论文导读 | IEEE T-ITS 2024年第25期核心论文导读（二）

点击蓝字 关注我们

核心论文

21. Epurate-Net: Efficient Progressive Uncertainty Refinement Analysis for Traffic Environment Urban Road Detection

Epurate-Net：用于交通环境城市道路检测的高效渐进不确定性精炼分析

作者：T. Han, S. Chen, C. Li, Z. Wang, J. Su, M. Huang, and G. Cai

摘要：An efficient and high-precision urban road detection algorithm is proposed to identify ambiguous road contours in a traffic environment. The algorithm employs a multi-model strategy and progressive uncertainty analysis through deep learning methods. The experimental results demonstrate that the proposed method refines road contours, achieving exquisite road margins and ensuring the safety of autonomous driving.

关键词：Intelligent transportation, urban road detection, autonomous driving, uncertainty analysis, deep learning.

22. TCGNet: Type-Correlation Guidance for Salient Object Detection

TCGNet：用于显著目标检测的类型相关引导

作者：Y. Liu, L. Zhou, G. Wu, S. Xu, and J. Han

摘要：This article probes into the issue of saliency prediction involving the complementary properties of convolutional neural networks (CNNs) and capsule networks (CapsNets) semantic cues, and proposes a type-correlation guidance network (TCGNet) for salient object detection. Specifically, a multi- type cue correlation (MTCC) covering CNNs and CapsNets is designed to extract the contrast and part-whole relational semantics, respectively. Using MTCC, two correlation matrices containing complementary information are computed with these two types of semantics. In return, these correlation matrices are used to guide the learning of the above semantics to generate better saliency cues. Besides, a type interaction attention (TIA) is developed to interact semantics from CNNs and CapsNets for the aim of saliency prediction.

关键词：Salient object detection, part-object relationship, capsule network.

23.Toward Data-Driven Simulation of Network-Wide Traf- fic: A Multi-Agent Imitation Learning Approach Using Urban Vehicle Trajectory Data

面向数据驱动的全网络交通仿真：使用城市车辆轨迹数据的多智能体模仿学习方法

作者：J. Sun and J. Kim

摘要：This article aims to model network-wide vehicle movements and traffic state evolution by considering the interac- tions among vehicles and link traffic conditions. The authors thus develop a data-driven simulation model that applies the existing multi-agent generative adversarial imitation learning (MAGAIL) framework considering both vehicle and link agents to learn vehicles’ link-to-link transitions and within- link movements (travel speed), and road links’ traffic state evolution patterns simultaneously.

关键词：Traffic simulation, multi-agent imitation learning, vehicle trajectory, link traffic state, MAGAIL.

24. Adaptive Segmentation Enhanced Asynchronous Federated Learning for Sustainable Intelligent Transportation Systems

用于可持续智能交通系统的自适应分割增强异步联邦学习

作者：X. Zhou, W. Liang, A. Kawai, K. Fueda, J. She, and K. I.-K. Wang

摘要：The proliferation of advanced embedded and communication technologies has facilitated the possibility of modeDecentralized federated learning, adaptive segmentation, sustainable computing, intelligent transportation systems. rn intelligent transportation systems (ITSs). The hierarchical nature of such large-scale and distributed systems brings obvious challenges in creating a scalable and sustainable computing environment, and hence the development and application of edge intelligence become critical. Federated learning (FL), as an emerging distributed machine learning paradigm, aims to offer secure knowledge sharing and effective learning across multiple devices. However, conventional FL may fall into trouble when facing large-scale and network-agnostic systems with fast-moving devices and changing network attributes. In this study, the authors propose an adaptive segmentation-enhanced asynchronous FL (AS-AFL) model, aiming to improve learning efficiency and reliability in sustainable ITS via a decentralized fashion. Specifically, a meta-learning-based adaptive segmentation scheme is designed to automatically separate the client nodes (e.g., vehicles) into multiple edge groups according to their homogeneous attributes. An integrated aggregation mechanism is then developed to realize the horizontal FL among a group of similar client nodes via the so-called intra-group synchronous aggregation while allowing the vertical FL across different groups via the so-called inter-group asynchronous aggregation. The experiment and evaluation results based on an open-source dataset demonstrate the outstanding learning and communication performance of the proposed model, compared with several conventional FL schemes in a distributed ITS application scenario.

关键词：Autonomous vehicles, collision avoidance, pla tooning, visible light positioning.

25. Leveraging Dynamic Right-of-Way Allocation and Tolling Policy for CAV Dedicated Lane Management to Promote CAV and Improve Mobility

利用动态优先通行权分配和收费政策管理自动驾驶车辆专用车道以促进自动驾驶车辆并改善出行

作者：H. Chen, F. Wu, K. Hou, and T. Z. Qiu

摘要：The connected and automated vehicle dedicated lanes (CAV-DLs) were proposed to maximize the benefit of CAVs. However, the CAV-DLs may be underutilized and reduce traffic efficiency, especially in low CAV penetration rate (PR) cases. To solve this problem, two novel strategies were pro- posed: The first one is to dynamically allocate the right-of- way for CAV-DLs so that the human-driving vehicles (HDVs) can be allowed to use the dedicated lanes when they are not adequately occupied. The second strategy allows HDVs to use the CAV-DLs by paying a toll. The toll was determined by the travel time difference between CAV-DL and general lane (GL), and these tolls can be utilized as subsidies to stimulate drivers to purchase CAVs to promote their adoption. The two strategies were evaluated based on the network of Edmonton’s downtown area in Canada, and the results demonstrated that both methods can significantly improve traffic mobility.

关键词：CAV dedicated lane, right-of-way allocation, toll, reduced travel time, promote CAV.

26. A Multi-Task Learning Network With a Collision- Aware Graph Transformer for Traffic-Agents Trajectory Prediction

一种用于交通代理轨迹预测的多任务学习网络与碰撞感知图转换器

作者：B. Yang, F. Fan, R. Ni, H. Wang, A. Jafaripournimchahi, and H. Hu

摘要：Accurately forecasting the future trajectories of surrounding agents is a crucial task for autonomous vehicles to avoid collisions. However, capturing the complex interactions between agents in complex urban scenes poses a significant challenge. To address this issue, the authors propose an enhanced graph transformer (TP-EGT) to forecast the future trajectories of traffic-agents. Specifically, a collision-aware graph trans- former is introduced to capture the intricate social interactions between traffic-agents. Furthermore, an additional interaction prediction task is proposed that predicts the interaction prob- abilities between agents. This task aims to mitigate the over- smoothing issue of the graph transformer by employing a multi-task learning strategy. The trajectory prediction performance is subsequently improved with additional interaction probabilities, which are useful for the decision-making and planning of autonomous vehicles. Overall, TP-EGT offers a promising approach to accurately predict the future trajectories of agents, thereby enhancing the safety and efficiency of autonomous vehicles in complex urban environments.

关键词：Trajectory prediction, autonomous vehicles, social interactions, multi-task learning, graph transformer.

27. Deep Reinforcement Learning Enabled Multi-UAV Scheduling for Disaster Data Collection With Time- Varying Value

基于深度强化学习的灾害数据收集多无人机调度

作者：P. Wan, G. Xu, J. Chen, and Y. ZhouIn order to cope with information infrastructure congestion and disruption during disasters and provide timely information for rapid response activities, UAVs are proposed as temporary and mobile relays for disaster data collection and their scheduling method is studied based on deep reinforcement learning (DRL). Since the disaster data value always varies with data collection time and durations, the problem is modeled as a team orientation problem with time-varying values. Through analyzing the relationships between UAV route selection and service time at each node, as well as the dynamics of disaster data value, an attention-based DRL method is developed. Extensive experiments have been conducted and results show that the proposed method is superior to many existing methods in UAV scheduling, especially in large and complex scenarios.

关键词：Disaster response, unmanned aerial vehicle, multi-UAV scheduling, data collection, deep reinforcement learning, time-varying value.

28. Improving Freeway Merging Efficiency via Flow-Level Coordination of Connected and Autonomous Vehicles

通过连接和自动驾驶车辆的流量级协调提高高速公路合并效率

作者：J. Zhu, L. Wang, I. Tasic, and X. Qu

摘要：In this article, a traffic coordination strategy that improves the on-ramp merging of connected and autonomous vehicle (CAV) is proposed. The strategy consists of proactive gap creation on the mainline freeway and platoon merging on the ramp. The coordination is solved as a constrained optimization problem, incorporating macroscopic and microscopic traffic flow models, to determine the optimal coordination plan adaptive to real-time traffic conditions. The simulation results show that the strategy is compatible with real-world implementation and substantially improves the overall on- ramp merging efficiency, especially under high traffic volumes where recurrent traffic congestion is prevented, and merging throughput increased.

关键词：Connected and autonomous vehicles, coordinative merging strategy, freeway on-ramp merging, microscopic simulation, optimization.

29. MADRL-Based URLLC-Aware Task Offloading for Air- Ground Vehicular Cooperative Computing Network

基于多智能体深度强化学习的超可靠低时延通信感知任务卸载，用于空-地车辆协作计算网络

作者：P. Qin, Y. Wang, Z. Cai, J. Liu, J. Li, and X. Zhao

摘要：With the rapid development of 5G and Internet of Vehicles technologies, vehicles have evolved from mere transportation devices to mobile living spaces for humans, making relying solely on on-board processing capabilities insufficient. Traditional RSU-based edge computing has limitations such as high deployment cost and finite coverage. To address those issues, the authors construct an air-ground vehicular cooperative computing network that introduces cooperative vehicles to reduce deployment costs while incorporating UAVs to expand communication coverage. They aim to balance offloading efficiency and environmental sustainability by proposing a system cost minimization problem. However, it faces new challenges such as the lack of global state information, and the URLLC queue delay constraints, rendering traditional methods inadequate. Therefore, they employ Lyapunov optimization to partition the initial problem into two distinct sub-problems and solve them by deep reinforcement learning and greedy algorithm, respectively. The experimental results showcase notable achievements of the approach.

关键词：Air-ground vehicular cooperative computing network (AVC2N), task offloading, URLLC-aware, multi-agent deep reinforcement learning (MADRL).

30. Vertical-Longitudinal Comprehensive Control for Vehicle With In-Wheel Motors Considering Energy Recovery and Vibration Mitigation

考虑能量回收和振动减弱的带轮内电机车辆的纵向-纵向综合控制

作者：C. Xing, Y. Zhu, J. Wang, and W. Wang

摘要：Considering energy recovery and vibration mitigation, a vertical-longitudinal comprehensive control method for the vehicle with in-wheel switched reluctance motors is proposed. A braking force distribution controller is built to improve vehicle endurance mileage. An optimized linear quadratic regulator controller for an active suspension system is built to improve vehicle ride comfort. A coordination controller is designed to improve vehicle comprehensive performance. The simulation and hardware-in-the-loop results under several braking conditions show the proposed method can effectively enhance vehicle endurance mileage and braking comfort.

关键词：Electric vehicle, in-wheel switched reluctance motor, braking energy recovery, vibration mitigation, active suspension.

31. Maneuver Coordination for CAVs: ITS-G5 Channel and Service Performance Evaluation

自动驾驶车辆的操作协调：ITS-G5信道和服务性能评估

作者：K. Garlichs, A. Willecke, A.-C. Hagau, and L. C. Wolf

摘要：After cooperative awareness and collective perception, maneuver coordination will be the next key functionality for connected automated vehicles. It will enable them to exchange planned trajectories and actively cooperate via maneuver coordination messages (MCMs). The study presented in this article analyzes the capacity and performance of an ITS-G5 communication channel employed to exchange these messages, varying a set of different parameters. One of Europe’s most frequented highway intersections was modeled for the simulations, including realistic traffic patterns. The study pro- vides insights into what to expect from maneuver coordination for researchers, engineers, and regulators on both ends of the communication stack.

关键词：Connected vehicles, cooperative systems, V2V communication, maneuver coordination.

32. A Novel Framework Combining MPC and Deep Reinforcement Learning With Application to Freeway Traffic Control

将模型预测控制与深度强化学习相结合的新框架在高速公路交通控制中的应用

作者：D. Sun, A. Jamshidnejad, and B. De Schutter

摘要：The authors present a novel framework that integrates model predictive control (MPC) and deep reinforcement learning (DRL). This combined MPC-DRL framework strategically leverages the strengths and mitigates the weaknesses inherent in both MPC and DRL methods. The application of this framework is demonstrated in freeway traffic management. The simulation results reveal that the proposed method exhibits accelerated learning compared to conventional DRL algorithms and surpasses the individual performances of standalone MPC and DRL methods. Importantly, this versatile framework holds applicability beyond freeway traffic management and can be effectively utilized in various other domains.

关键词：Freeway network management, model predictive control, deep reinforcement learning, hierarchical structure.

33. Goal-LBP: Goal-Based Local Behavior Guided Trajectory Prediction for Autonomous Driving

Goal-LBP：基于目标的局部行为引导的轨迹预测，用于自动驾驶

作者：Z. Yao, X. Li, B. Lang, and M. C. Chuah

摘要：In trajectory prediction, predicting future locations is challenging due to the difficulty of learning accurate intentions and modeling multimodality. Historical paths at a certain location can help predict the future trajectory of an agent currently located in that position and address these limitations. In this work, the authors propose a goal-based local behavior-guided model, goal-LBP, using such information (referred to as local behavior data) to generate potential goals and guide the trajectory prediction conditioned on such goals. Goal-LBP uses a transformer to extract homogeneous features and attention mechanism to represent the heterogeneous interactions and subsequently uses an encoder–decoder gated recurrent unit to generate predictions. They evaluate goal-LBP using two large-scale real-world autonomous driving datasets, namely, nuScenes and Argoverse and goal-LBP can achieve SOTA performance. In addition, the local behavior estimator block can be easily added to existing SOTA methods and improves the performance by at least 10%.

关键词：Trajectory prediction, historical local data, goal based method, attention mechanism.

34. CEAMP: A Cross-Domain Entity Authentication and Message Protection Framework for Intra-Vehicle Network

CEAMP：用于车内网络的跨域实体认证和消息保护框架

作者：C. Shang, J. Cao, J. Liu, Y. Zhang, B. Niu, and H. Li

摘要：

Controller area network (CAN) is the most wide-used bus system in intra-vehicle networks (IVNs). However, the nature of broadcast communication and the lack of security mechanisms make the CAN bus extremely fragile against malicious attacks. In this article, the authors propose a security framework for the CAN bus, covering ECU entity identity management and authentication, symmetric key generation and update, intra-domain, cross-domain secure transmission, and sensitivity-based security classification methods. They formally verify the protocols using the up-to-date tool Tamarin and simulate real attacks on Arduino boards. The performance analysis results show that the increased time of a frame for a single ECU in the proposed cross-domain scheme is 134.30–164.80 µs on Arduino DUE, which takes up 6.72%–8.24% of a 10 ms frame. To the best of our knowledge, this is the first proposed IVN cross-domain secure transmission protocol without changing the IVN network topology or the CAN protocol.

关键词：Intra-vehicle network, CAN bus, cross-domain, encryption, MAC.

35. Predictive Cruise Cloud Control Scheme Design on Notable Vehicles-Under the Perspective of Cyber-Physical Systems

网络物理系统视角下显著车辆的预测巡航云控制方案设计

作者：J. Lin, Y. Li, and H. Xiao

摘要：Under the perspective of cyber–physical systems (CPSs), the multi-scale relationship between macro-scale traffic flow and micro-scale notable vehicles (NVs) in complex highways and their tunnel scenarios is explored. The authors propose a collaborative optimization layered cloud control architecture for NV. The proposed scheme introduces the multi-source heterogeneous information data-driven model to predict the traffic state after pre-factorization processing and the predictive cruise cloud control (PCCC) algorithm solver to optimize the proposed objects of study under restrictive constraints with the prediction information and digital information. The effectiveness and feasibility of the architecture under several traffic scenarios are fully evaluated through extensive simulations in traffic ground truth.

关键词：Cyber-physical systems, predictive cruise cloud control, cloud control system, traffic prediction, deep-learning algorithm, vehicle-cloud layered control.

36. CapsLoc3D: Point Cloud Retrieval for Large-Scale Place Recognition Based on 3D Capsule Networks

CapsLoc3D：基于3D胶囊网络的大规模地点识别点云检索

作者：J. Zhang, Y. Zhang, M. Liao, R. Tian, S. Coleman, and D. Kerr

摘要：This work presents a novel point cloud-based place recognition framework tailored for large-scale scenes. The proposed method enables robust global localization in GPS-denied environments and reliable loop-closure detection within SLAM systems. Extensive experiments demonstrate the detrimental impact of dynamic objects on recognition accuracy, under- scoring the importance of their removal. Leveraging a multi- scale perception field and feature correlation modeling, the network extracts discriminative point cloud features, leading to significant performance gains compared to existing techniques.

关键词：Lidar place recognition, moving object segmentation, capsule network, global localization, place feature learning.

37. A Privacy-Preserving-Based Distributed Collaborative Scheme for Connected Autonomous Vehicles at Multi-Lane Signal-Free Intersections

基于隐私保护的分布式协作方案，用于多车道无信号交叉口的自动驾驶车辆

作者：Y. Zhao, D. Gong, S. Wen, L. Ding, and G. Guo

摘要：This article proposes a privacy-preserving distributed collaboration (PPDC) scheme for connected autonomous vehicles (CAVs) to cross signal-free intersections based on the cloud while securing the private data of the vehicles. First, this article converts the cooperation problem into a multi-objective problem that aims to improve the efficiency of traffic and fuel economy. Second, to prevent the privacy of the transmitted data of vehicles from being inferred by untrusted cloud servers or external attackers, an affine masking-based privacy strategy is designed. Simulation examples show that the proposed PPDC scheme can guarantee collision avoidance and the privacy protection of transmitted data of CAVs, improve traffic efficiency as well as fuel economy, and avoid extensive computation burden.

关键词：Connected and automated vehicles, privacy preserving, signal-free intersections, vehicle-cloud collaboration system.

38. Battery Charging and Swapping System Involved in Demand Response for Joint Power and Transportation Networks

需求响应中涉及的电池充电和交换系统，用于联合电力和交通网络

作者：J. Bai, T. Ding, W. Jia, C. Mu,  P. Siano, and M. Shahidehpour

摘要：A hybrid swapped battery charging and logistics dispatch model in the continuous-time domain is proposed to dispatch battery charging and swapping system (BCSS) optimally. BCSS improves the shortcomings of traditional BSS with potential safety and space issues associated with local charging by separating the charging stations from the switching stations and coupling them through the transportation network. Considering the demand response in the model, this article proposes a rectangular boxing algorithm based on time-of-use tariffs and a vehicle routing algorithm that considers the cost of traffic congestion to complement the models of the power grid and the transportation network, respectively. The effectiveness and benefits of the proposed model are discussed.

关键词：Electric vehicles, transportation network, rectangle packing problem, battery charging, and swapping system, demand response.

39. Real-Time Asphalt Pavement Layer Thickness Prediction Using Ground-Penetrating Radar Based on a Modified Extended Common Mid-Point (XCMP) Approach

基于修改后的扩展公共中点方法的地面穿透雷达实时沥青路面层厚度预测

作者：S. Wang, Z. Leng, X. Sui, W. Zhang, T. Ma, and Z. Zhu

摘要：The conventional surface reflection method has been widely used to measure the asphalt pavement layer dielectric constant using ground-penetrating radar (GPR). This method may be inaccurate for in-service pavement thickness estimation with dielectric constant variation through the depth, which could be addressed using the extended common mid-point method (XCMP) with air-coupled GPR antennas. However, the factors affecting the XCMP method on thickness prediction accuracy have not been studied. Manual acquisition of key factors is required, which hinders its real-time applications. This study investigates the affecting factors and develops a modified XCMP method to allow automatic thickness prediction of in-service asphalt pavement with non-uniform dielectric properties through depth. A sensitivity analysis was performed, necessitating the accurate estimation of time of flights (TOFs) from antenna pairs. A modified XCMP method based on edge detection was proposed to allow real-time TOFs estimation, then dielectric constant and thickness predictions. Field tests using a multi-channel GPR system were performed for validation. Both the surface reflection and XCMP setups were conducted. The results show that the modified XCMP method is recommended with a mean prediction error of 1.86%, which is more accurate than the surface reflection method (5.73%).

关键词：Asphalt pavement, ground-penetrating radar, layer thickness, extended common mid-point method.

40. User Fairness Optimization of IRS-Assisted Cooperative MISO-NOMA for ITS With SWIPT

用于智能交通系统的IRS辅助合作MISO-NOMA的用户公平优化与能量收集

作者：Z. Yang, L. Xia, J. Cui, X. Li, Y. Wu, Z. Dong, and Z. Ding

摘要：The authors consider a cooperative multiple-input single- output non-orthogonal multiple access (MISO-NOMA) for ITS with intelligent reflecting surface (IRS) and simultaneous wireless information and power transfer (SWIPT). A user fairness optimization problem is formulated to maximize the fairness rate of the vehicles, subject to the quality of service requirements of the vehicles and the successive interference cancellation. For solving the challenging problem, an iterative successive convex approximation and semi-definite relaxation- based algorithm is proposed. The experimental results illustrate that the user fairness of the proposed cooperative MISO-NOMA for ITS with IRS and SWIPT is better than that of both the IRS-NOMA for ITS without SWIPT and the IRS-OMA for ITS.

关键词：Intelligent reflecting surface, intelligent transportation systems, non-orthogonal multiple access, simultaneous wireless information and power transfer, user fairness optimization.

创立宗旨：交通邦旨在构建泛交通领域最大的交流分享平台，促进产学研融合，最小化信息不对称。欢迎各位学生/教授/专家/企业在学术成果、招生计划、书籍推介、内推岗位、比赛冠名、会议合作等与交通邦合作！

联系方式：添加交通邦小助手微信（jiaotongbang），备注“姓名-学校-合作内容”。