Tactile Internet, as a novel industrial network, allows fully immersive multisensory remote exploration of real or virtual environments. An important technological aspect in tactile Internet is the acquisition, compression, transmission, and display of haptic information. This article focuses on the cross-modal acquisition of fingertip’s tactile friction from visual measurements. In tactile Internet applications, these tactile friction data are transmitted to surface haptic devices for high-fidelity haptic rendering of shapes and textures on touchscreens. To ensure the reliability and latency for such tactile friction acquisition, we develop a federated closed-loop learning (FedCLL) method that is based on the optimized federated learning and the closed-loop learning. The former builds the global model in the centric server, by utilizing deep reinforcement learning to determine aggregation weights of local tactile devices, which improves the acquisition accuracy; The latter generates tactile friction for local devices, by exploring feedback mechanism to achieve improved accuracy and reduced complexity. The proposed FedCLL is numerically evaluated, using HapTex dataset. The results show that FedCLL outperforms existent methods in both acquisition accuracy and computational complexity.

Edge computing is considered as a promising paradigm to support vehicular applications in the upcoming sixth-generation (6G) vehicular networks. In the context of vehicular edge computing (VEC), the self-interested vehicular users and edge servers work towards incongruous goals. Such mixed-motive setting is detrimental to the collective good, sometimes leading to social dilemmas. To resolve such a conflict, we first formulate a bi-level optimization problem to model mixed-motive task offloading. In this case, vehicular users aim to improve energy efficiency under strict low-latency requirements, whereas edge servers attempt to increase serving efficiency. To address it, we propose a scheme based on bi-level reinforcement learning, i.e., bi-level multi-agent actor-critic (BLMAAC) framework. Specifically, upper-level edge servers make iterative optimization under the best responses of lower-level vehicular users, which can be regarded as a Stackelberg game. Theoretically, we identify the conditions and prove the convergence of the framework that is able to reach Stackelberg equilibrium strategy. By numerical evaluation, the high-utilization edge servers and energy-efficient vehicular users demonstrate the superiority of the bi-level structure. Moreover, the proposed scheme outperforms other actor-critic based learning algorithms and two-stage methods exploring Nash equilibrium strategy.

In the pursuit of achieving optimal functionality for internet of vehicles (IoV), the integration of multi-access edge computing (MEC) emerges as a solution, offering high bandwidth, low latency, robust security, and reliability services. In this article, we consider a multi-user multi-server vehicular network scenario, where the non-orthogonal multiple access (NOMA) technology in 5G is used to optimize spectrum resource utilization. We firstly formulate the problem using mixed integer non-linear programming (MINLP) and propose a task scheduling scheme based on deep reinforcement learning (DRL) to handle high-dimensional state and action spaces and to approximate the optimal solution. We then proposed solutions to the NOMA clustering and power allocation problems in order to further reducing system latency in the uplink transmission stage. Simulation results underscore the efficacy of our proposed algorithm in systems with unevenly distributed computing resources, showcasing superior performance compared to alternative algorithms.

Nonintrusive load monitoring (NILM) estimates the power of all loads in a smart home, using only the aggregated power on the bus. By leveraging NILM, smart grids implement demand-side management and anomaly diagnosis, optimizing energy allocation, and reducing casualties. To date, some NILM methods have been well developed. However, their estimation accuracy can be further improved when dealing with loads with more than two states. For this reason, a method is proposed using an actor–critic architecture with an attention mechanism to improve accuracy. Both actor and critic are neural networks: the former transforms the aggregated power into estimated load power, while the latter provides feedback on the estimated load power. To capture the dependencies of diverse states, an attention mechanism is equipped to the actor to enhance its representational ability. The proposed method is evaluated on the U.K.-DALE and REDD datasets and demonstrates improved accuracy compared to state-of-the-art.

Unbalanced material classification with visual–haptic data has become one of the primary concerns of tactile internet, existent classifiers suffer from performance degradation, since: the imbalance across both material classes and visual–haptic samples causes overestimation, some clients experience dropouts in global classifier training, and determining the theoretical convergence of a global classifier is challenging. This article develops the FedDDQN method to cope with the above issues. Specifically, for the server, the optimized federated learning scheme efficiently deals with unbalanced material classes and clients dropouts, and for clients, the adjusted double deep Q-networks (DDQN) effectively avoids the overestimation caused by unbalanced visual–haptic samples and improves representation ability. The convergence of FedDDQN are theoretically examined, using autonomous ordinary differential equation systems and Lyapunov stability. Numerical evaluations show that FedDDQN outperforms existing methods in both classification accuracy and computational complexity.

Spectrum sharing in cellular vehicle-to-everything (C-V2X) has been conceived as a promising solution to improve spectrum efficiency. However, the co-channel interference incurred with it may cause severe performance degradation to vehicular links. Thereby, radio resource management (RRM) is motivated and designed to ensure communication reliability and increase system capacity. One challenge is that RRM involves channel allocation and power control, which are tightly coupled and hard to optimize simultaneously. Another challenge for this is the difficulty adapting centralized RRM schemes, requiring global channel state information (CSI) and causing high signaling overhead. To tackle these challenges, we propose the hybrid centralized-distributed RRM scheme and the distributed RRM scheme. Specifically, we prove a decoupling method that provides a theoretical lower bound so that channel allocation and power control can be optimized independently. Given the decoupling method, the hybrid centralized-distributed RRM scheme is based on graph matching and reinforcement learning (GMRL) to maximize system capacity and guarantee reliability requirements. Further, to decrease computation complexity and signaling overhead, the distributed RRM scheme that only requires local CSI with hybrid-framework reinforcement learning (HFRL) is exploited. Finally, both schemes are numerically evaluated through experiments and outperform other deep Q-network (DQN)-based schemes.

Moving ground-target detection system is widely used to monitor illegal activities of pedestrians and vehicles. However, existing detection methods are restricted by the power consumption in hardware and are usually based on some single feature of the seismic signal, which leads to low detection accuracy and false alarms. To address these issues, we propose a new moving ground-target detection method for detecting the weak seismic signals generated by distant moving ground targets. This method combines an adaptive strategy and support vector machines (SVMs). Both time- and frequency-domain features of seismic signals are considered in the detection method. Additionally, we carry out field experiments to evaluate the performance of the proposed method. The results show that the proposed moving ground-target detection method can detect distant moving ground targets and avoid false alarms as many as possible, which indicates good performance.