designing an optimal offloading and resource allocation policy is extremely challenging. To solve this issue, a deep reinforcement learning-based intelligent offloading and power allocation scheme is proposed for minimizing the total delay cost and energy consumption in dynamic heterogeneous VEC networks. Specifically, we first construct an end-edge-cloud offloading model in a bidirectional road scenario, taking into account stochastic task arrival, time-varying channel conditions, and vehicle mobility. With the objective of minimizing the long-term total cost composed of the energy consumption and task delay, the Markov Decision Process (MDP) can be employed to solve such optimization problems. Moreover, considering the high-dimensional continuity of the action space and the dynamics of task generation, we propose a deep deterministic policy gradient-based adaptive computation offloading and power allocation (DDPG-ACOPA) algorithm to solve the formulated MDP problem. Extensive simulation results demonstrate that the proposed DDPG-ACOPA algorithm performs better in the dynamic heterogeneous VEC environment, significantly outperforming the other four baseline schemes.

The autonomous and security are the crucial requirements in fields of the polar transportation. This paper proposes a newly iterative learning control framework for the autonomous surface vessels (ASV) to implement the path-following operation in the ice floes scenario. The proposed framework is divided into two parts: the guidance and control. For the former, the ice floes are firstly identified into threatening and non-threatening based on the size. Subsequently, the obstacle area of each threatening ice floe is programmed considering the underwater portion. Then the ice floes avoidance guidance with auto-select mechanism for ice-zone traversal mission is constructed by setting the hazard threshold and target point. For the latter, a robust adaptive iterative learning control (ILC) system is designed for the path-following mission, where the control accuracy increases with the number of iterations. The stability of the closed-loop control system is proved with utilization of the Lyapunov theorem. Finally, two numerical examples are provided to evaluate the advantages and accuracy of the proposed algorithm, where the ice floes are generated with irregular.

This article investigates an adaptive prescribed performance path-following control algorithm for rotor-assisted vehicles, incorporating reinforcement learning (RL) to execute energy-saving cruising missions. For obtaining a high-performance path-following controller, a concise prescribed performance control (PPC) algorithm is designed to tightly constrain the output errors within the defined boundaries, while a shifting function is introduced to solve the problem of initial condition restrictions. Furthermore, through integrating the Backstepping method and the optimal control technique, an improved RL with the form of actor–critic neural networks (AC-NNs) is proposed to offer an innovative approach to the challenges of the model uncertainties and external disturbances. In this approach, the actor NN is employed to create an appropriate control policy, while the critic NN is aimed at evaluating the cost-to-go function to modify the system action. Semi-global uniform ultimate bounded (SGUUB) stable properties of the proposed algorithm are guaranteed via the Lyapunov theory. Finally, the superiority and feasibility of the proposed algorithm are verified by two numerical experiments.

In a cellular vehicle-to-everything (C-V2X) enabled computation-offloading network, vehicular users may deliver computation tasks to a cellular base station (BS) through vehicle-to-infrastructure (V2I) transmission links in order to accommodate computation-intensive applications, where the BS is usually equipped with a mobile edge computing (MEC) server. However, due to the broadcast nature of wireless communications and high-mobility of vehicles, the computation-offloading information may suffer from an eavesdropping threat. In practice, the interference generated by device-to-device-based V2V (D2D-V) links can be utilized for protecting the offloading information against eavesdropping. This observation motivates a security provisioning for C-V2X computation-offloading network. Specifically, a dynamic threshold-based access scheme is required to maintain the interference under control for different channel conditions. Unfortunately, the previous studies that are based on the assumption of perfect channel state information (CSI) can not be applied for the dynamic vehicular networks. In this paper, we propose a dynamic threshold-based access scheme for security provisioning of C-V2X computation-offloading network by considering an imperfect CSI. In this scheme, an optimized access threshold is set to update adaptively in terms of channel estimation error for balancing both the security and reliability of the offloading link. Furthermore, the proposed scheme can maximize the secrecy throughput under a connection outage constraint of the D2D-V links, with the total area spectral efficiency optimized under the security performance criterion for the offloading link. Numerical results are provided for validating the proposed theoretical analysis. A useful design insight is provided for attaining an optimal configuration of C-V2X computation-offloading network.

Channel estimation is considered a fundamental task in achieving reconfigurable intelligent surface (RIS)-aided communication systems. However, the location distribution of users in a cell is nonuniform, which leads to worse channel estimation performance for a single neural network. In this letter, we propose two federated learning (FL)-based hierarchical networks to improve channel estimation performance. Specifically, a hierarchical neural network (HNet) is proposed to enhance the channel estimation accuracy, which can perform different channel feature extraction and mapping tasks for users in different regions. Furthermore, a hierarchical residual neural network (HReNet) is presented to reduce the communication overhead during model training. Simulation results reveal that the FL-based HNet and HReNet schemes yield better channel estimation performance when users are nonuniformly distributed.