The number of logistics distribution vehicle in the urban road network has a significant impact on the traffic flow operation. In order to quantitatively evaluate its impact and develop more refined control strategies, a Multiple Modes of Travel in Urban Road Network has been established based on a three-dimensional macroscopic fundamental diagram (MFD). It has been found that when the number of private cars in the experimental road network reaches 1100 and the number of logistics vehicles reaches 80, the driving speed of the road network begins to decrease and enter a congested state. It could be concluded that ① after the intensive optimization of distribution points, the number of private cars in the road network can be increased by up to 300, while logistics distribution vehicle can be increased by 50, resulting in an overall increase in traffic capacity of 9.33

Transportation systems are complex mega systems fraught with uncertainty and risk. Traditional traffic control and management methods, often based on deterministic or expected-value optimization, struggle to effectively address tail risks arising from traffic accidents, congestion propagation, and emergencies. As an emerging paradigm, risk-driven control introduces rigorous risk measurement tools to directly quantify and manage the distributional risks of transportation system performance, enabling more nuanced trade-offs between safety, efficiency, and robustness. This paper systematically reviews the theoretical advancements and practical applications of risk-driven control in transportation. First, it elucidates the necessity and core concepts of risk-driven control within the transportation context. Subsequently, it summarizes research outcomes across key subfields: autonomous vehicles, intelligent connected transportation, traffic network management and control, as well as public and multimodal transportation. Finally, it delves into current research challenges, such as computational complexity and verification difficulties, while outlining future research directions.

Low-altitude UAV scheduling and planning has become a critical technological pillar in disaster response systems; however, systemic challenges in complex environments and under uncertain risk conditions remain insufficiently understood. Although substantial progress has been achieved in model formulation and algorithm design in recent years, scheduling and planning frameworks still lack a systematic representation of key risk factors, such as meteorological disturbances, terrain damage, and communication constraints, thereby undermining operational safety and decision reliability. This study conducts a systematic review of low-altitude UAV scheduling and planning research over the past decade, covering representative disaster scenarios including forest fires, large building fires, earthquakes, floods, major public health emergencies, and traffic accidents. By comparatively analyzing scheduling objectives and technical pathways across the pre-disaster, during-disaster, and post-disaster stages, this paper summarizes the dominant research paradigms and limitations of multi-UAV coordination, air–ground coordination, and risk reduction-oriented scheduling and planning. This review reveals that existing approaches generally lack explicit modeling of dynamic risks and uncertainties, highlighting an urgent need to incorporate risk-aware considerations and reliability analysis frameworks into scheduling and planning to enhance the overall robustness and decision credibility of UAV systems in disaster environments.

This paper proposes a novel method for multi-robot collision avoidance during a hunting task, within a probabilistic uncertainty framework. First, to minimize the total hunting time, this paper transforms the multi-robot hunting task assignment issue to a multi-objective problem by considering some necessary factors that may affect hunting efficiency, including distance, the number of obstacles, and the adaptation between pursuers and evaders. Then, an improved K-means clustering algorithm is proposed to allocate the pursuers to evaders, and the auction algorithm is designed to solve the multi-objective problem. Additionally, by taking into account the positional uncertainty of robots and obstacles, the Buffered Uncertainty-Aware Voronoi Cells (BUAVC) of robots are constructed to guarantee the probabilistic conditional anti-collision measures between robots, as well as between robots and obstacles. In the Buffered Uncertainty-Aware Voronoi hunting framework, a greedy switch pursuer control strategy is designed to enhance hunting capability, which minimizes hunting time as much as possible while satisfying the probability anti-collision condition and considering the ‘deadlock’ problem. Finally, simulation experiments are conducted to illustrate the superiority of the proposed strategy with shorter global hunting time and total travel distance by comparing it with other existing methods.

3D object detection from LiDAR point clouds is a fundamental task in intelligent driving and urban scene perception. However, due to the inherent sparsity, uneven distribution, and severe long-range degradation of point clouds, accurately detecting distant vehicles remains challenging. To address these issues, we proposes CA-RCNN, a cascade attention-based 3D object detection network for LiDAR point clouds. Building upon a point-voxel fusion framework, CA-RCNN introduces three key components: a Cascade Attention-based Proposal Refinement (CAPR) module, a Semantic-Guided Farthest Point Sampling (SG-FPS) strategy, and a Multi-source Feature Fusion (MSFF) module. The CAPR module performs multi-stage proposal refinement with cascade attention, enabling progressive feature interaction across stages and iteratively improving the quality of bounding box regression and classification. The SG-FPS strategy assigns semantic-aware weights during keypoint sampling to enhance foreground representation while suppressing background interference, especially for distant and sparse objects. The MSFF module integrates point, voxel, and BEV features in a unified manner, enabling fine-grained interaction across multiple representations, multiple scales, and multiple points, thereby strengthening feature discriminability. Experiments on the KITTI datasets demonstrate that CA-RCNN consistently improves overall detection performance and achieves significant gains in distant vehicle detection tasks.

The increasing complexity and persistent network security challenges in traffic signal control are key issues requiring urgent attention to meet growing traffic demands. To address these issues, this paper proposes a resilient distributed model-free adaptive traffic signal control strategy (CDL-DMFAC) that integrates controller dynamic linearization (CDL) with multi-agent modeling. In the proposed framework, each signal phase at an intersection is modeled as an independent agent, and a compact form dynamic linearization (CFDL) is employed to construct an unknown ideal controller, enabling balanced control of multi-phase queue lengths. Furthermore, a denial-of-service (DoS) attack compensation mechanism is designed to mitigate the negative impact of communication interruptions or delays on signal timing decisions. Experimental results show that CDL-DMFAC effectively suppresses queue growth and delay accumulation under various attack intensities, with its performance advantage becoming more pronounced as attack severity increases. Notably, under the most challenging scenario-high traffic demand with multiple intersections simultaneously subjected to DoS attacks-the proposed method achieves reductions of 28.3% in average queue length and 36.32% in average waiting time compared to conventional signal control methods. These results highlight the method's strong resilience against attacks, operational stability, and potential for deployment in larger-scale urban traffic networks.

To address the limitation of single-view detection in complex underground environments, a novel dual-branch network architecture for air-ground cooperative detection is proposed in this paper. The proposed method consists of two independent encoder-decoder branches, which are used to extract aerial and ground features separately. A cross-attention mechanism is introduced in the encoders to enable feature interaction between different viewpoints. Furthermore, a dynamic weight fusion module is designed to adaptively fuse aerial and ground features, and a joint loss is incorporated for backpropagation to optimize both the branches and the fusion module. Experimental results show that the dual-branch DETR achieves a 19.8

To solve the problem that a single simulation platform is difficult to balance the authenticity of the scene and the accuracy of traffic flow modeling in the air-ground cooperative traffic control simulation, this paper proposes a joint simulation scheme based on Carla and SUMO: build a custom traffic environment by integrating RoadRunner, rely on Carla to achieve high-fidelity scene rendering and sensor simulation, and use SUMO to complete refined traffic flow modeling, thus forming a virtual experimental scene with both realism and accuracy. At the same time, the YOLOv12 target detection algorithm and Bytetrack tracker are combined to realize the accurate calculation of vehicle speed in BEV traffic video based on multi-frame position information. The experimental results show that the system performs well in traffic flow restoration and target tracking accuracy, achieving an MAE of 0.57 and an MSE of 0.51. It provides a low-cost and reproducible software-in-the-loop simulation environment for BEV traffic image detection algorithm test and traffic control strategy verification.

Multi-UAV swarms offer significant potential in military and civilian applications due to their high efficiency, robustness, and flexibility. Cooperative envelopment, a key application for target interception, area blockade, and surveillance, faces technical challenges from non-cooperative targets, dynamic environments, and limited communication capabilities. This paper systematically reviews Multi-UAV cooperative envelopment, focusing on three core modules. First, Task Allocation methods are analyzed, including heuristic intelligent algorithms (e.g., Genetic Algorithms (GA), Particle Swarm Optimization (PSO)), distributed market mechanisms (e.g., auction algorithms, contract net protocol), and clustering algorithms. Second, Motion Planning and Cooperative Control techniques are examined, covering geometry/topology-based methods (e.g., Voronoi diagrams, Apollonian circles), Reinforcement Learning (RL), and consensus theory. Finally, the paper summarizes the field's trend from classical control toward AI integration. It also highlights future challenges in scalability and robustness, noting the broad application prospects for this technology.

Autonomous new energy heavy-duty trucks are a crucial component of the future logistics and transportation industry. However, their dynamic control becomes more challenging due to characteristics such as complex operating environments, response delays, and large inertia—yet autonomous driving systems still demand precise dynamic control performance. Addressing the issue of steering resistance with unknown and time-varying dynamics coefficients caused by complex road conditions and terrains, this study decomposes the motion control problem of new energy heavy-duty trucks by utilizing two key motion variables: longitudinal velocity and yaw rate. This paper presents an adaptive feedback control architecture for vehicle speed and steering angular velocity, based on the Model Reference Adaptive Control (MRAC) method. The strategy refers to the vehicle dynamics model, and uses the proposed adaptive control law to adjust the motor output torque in real time, so that the actual vehicle speed and angular speed can accurately track the corresponding output of the reference model. The simulation results show that the tracking performance is improved by 10

Objective With the rapid development of vehicular networks (vehicle-to-everything) and autonomous driving technologies, cooperative perception has become a crucial technology to enhance the environmental perception capability of connected and autonomous vehicles (CAVs). Individual perception information is shared among CAVs, and cooperative perception can effectively expand the sensing range, reduce occlusion effects, and improve perception redundancy. However, vehicle localization errors are unavoidable in real driving scenarios due to sensor noise, environmental interference, and communication uncertainty. Localization errors often lead to spatial misalignment among point clouds from multiple vehicles, thereby reducing the performance of multivehicle cooperative perception. Mitigating the impact of localization errors on cooperative perception and improving computational efficiency for on-board deployment remain challenging problems. Methods To address the above issues, this paper proposed a cooperative perception method of CAVs based on point feature fine alignment. First, a lightweight point feature extraction module was designed using PointConvFormer to process point cloud data collected by individual vehicles. By integrating PointConvFormer layers into bottleneck residual blocks, the proposed feature extraction module preserves the three-dimensional spatial structure of the point cloud while capturing local geometric features and global contextual information. Second, the cross-vehicle point feature hierarchical fine alignment module was designed to address spatial misalignment in cross-vehicle data fusion. This module used the global poses of multiple CAVs, collected from positioning systems, to achieve coarse alignment of point features between the surrounding CAVs and the ego-vehicle. The fine-grained alignment strategy was further implemented using local overlapping point-cloud registration to improve the spatial feature consistency of the aggregated point cloud, and the point feature similarity within overlapping regions was exploited to maximize cross-vehicle feature correspondence and alleviate feature alignment deviation caused by localization errors. Furthermore, the multiscale feature fusion module was built to integrate local fine-grained features with global contextual information; it employed multiscale mask sampling to retain the structural information of the aligned aggregated point cloud at various spatial resolutions. Results Extensive experiments and ablation studies were conducted on V2V4real and V2XSet datasets to comprehensively evaluate the performance of the proposed method. The experimental results demonstrated that the proposed approach achieved superior perception accuracy and robustness compared to other state-of-the-art methods across traffic scenarios with varying levels of localization errors. Moreover, the proposed method maintained high computational efficiency and satisfied the real-time requirements of on-board deployment. Conclusions The proposed cooperative perception method, based on point feature fine alignment, integrates a lightweight point feature extraction module, a cross-vehicle point feature fine alignment module, and a multiscale feature fusion module. It effectively addressed the perception performance degradation problem caused by vehicle localization errors and improved the accuracy and robustness of cooperative perception among CAVs. In future work, we will enhance the collaborative perception performance of CAVs in complex scenarios, such as rain and fog, by integrating information from multimodal sensors, including cameras and millimeter-wave radars.

The Macroscopic Fundamental Diagram (MFD) is a promising paradigm for network-level traffic management, providing a robust framework for monitoring congestion and maximizing network production. This paper surveys recent MFD advances across three key domains: theoretical modeling, state estimation, and coordinated control. It highlights key methods, from classical dynamic models and EKF/MHE estimators to data-driven (ML/DL) modeling, and control frameworks from PID to Model Predictive Control (MPC) and multi-region coordination. This review analyzes major challenges, including network heterogeneity, data sparsity, boundary queuing, and multi-modal interactions. Looking ahead, the paper discusses emerging trends like hybrid model-driven/data-driven frameworks, multi-modal 3D-MFDs, and control strategies for mixed-autonomy (CAV) environments. Ultimately, this paper's goal is to provide theoretical insights and practical guidance for developing robust MFD-based control systems for complex urban networks.

IntroductionUnderstanding the temporal dynamics and regional variation of study-abroad search attention is important for interpreting educational mobility intentions in a highly digitized information environment.MethodsThis study applies visibility graph theory to Douyin/Juliang Suanshu study-abroad search indices across 31 mainland Chinese provincial-level units from 4 June 2022 to 31 May 2025. Provincial series are aggregated into seven major regions, with summation used as the primary aggregation method and PCA used for aggregation sensitivity analysis. Regional natural visibility graphs are benchmarked against 100 size- and density-matched random graphs, alternative degree distributions are fitted, and regional network complexity is evaluated using the entropy weight method (EWM) with bootstrap uncertainty.ResultsThe results show that all regional visibility graphs have substantially higher clustering than random benchmarks and small-world coefficients above 42, while the degree distributions are better interpreted as heavy-tailed than as uniquely confirmed power laws. Regional time series are strongly synchronized, with a mean off-diagonal zero-lag correlation of 0.987 and no systematic lead-lag pattern within a 30-day window. EWM ranks Central China highest in the daily analysis, followed by East China and South China, but bootstrap intervals overlap and the ordering is sensitive to weekly aggregation. Weekly visibility-graph community detection identifies 5–7 temporal communities per region and recurring transition dates around February 2023, September 2023, March 2024, and late 2024.DiscussionThese findings clarify the temporal organization of study-abroad search attention and provide a network-based framework for analyzing regional educational search behavior.

Air-ground heterogeneous cooperative sensing systems have emerged as a promising paradigm in multi-agent perception and intelligent decision-making, and they are increasingly deployed in smart cities, disaster relief, and environmental monitoring. Specifically, these systems leverage the complementary strengths of unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) to achieve robust perception, precise localization, and coordinated task execution in complex, dynamic environments. Against this backdrop, this paper presents a comprehensive survey of recent advances in air-ground cooperative systems, which is structured around four key areas: multimodal perception fusion, cooperative localization and mapping, path planning, and control strategies. It also highlights representative methods such as distributed state estimation, cross-view semantic feature alignment, and integrated perception–decision–control frameworks. Moreover, this review analyzes typical application cases and identifies major challenges, including heterogeneous data fusion, limited communication bandwidth, and the absence of standardized datasets and evaluation metrics. Looking ahead, the paper discusses emerging trends such as swarm-intelligent coordination mechanisms, unified cross-modal representation models, and lightweight end-to-end architectures. Ultimately, this paper’s goal is to provide theoretical insights and practical guidance for building high-robustness, high-autonomy air-ground cooperative systems suitable for deployment in real-world multi-agent scenarios.

In this work, a novel lightweight FusionMamba-YOLO algorithm is introduced to address the challenges of real-time scene understanding in the navigation of quadruped robots. Specifically, the proposed algorithm is based on a novel Mamba-YOLO hybrid architecture. First, the FusionMamba backbone network employs a dual-stream interactive fusion mechanism, integrating global four-directional Mamba scanning with Mamba operations based on local windows. This design enables the effective combination of local feature information while efficiently processing global information. Second, a novel small object enhancement pyramid (SOEP) module is designed to dynamically fuse low-level P2 features with high-level semantic features. Consequently, the SOEP module is a lightweight structure that significantly enhances small object detection capabilities without increasing computational overhead. Third, a new pedestrian traffic light instance (PTL-Instance) dataset is designed. This dataset extends the original pedestrian traffic light (PTL) dataset by incorporating data from complex traffic scenes in Beijing with instance-level annotations for traffic signals and pedestrian crossings. The experimental results demonstrate that the proposed algorithm achieves superior performance, with a mean average precision (mAP) of 88.5

Traffic holographic perception refers to the real-time, high-fidelity, and multi-dimensional sensing of traffic states through the fusion of heterogeneous sensors, including cameras, radars, and connected vehicle data. The multi-source perception data obtained thereby can provide a complete digital representation of the road network for the Intelligent Transportation System (ITS). However, sensors are vulnerable to environmental interference, which can result in data loss at specific points or along arterial highways for certain periods, potentially undermining system safety and decision-making reliability. To address these challenges, a deep learning method based on Graph Convolutional Networks (GCN) and Gated Recurrent Units (GRU) is proposed, leveraging Artificial Intelligence (AI) and intelligent connected technologies for real-time acquisition of multi-sensor perception data. A feature-level fusion integrates multi-source perception data. GCN captures spatial dependencies from the road network topology, while GRU extracts temporal features from time series, enabling accurate imputation of missing traffic data. The method is evaluated at intelligent connected intersections in the Beijing High-level Autonomous Driving Demonstration Area. Results show that the accuracy of long-term traffic state completion reaches 89.36%, and the Root Mean Square Error (RMSE) is reduced by 17.2% compared to the Long Short-Term Memory (LSTM) baseline. This framework provides a practical solution for deploying traffic holographic perception technology in secure and trustworthy ITS.

Deep learning methods face several limitations in the real-time data-driven control of complex nonlinear systems. These limitations include low computational efficiency, weak theoretical reliability, and a lack of interpretability. The root causes are the network’s reliance on static nonlinear activation functions and the inadequacy of conventional single-step parameter update mechanisms. To address these limitations, this paper proposes a novel Model-Free Adaptive Neural Network (MFA-NN) control algorithm. First, the proposed method eliminates nonlinear activation functions. Instead, it constructs an underlying architecture composed of cascaded multi-layer purely linear weight matrices. This structural design effectively mitigates both the vanishing gradient and dead neuron problems. Second, this paper introduces an online dual-axis optimization mechanism. This mechanism performs multiple error-driven iterations within a single sampling period. Consequently, it effectively improves the utilization of single-step data and enhances the controller’s robust adaptability to abrupt system changes. Theoretically, the proposed framework directly maps the cascaded network matrices into a high-dimensional Pseudo Partial Derivative (PPD) with clear physical significance. This mapping effectively overcomes the black-box limitations of traditional neural networks. Furthermore, based on the differential mean value theorem, this paper proves the existence and boundedness of the equivalent mapping parameters. This proof provides a theoretical basis for the stability of the closed-loop system. Comparative simulation results on discrete-time nonlinear systems are presented. The results demonstrate that the proposed method maintains high-precision steady-state tracking and strong robustness, even under severe abrupt structural changes.

Abstract Embodied intelligence (EI) emphasizes the ability of an agent to achieve adaptive behavior through physical perception, environmental interaction, and real-time decision-making. Its core lies in situating cognition, perception, and action within the dynamic interactions of the physical world. Unlike Intelligent Transportation System (ITS), which primarily emphasizes the intelligent upgrading of vehicles and infrastructure, the new generation of Autonomous Transportation System (ATS), focuses on the organic integration of “humans (decision-makers) - machines (carriers) - infrastructure (foundation) - environment (context)”, requiring the system to respond as flexibly as a human driver in complex scenarios. The core requirements of EI and ATS are highly aligned: embodied intelligence provides a theoretical framework for ATS, while ATS offers a large-scale application scenario for the implementation of embodied intelligence. Through the lens of embodied intelligence, challenges in ATS related to perception, decision-making, control, and social interaction can be addressed more systematically. In this regard, this paper first elaborates on the embodied intelligence aspects of ATS from a multi-faceted systemic perspective, including perceptual embodiment, motor embodiment, cognitive embodiment, evolutionary embodiment, and social embodiment, establishing a framework for ATS from the perspective of embodied intelligence. Subsequently, from the standpoint of embodied intelligence, it summarizes the key technologies required for realizing ATS. Finally, it outlines future development directions for ATS under the theoretical framework of embodied intelligence. By analyzing and understanding the principles of ATS through the novel perspective of embodied intelligence, this paper provides important guidance for promoting the development and realization of ATS.

This paper investigates the optimal tracking control problem for a class of nonlinear multi-agent systems (MASs) with affine-in-control dynamics. Firstly, to handle the full-state constraints, time-varying barrier functions are introduced, transforming the constrained system into an equivalent unconstrained form. Secondly, optimal consensus control of the system is achieved by establishing an enhanced performance index function. Novel local value functions are designed for each agent within the parallel control framework. These functions comprehensively consider the consensus error, the dynamics of the local agent, and the parallel control strategies of its neighbors, thereby transforming the tracking control problem into an optimal consensus problem. To solve the resulting coupled Hamilton-Jacobi (HJ) equations, a critic-only neural network (NN)architecture is constructed. The uniform ultimate boundedness of the tracking errors for all agents under the designed control law is rigorously proven using Lyapunov’s direct method. Finally, numerical simulations are provided to verify the effectiveness and superiority of the proposed method.