Accurately modeling car-following decisions is fundamental for understanding traffic flow dynamics and enabling the safe operation of connected and automated vehicles (CAVs) in mixed traffic environments. While theorydriven models offer interpretability, they often struggle to capture the complex, nonlinear stochasticity of human driving behavior. Conversely, prevalent data-driven models, such as those based on Long Short-Term Memory (LSTM) networks, still face challenges in achieving robust long-term prediction accuracy and maintaining low collision rates. To address these limitations, this paper proposes a novel microscopic carfollowing model based on the Temporal Convolutional Network (TCN). The proposed architecture leverages dilated causal convolutions and residual connections to effectively capture both long- and short-term temporal dependencies in vehicle trajectory data, ensuring strict causality in prediction. The model is trained and validated using the HighD trajectory data. A comprehensive Bayesian optimization framework is employed to systematically tune the model's hyperparameters, enhancing its predictive performance and generalization capability. Simulation results demonstrate that the optimized TCN model significantly outperforms several state-of-the-art benchmarks, including the Intelligent Driver Model (IDM), LSTM, and Transformer, in terms of mean squared error on spacing prediction. Furthermore, the proposed model achieves a substantial reduction in collision rate compared to other deep learning approaches, highlighting its superior balance between prediction fidelity and safety in microscopic traffic simulation.

The crossing behavior of pedestrians in virtual reality (VR) environments has been extensively used in traffic safety research; however, the kinematic differences compared to real-world situations remain insufficiently explored. This study constructed a virtual scene that accurately matches real road conditions and recruited 179 participants of various ages to conduct crossing experiments at three speeds: normal walking, fast walking, and normal running, within three distinct environments: real, spacious virtual, and narrow virtual. Using motion capture and questionnaires, parameters such as gait, trajectory, and limb angles, as well as subjective experiences, were quantified. The findings indicate that differences between real and virtual environments primarily manifest in speed, stride length, and trajectory deviation, while limb angles representing core movements showed no significant differences. In the spacious virtual environment, pedestrian trajectories maintained a straight line during the initial 50% (approximately 2.81 ± 0.15 m), with deviations increasing with age and speed; conversely, the narrow virtual environment resulted in significant distortions in movement patterns. Post-experiment, participants estimated the onset of dizziness at 16.9 ± 2.79 min. Analysis of individual differences revealed high consistency in movements within both environments, yet substantial variability existed among participants. Based on these findings, this research recommends conducting VR crossing experiments in spacious settings, at normal speeds, and limiting walking distance to approximately 2.8 m to ensure participant safety and comfort while obtaining data that closely reflect real-world pedestrian kinematics. This guideline provides empirical support for future high-fidelity traffic safety studies using VR technology.

A key factor affecting the motion fidelity of vehicle driving simulators is the design of scaling methods in washout algorithms (WAs). Conventional scaling methods use fixed gain parameters with poor consistency across drivers and scenarios, while recent offline-calibrated polynomial nonlinear scaling methods are constrained to static pre-tuning and cannot adapt to complex driving conditions. To address this, this study proposes a deep reinforcement learning-based nonlinear scaling (DRL-NS) method that dynamically adjusts the gain coefficients of the MCA. Formulating the gain adaptation as a sequential decision-making problem, a modified Proximal Policy Optimization (PPO) agent is adopted to learn an optimal nonlinear scaling strategy for high-fidelity motion cueing. Experimental results on a simulated motion platform show that the proposed DRL-NS framework achieves high scenario adaptability, significantly reducing errors in sensed specific force and angular velocity and improving workspace utilization.

With the advancement of communication and autonomous technologies, regular vehicles, connected vehicles, and connected autonomous vehicles (CAVs) will coexist on roads for a prolonged period, forming a mixed traffic environment. This paper presents a car-following model for CAVs based on the driving risk field (DRF), aiming to enhance traffic flow efficiency, stability, and safety within this mixed traffic environment. The proposed model employs the driving risk field theory to quantify interactions between vehicles, integrating multiple risk components: a risk item about the individual leading vehicle, a rear-end crash risk item accounting for the influence of the following vehicle, and an extra risk item to catch the dynamics of multiple leading vehicles. Based on the genetic algorithms (GA), the model's parameters were calibrated and optimized leveraging highD trajectory data and Sumo simulation. The numerical simulations confirm the model's effectiveness in improving traffic flow stability within mixed traffic environments. By accurately quantifying the risks associated with multi-vehicle interactions and ensuring more controlled acceleration adjustments, the proposed model demonstrates its capability to handle complex traffic dynamics. Compared to previous models, it provides enhanced performance, as evidenced by smoother response curves and better evaluation indicators observed across various driving scenarios. The study could support the development and practical deployment of driving assistance systems for CAVs and shows the potential to advance the safety and efficiency of traffic management systems for the mixed traffic environment.

Due to the obvious randomness, pedestrian crossing behavior is hard to predict, which challenges the decision-making of autonomous vehicles (AVs). Recent solutions have been able to adapt to structured road scenes with crossing signals or markings. However, there is still a gap in extending the pedestrian-vehicle interaction (PVI) performance in structured road scenes to unstructured road scenes. Therefore, this paper proposed a vehicle decision-making model considering pedestrian intention based on game theory and reinforcement learning (RL). We designed and conducted a simulation experiment based on a virtual reality platform. Then, leveraging game theory, we established a pedestrian crossing decision-making model considering pedestrian heterogeneity evoked by time pressure (TP). A reward function was developed to enhance driving performance by combining safety, efficiency, and comfort. The RL agent of AVs learns to control the vehicle speed in a pattern that maximizes cumulative rewards through trials and errors by interacting with pedestrians in the simulation environment. The results show that AVs can effectively and safely interact with heterogeneous pedestrians on unstructured roads based on the proposed model. This study contributes to developing AVs that interact better with pedestrians and improve traffic safety, efficiency, and user acceptance of autonomous vehicles.

Vehicle driving simulators have been widely used in fields including road design, automotive development, and driver training. As a core component of the simulators, motion cueing algorithms (MCAs) aim to reproduce realistic vehicle motion sensations while respecting workspace limitations for a high-fidelity driving experience. Recent advancements have identified model predictive control (MPC) as a promising approach for MCA. However, conventional MPC-based MCA faces computational bottlenecks that limit its performance. Although recent studies have employed swarm intelligence optimization algorithms, such as the genetic algorithm and grey wolf optimizer, to enhance the performance of MPC-based MCA through horizon adjustments, the improvement path from the computational cost perspective is still overlooked. Therefore, this paper proposes a costs-optimization framework for MPC-based MCA (COMPC-based MCA), which integrates two key components: an Operator Splitting Quadratic Program (OSQP) solver and a Snow Ablation Optimizer (SAO) that considers computational costs and motion sensation errors for parameter selection. Experiment results validated the performance of the proposed framework. COMPC-based MCA can achieve fast quadratic programming (QP) solving and select appropriate parameters for performance improvement while considering computational costs.

The pedestrian-vehicle conflict at uncontrolled mid-block locations often results in severe injuries. The present study developed a game theory-based pedestrian crossing decision-making model considering the effects of time pressure, vehicle speed, pedestrian-vehicle distance, and vehicle length. We used a pedestrian simulator to conduct a pedestrian crossing experiment with 54 subjects (40 males and 14 females. The experimental independent variables include vehicle gap (2s-6s, with an interval of 0.5s), time pressure (with or without time pressure), and driving status of the vehicle (decelerating to a stop or passing at a constant speed). The experimental dependent variable is the gap acceptance behavior. Based on the experimental data, we developed a pedestrian crossing decision-making model based on game theory and solved it with the mixed strategy algorithm. The parameters of the model were calibrated, and the validity was verified. The results show that the time pressure and vehicle gap have significant impacts on gap acceptance behavior, and the proposed pedestrian crossing decision-making model can effectively predict pedestrian crossing decision-making behavior under different time pressure conditions.

This paper presents a control strategy for the handling stability of a four-wheel drive formula student electric car. The wheel speed correction algorithm based on double track model and the car speed estimation algorithm based on Kalman filter are designed to lay the foundation for good handling and stability control. Based on the hierarchical control theory, direct yaw moment control is divided into three control layers: motion tracking layer, torque distribution layer and slip rate control layer. In the motion tracking layer, combined with the linear two-degree-of-freedom vehicle model, the anti-saturation integral PID algorithm is used to obtain the ideal value of the yaw rate. In the torque distribution layer, an additional yaw moment is generated by the method of equal distribution of inner and outer sides and proportional distribution of front and rear axles to realize the control of the driving stability of the car. In the slip rate control layer, a synovial variable structure control method is adopted to make the actual wheel slip rate the same as the target slip rate. A joint simulation model was built based on Carsim and Simulink, and the control strategy in-loop simulation verification was completed. An electric formula racing test platform was built, and the feasibility of the control strategy was further verified in the four-wheel drive formula student electric car.

With level 3 automated vehicles poised to appear on the roads soon, takeover remains a major challenge. At present, the effect of manual driving experience on takeover performance is unknown. Therefore, a simulator study was conducted to investigate the influence of driving experience (novice and experienced) on takeover performance in different takeover time budgets (7 s and 5 s) and in combination with a visual secondary task (i.e., surrogate reference task). Data from 48 young and middle-aged participants consisting of 24 novice and 24 experienced drivers were used for this study. Researchers found that the overall stability of evasive maneuvers by novice drivers was considerably worse than that by experienced drivers. A detailed analysis showed that the influence of driving experience on takeover stability was mainly reflected in longitudinal control rather than lateral control. A significant interaction between driving experience and visual secondary task showed that the latter had a substantial impact on the takeover stability of experienced drivers but not on that of novice drivers. Researchers also found that rich manual driving experience cannot make the takeover process of experienced drivers more stable than that of novice drivers under conditions of eye-off-road. In addition, no significant difference was found between novice and experienced drivers in automation disengagement time, takeover time and minimum time to collision. Results indicate that novice drivers have poor takeover stability and weak adaptability, but their longitudinal collision risk is not deteriorated by the lack of manual driving experience.

Eco-driving through multiple intersections has significant fuel benefit for road transportation. Most existing studies assume that vehicles travel at a constant speed between intersections, which naturally leads to large errors in fuel prediction and trajectory planning. This article focuses on eco-driving operation through multiple signalized intersections considering more realistic powertrain dynamics, which directly contain the engine and the transmission as well as aerodynamic drag, rolling resistance, and so on. This feature enables explicit descriptions of dynamic behaviors and fuel characteristics in acceleration, deceleration, and constant speed driving. An open-loop optimal control problem (OCP) is formulated to minimize fuel consumption between two red-signalized intersections, and the two- or three-stage operation rules are then proposed to approximate its optimal solution. The eco-driving of passing multiple intersections is then numerically solved by combining the two- or three-stage driving rule and the Dijkstra algorithm. This method can consider the acceleration and deceleration processes at each intersection and achieve improved fuel economy.

自动驾驶汽车是解决未来交通问题的重要途径,而我国自动驾驶人才不足,人才短缺将制约我国自动驾驶汽车产业的发展.为此,北航交通学院进行了自动驾驶与智能网联高级专门人才培养的探索与实践.设计了包括学位理论课程和综合实践环节两部分构成的课程培养体系,编写了6册入选"十三五"国家重点图书出版规划项目的自动驾驶系列教材,开设自动驾驶讲堂并邀请自动驾驶企业的技术负责人开讲座.实践方面主要包括开设实验课程、参加自动驾驶相关比赛及企业实习.为自动驾驶人才培养探索了切实可行的方法,为解决我国自动驾驶人才紧缺问题提供了思路.

Abstract Objective At conditionally automated driving, the driver can temporarily engage in non-driving related tasks (NDRTs). However, they must safely take over control when the automated driving system reaches its operation limit. Thus, understanding the effects of the NDRTs on driver take-over performance is essential. The present work investigates the effects of various NDRTs on motor readiness in take-over scenarios during conditionally automated driving. Methods Three driving simulator studies were conducted. 48, 49, and 22 participants were recruited in three experiments, respectively. The participants were distracted by different NDRTs (everyday task in Experiment 1, arrow task in Experiment 2, and SuRT in Experiment 3) on a tablet mounted in the vehicle. The everyday task included reading the news and watching a video, and the arrow task included a set of arrow matrices presented to the participants in sequence. The time budgets in Experiment 1 included 3 s, 4 s, and 5 s, and the time budgets in Experiment 2 and 3 included 5 s and 7 s. A take-over request (TOR) warning was issued in the automated driving condition when the participants encountered a broken-down car in front. The participants must regain control of the vehicle with the given time budget. The hands-on time was evaluated, measuring the time from the TOR until the hands touch the steering wheel. Results The task (arrow task and SuRT), time budget (5 s and 7 s), and gender did not affect the hands-on time. However, the hands-on time for the drivers with the everyday task was significantly shorter than that for the drivers with the arrow task in the 5 s time budget. Conclusions In conditionally automated driving, the arrow task and SuRT imposed a similar workload on readiness to take over control. Compared to the everyday task, the engagement in the arrow tasks consumed more workload on readiness to take over control.

Understanding driver behavior of conditionally automated driving is necessary to ensure a safe transition from automated to manual driving. This study aimed to examine the difference in take-over performance between high crash risk (HCR) and lower crash risk (LCR) drivers in emergency take-over situations during conditionally automated driving. In the current simulator study, a 3 x 3 (within-subjects) factorial design was used, including the task factors (no task, reading the news, and watching a video) and time budget factors (time budget = 3 s, 4 s, and 5 s). Forty-eight participants completed a WA drive on an approximately 10 km long two-way six-lane urban road. The participants firstly were in manual control and then switched to the automated driving mode at a speed of 50 km/h. The automated driving system was able to detect a broken car in the ego-lane and requested the driver to take over the control of the vehicle. There are at least one or two other vehicles or motorcycles on each side of the ego-vehicle, resulting in fewer escape paths. For the two non-handheld non-driving-related tasks (NDRTs), the participants were asked to be fully engaged in a task without any need to monitor the road environments. Each participant completed nine emergency take-over situations. The participants were classified into two groups that were labeled LCR (N <= 2) and HCR drivers (N >= 3) according to the number of accidents per driver. The results show that LCR drivers had shorter brake reaction time compared to HCR drivers. For all drivers, the engagement in a task led to longer response times, and the time budget affected the longitudinal vehicle control. In addition, the task affected the response times for LCR and HCR drivers, but only the time budget affected the longitudinal vehicle control for LCR drivers. For all drivers, LCR and HCR drivers, the time budget and task affected the safety of take-over. Especially, the two non-handheld everyday tasks seem to have a similar effect on the drivers' workload. Therefore, the HCR drivers had a lower hazard perception compared to the LCR drivers, and the factor regarding the individual difference of driving ability in take-over situations should be considered to design safe take-over concepts for automated vehicles.

Accurately estimating driving style is crucial for designing personalized autonomous driving to enhance market acceptance. Focusing driving style estimation while driving, a novel model defined as deep clustering is proposed. Since the next generation simulation (NGSIM) dataset is complex and high-dimensional, a parameterized non-linear embedding from the original data space to a low-dimensional feature space by using deep neural networks (DNNs) is proposed to alleviate the “curse of dimensionality.” We then propose a novel clustering layer to estimate the driving style of the encoded NGSIM data. Experimental results demonstrate that the NGSIM data divided into four groups shows better performance. Furthermore, compared with K-means, fuzzy C-means (FCM) and Gaussian mixture model (GMM), the proposed deep clustering model is capable of achieving superior performance in behavior analysis on public NGSIM dataset. Moreover, the deep clustering model has a stable performance on driving style estimation for different vehicle classes.

为了分析城市道路环境下高度自动驾驶中非驾驶相关任务和接管紧迫度对接管绩效的影响,基于驾驶模拟器设计了自动驾驶紧急接管场景并开展驾驶模拟试验,接管请求时间分别设定为3,4,5 s,非驾驶相关任务为读新闻、看视频、玩游戏,自动驾驶车速为50 km·h-1,试验中共招募了49名被试(男性30名,女性19名),被试的平均年龄为31.06岁(标准差为7.1岁),驾驶人在自动驾驶阶段始终执行非驾驶相关任务,听到接管请求提示后需要接管车辆的控制权,并实施紧急避让操作.研究结果表明:在紧急接管情况下,接管紧迫度对合成加速度和最小TTC有影响,而对接管时间无影响,与5 s的接管请求时间条件相比,3,4 s的接管请求时间条件下的合成加速度明显增加,而最小TTC则随接管请求时间的减少而降低;非驾驶相关任务对接管时间和最小TTC有影响,而对合成加速度无影响,与无非驾驶相关任务相比,非驾驶相关任务会显著增加接管时间和降低最小TTC;碰撞几乎都发生在3 s和4 s的接管请求时间下,5 s的接管请求时间能够基本保证接管的安全性.

为评价L3级自动驾驶车辆接管的安全性,基于驾驶模拟器设计了双向六车道高速公路环境下的接管场景并进行驾驶模拟实验,驾驶人在自动驾驶过程中始终执行视觉次任务操作,次任务为观看3种难度等级的箭头图,接管场景为自车行驶中遇到同车道前方的一辆抛锚车辆,接管请求时间设为7 s,自动驾驶车辆的速度为110 km/h.实验共计招募了49名被试(男性30名,女性19名),平均年龄为31.06岁(标准差为7.10岁).当车辆发出听觉+视觉的接管请求信号后,被试应通过按下转向盘上的切换按钮来获取车辆的控制权.以最小TTC的组别为因变量,设定最小TTC小于等于1 s为危险组,大于1 s为安全组,利用二元logistics回归建立接管安全性评价模型.研究结果表明:7s的接管请求时间条件下,影响接管安全性的因素主要是接管反应时间和次任务,本文中建立的接管安全性评价模型的预测准确率达85.5％.

This study aimed to investigate the take-over response characteristics of young drivers in conditional automated driving. It was focused on the influence of the different visual tasks(3× 3and 4×4arrow tasks)and the take-over-request lead time(TORlt, TTC is 5sand TTC is 7s) on the take-over time. We designed automated driving take-over scenarios including the different visual secondary tasks and take-over request lead time based on a driving simulator, and conducted the driving simulation tests with 29young participants. The effects of the different visual tasks and take-over-request lead time on the take-over time were studied using the two-factor variance analysis. The Pearson correlation test was used to analyze the correlation of different take-over times. The results show that the take-over time with secondary tasks increase significantly compared to that without secondary tasks, and different tasks have no significant effect on the take-over time. Further, no significant interaction occurred between the secondary tasks and take-over time with an obstacle in the front. The take-over time decreased significantly in the take-over scenarios with an obstacle in the front compared to that in scenarios without an obstacle. Additionally, different TORlts had no significant effect on the steering reaction time and the brake reaction time when the participants engaged in different secondary tasks. For the take-over scenarios with an obstacle in the front, the participants tended to apply the operation of combined brake and swerve, and the ratio of the brake only to the combined operation for different TORlts is the same. A strong correlation exists between the take-over reaction time and the brake reaction time(r =0. 7).

In the two-echelon city logistics system, satellites enable cargo transshipment. Considering on-street satellite phenomena in city logistics practices and the concept of road space sharing between delivery operations and city traffic, we introduce the two-echelon city logistics system with on-street satellites (2E-CLS-OS). The 2E-CLS-OS uses time windows and satellite transshipment constraints to make routes of each of the two echelons interacting. At various time windows of on-street satellites, the real-time transshipment capacity needed by the transshipment operation is constrained by the permitted maximal transshipment capacity and the occupied transshipment capacity. The mathematical formulation for the 2E-CLS-OS is developed. The variable neighborhood search (VNS) heuristic is provided, based on the initial solution constructed by the savings-based heuristic (CW). The model formulation and the VNS are tested by using 20 randomly-generated small-scale instances. The VNS is used to solve 33 realistic instances with up to 30 on-street satellites and 900 customers. The economic difference between the electric vehicles and diesel ones on the second echelon and the impact of varying the truck capacity and the electric tricycle capacity are experimentally analyzed.

Vehicle electrification can help to decrease environmental pollution. But fast electricity consumption and inconvenient recharging service restrict the widespread adoption of electric vehicles, especially in the taxi industry. This paper attempts to solve a problem of locating battery supplying infrastructures for electric taxies. The problem involves the integration of two inter-related decision sub-problems: location optimization of charging centers, location optimization of battery-exchanging stations (BESs). To solve it, minimal cost model and corresponding algorithm are designed. Then, a case study of Dalian, China is conducted to show the effectiveness of our model and algorithm. Computational results can provide the government with good insight into how to realize taxi electrician and therefore contribute to the further planning of vehicle electrification system.