Reconfigurable intelligent surface (RIS) represents a promising technology due to its ability to create a controllable wireless propagation environment. However, this advantage can only be realized when the passive RIS is of a sufficiently large size, for which the conventional uniform plane wave (UPW)-based far-field model may become invalid. In this paper, we consider the directional reflecting element and establish an extremely large-scale RIS (XL-RIS) signal transmission model that incorporates distance/angle correlation based on the non-uniform spherical wave (NUSW). Based on the proposed model, we first analyze the optimal rotation angle of the RIS in both far-field and near-field scenarios, concluding that optimal system performance can be achieved when the angle of arrival (AoA) and angle of departure (AoD) are equal through the rotation of the RIS. Then, we introduce the concept of the effective activated area and analyze the number of effective activated elements of the XL-RIS under narrow beam incidence, and the relationship between transmission distance and the optimal number of reflecting elements is obtained.We also analyze the asymptotic performance and convergence characteristics of the system. Theoretical results indicate that the relationship between the quadratic of the XL-RIS element number and the received signal-to-noise ratio (SNR) is nonlinear in the near-field. Numerical results validate our analysis and demonstrate the necessity of an appropriate XL-RIS transmission model within the effective activated area.

Segment Routing over IPv6 (SRv6) gives operators explicit path control and alleviates network congestion, making it a compelling technique for traffic engineering (TE). Yet two practical hurdles slow adoption. First, a one-shot upgrade of every traditional device is prohibitively expensive, so operators must prioritize which devices to upgrade. Second, the Segment Routing Header (SRH) increases packet size; if TE algorithms ignore this overhead, they will underestimate link load and may cause congestion in practice. We address both challenges with DRL-TE, an algorithm that couples deep reinforcement learning (DRL) with a lightweight local search (LS) step to minimize the network’s maximum link utilization (MLU). DRL-TE first identifies the smallest set of critical devices whose upgrade yields the largest drop in MLU, enabling hybrid IP/SRv6 networks to approach optimal performance with minimal investment. It then computes SRH-aware routes, and the DRL agent, augmented by a fast LS refinement, rapidly reduces MLU even under traffic variation. Experiments on an 11-node hardware testbed and three larger simulated topologies show that upgrading about 30% of devices allows DRL-TE to match fully upgraded networks and reduce MLU by up to 34% compared with existing algorithms. DRL-TE also maintains high performance under link failures and traffic variations, offering a cost-effective and robust path toward incremental SRv6 deployment.

We introduce LOcc, an effective and generalizable framework for open-vocabulary occupancy (OVO) prediction. Previous approaches typically supervise the networks through coarse voxel-to-text correspondences via image features as intermediates or noisy and sparse correspondences from voxel-based model-view projections. To alleviate the inaccurate supervision, we propose a semantic transitive labeling pipeline to generate dense and fine-grained 3D language occupancy ground truth. Our pipeline presents a feasible way to dig into the valuable semantic information of images, transferring text labels from images to LiDAR point clouds and ultimately to voxels, to establish precise voxel-to-text correspondences. By replacing the original prediction head of supervised occupancy models with a geometry head for binary occupancy states and a language head for language features, LOcc effectively uses the generated language ground truth to guide the learning of 3D language volume. Through extensive experiments, we demonstrate that our transitive semantic labeling pipeline can produce more accurate pseudo-labeled ground truth, diminishing labor-intensive human annotations. Additionally, we validate LOcc across various architectures, where all models consistently outperform state-of-the-art zero-shot occupancy prediction approaches on the Occ3D-nuScenes dataset.

Sensor simulation is pivotal for scalable validation of autonomous driving systems, yet existing Neural Radiance Fields (NeRF) based methods face applicability and efficiency challenges in industrial workflows. This paper introduces a Gaussian Splatting (GS) based system to address these challenges: We first break down sensor simulator components and analyze the possible advantages of GS over NeRF. Then in practice, we refactor three crucial components through GS, to leverage its explicit scene representation and real-time rendering: (1) choosing the 2D neural Gaussian representation for physics-compliant scene and sensor modeling, (2) proposing a scene editing pipeline to leverage Gaussian primitives library for data augmentation, and (3) coupling a controllable diffusion model for scene expansion and harmonization. We implement this framework on a proprietary autonomous driving dataset supporting cameras and LiDAR sensors. We demonstrate through ablation studies that our approach reduces frame-wise simulation latency, achieves better geometric and photometric consistency, and enables interpretable explicit scene editing and expansion. Furthermore, we showcase how integrating such a GS-based sensor simulator with traffic and dynamic simulators enables full-stack testing of end-to-end autonomy algorithms. Our work provides both algorithmic insights and practical validation, establishing GS as a cornerstone for industrial-grade sensor simulation.

The increasing demand for delay-sensitive services in industrial manufacturing, the Internet of Vehicles, and smart logistics imposes stringent delay requirements on intelligent machine (IM) networks. To reduce latency, short packet transmissions are widely used. However, their impact on network delay performance remains underexplored, particularly in large-scale deployments prone to packet collisions and queuing congestion. This paper develops a theoretical framework for modeling downlink communication and derives analytical expressions for three key delay metrics: transmission success probability, expected delay, and delay jitter. By incorporating finite blocklength constraints, we accurately characterize the effects of IM density and packet length on delay performance. Simulation results validate our model, offering valuable insights for optimizing IM network design and improving real-time communication efficiency.

This paper investigates the active and passive beamforming design of reconfigurable intelligent surface (RIS) aided multi-user multiple-input single-output (MU-MISO) systems to maximize the system sum rate. We proposed an evolution-guided policy gradient (EGPG) algorithm, which consists of the cross-entropy method (CEM) and the deep deterministic policy gradient (DDPG) algorithm to obtain the optimal beamforming design. Specifically, the experience replay diversity and policy network update of DDPG can be guided by CEM. Furthermore, the optimization objective of CEM is transformed from a high-dimensional policy network space into a low-dimensional action space for large-scale RIS beamforming design. The simulation results demonstrate that the proposed algorithm can bring about a significant performance improvement.

In perceptive mobile networks (PMNs), using 5G New Radio (NR) signals for direct sensing poses a significant challenge to practical implementation due to the high computational complexity involved in estimating sensing parameters. In this paper, an efficient sensing method is proposed to incorporate both downlink active sensing and passive sensing to estimate multiple sensing parameters, including delays, angle of arrival (AoA), angle of departure (AoD) and Doppler. In particular, it exploits the synchronization signal blocks (SSBs) to facilitate sensing with multiple remote radio units (RRUs). To reduce the computational complexity of direct sensing, a sparse model is developed to decouple multiple sensing parameter estimation, enabling efficient sensing method design. Then, leveraging unitary approximate message passing (UAMP) and sparse Bayesian learning (SBL), we propose an efficient method to achieve parameter estimation and association with corresponding RRUs. This method is further extended to general scenarios involving multiple path components with the same delay. Extensive simulations demonstrate the effectiveness of the proposed method, showing that it outperforms existing ones in terms of sensing accuracy and complexity.

In the dynamic moving process of vehicle clusters, there are several challenges, including inefficiencies, cross-domain trust issues and privacy leakage. To address these issues, we propose a blockchain-assisted group and cross-domain authentication key agreement, which implements distributed key management based on a threshold key sharing scheme, and realizes group authentication and group key distribution for vehicle clusters through the collaboration of roadside units (RSUs) and Key Generation Center (KGC). Meanwhile, a cross-domain trust chain is constructed based on blockchain to accomplish secure cross-domain authentication and key agreement without the participation of the original KGC, which solves the problem of trust deficiency and single-point vulnerability in the process of cross-domain communication. Finally, we employed Real-or-Random (ROR) formal security analysis and the ProVerif tool to verify that the proposed authentication scheme, the results show that the proposed scheme is secure and superior to existing schemes in terms of communication and computational overhead.

Federated Learning (FL) is a communication-efficient distributed machine learning method that allows multiple devices to collaboratively train models without sharing raw data. FL can be categorized into centralized and decentralized paradigms. The centralized paradigm relies on a central server to aggregate local models, potentially resulting in single points of failure, communication bottlenecks, and exposure of model parameters. In contrast, the decentralized paradigm, which does not require a central server, provides improved robustness and privacy. The essence of federated learning lies in leveraging multiple local updates for efficient communication. However, this approach may result in slower convergence or even convergence to suboptimal models in the presence of heterogeneous and imbalanced data. To address this challenge, we study decentralized federated averaging via random walk (DFedRW), which replaces multiple local update steps on a single device with random walk updates. Traditional Federated Averaging (FedAvg) and its decentralized versions commonly ignore stragglers, which reduces the amount of training data and introduces sampling bias. Therefore, we allow DFedRW to aggregate partial random walk updates, ensuring that each computation contributes to the model update. To further improve communication efficiency, we also propose a quantized version of DFedRW. We demonstrate that (quantized) DFedRW achieves convergence upper bound of order$\mathcal {O}(\frac{1}{k^{1-q}})$under convex conditions. Furthermore, we propose a sufficient condition that reveals when quantization balances communication and convergence. Numerical analysis indicates that our proposed algorithms outperform (decentralized) FedAvg in both convergence rate and accuracy, achieving a 38.3% and 37.5% increase in test accuracy under high levels of heterogeneities, without increasing communication costs for the busiest device.

Vision-centric 3D environment understanding is both vi-tal and challenging for autonomous driving systems. Re-cently, object-free methods have attracted considerable at-tention. Such methods perceive the world by predicting the semantics of discrete voxel grids but fail to construct continuous and accurate obstacle surfaces. To this end, in this paper, we propose SurroundSDF to implicitly predict the signed distance field (SDF) and semantic field for the continuous perception from surround images. Specifically, we introduce a query-based approach and utilize SDF con-strained by the Eikonal formulation to accurately describe the surfaces of obstacles. Furthermore, considering the absence of precise SDF ground truth, we propose a novel weakly supervised paradigm for SDF, referred to as the Sandwich Eikonal formulation, which emphasizes applying correct and dense constraints on both sides of the surface, thereby enhancing the perceptual accuracy of the surface. Experiments suggest that our method achieves SOTA for both occupancy prediction and 3D scene reconstruction tasks on the nuScenes dataset.

Low Earth Orbit (LEO) satellites are gradually developing towards to a large scale, in order to achieve a desired vision of ubiquitous connectivity and broadband access at anytime and anywhere. However, the increasing scale and highly dynamic nature of LEO constellation pose challenges to network management on its flexibility and scalability. Clustering is introduced as an effective approach to manage LEO satellite networks in a flexible manner. Unfortunately, satellite clusters encounter instability and high communication load due to frequent topology changes and traffic growth. In this paper, we design LEO satellites clustering models that jointly optimize cluster reliability and network communication load in the GEO/LEO network architecture. The coalition game framework is introduced to obtain a stable cluster structure by adopting an automated and centralized approach. A coalition formation algorithm based on the optimization of reliability and communication load is developed for the clustering problem. Finally, numerical simulations are carried out to evaluate the superiority and effectiveness of the proposed grouping and clustering scheme.

Benefiting from high mobility, unmanned aerial vehicles (UAVs) can reconstruct wireless connections for affected areas. Most of the existing work has usually ignored the influence of limited airborne energy on the dependability of UAV data transmission. Accordingly, this article proposes an UAV-assisted wireless powered system to achieve dependable data collections in Internet of Things (IoT). Specifically, an UAV leverages energy beamforming to transfer energy to ground users (GUs) in downlink subtimeslot, while the GUs transmit data to the UAV with the harvested energy in uplink subtimeslot. For this system, a joint optimization problem of subtimeslot allocation and UAV route planning is investigated to maximize the system energy efficiency subject to UAV dynamics, time slot duration, and GUs' rate threshold. To tackle the nonconvexity of the formulated problem, a low-complexity alternating iterative algorithm is proposed. The first subproblem optimizes subtimeslot allocation by using the bisection method and Lagrange multiplier method for the fixed UAV route, while the second optimizes the UAV route for the periodic and single flight modes with the given subtimeslot allocation. Then, the two subproblems are alternatively solved until convergence. The simulation results demonstrate that the proposed algorithm can not only optimize the UAV route, but also achieve a good compromise between system throughput and UAV propulsion energy consumption.

Modern deep-learning-based lane detection methods are successful in most scenarios but struggling for lane lines with complex topologies. In this work, we propose CondLaneNet, a novel top-to-down lane detection framework that detects the lane instances first and then dynamically predicts the line shape for each instance. Aiming to resolve lane instance-level discrimination problem, we introduce a conditional lane detection strategy based on conditional convolution and row-wise formulation. Further, we design the Recurrent Instance Module(RIM) to overcome the problem of detecting lane lines with complex topologies such as dense lines and fork lines. Benefit from the end-to-end pipeline which requires little post-process, our method has real-time efficiency. We extensively evaluate our method on three benchmarks of lane detection. Results show that our method achieves state-of-the-art performance on all three benchmark datasets. Moreover, our method has the coexistence of accuracy and efficiency, e.g. a 78.14 F1 score and 220 FPS on CULane. Our code is available at https://github.com/aliyun/conditional-lane-detection.

Amplitude-integrated EEG (aEEG) is becoming more and more useful in the monitoring of clinically ill neonates. If there is a method that can predict neonatal aEEG signals, doctors can forecast the possible abnormality of neonates' brain functions in advance and give early intervention. However, no such research on the prediction of aEEG signals has been found in the literature. In this paper, we combine aEEG signals with Long-Short Time Memory (LSTM) model and propose a method to predict aEEG signals based on LSTM. All of the aEEG signals after preprocessing were used as the input of the LSTM, a type of recurrent neural networks which can process long term signals with high accuracy. To assess the method, several experiments were conducted on 276 neonatal aEEG tracings including 217 normal cases and 59 abnormal ones. Experimental results show that the predicted aEEG signals are very close to the real aEEG signals. Our LSTM-based method might therefore help predict neonatal brain disorders in NICUs.