Journal of Transportation Engineering, Part A: Systems

The location of curb ramps, crosswalk markings, and curb radius are important contributors to pedestrian safety at intersections, particularly for conflicts and crashes involving right turns. Drivers tend to have higher turning speeds at locations with larger curb radii, which also requires greater stopping sight distance. If the crosswalk is placed further away from the intersection corner (i.e., offset or setback from the intersection corner), it could potentially provide drivers better visibility of pedestrians and increase the available stopping distance. However, no study has previously been conducted to investigate the relationship between crosswalk location and curb radius. This paper presents the results of research that quantifies the relationship between the offset of crosswalks (setback crosswalk) and indirect safety measures (driver stopping, yielding, and speed selection) at intersections with the consideration of other characteristics during simulated driving. The study included field-based observations and driving simulator experiments. A driving simulator experiment with 50 participants was conducted to determine how setback distances, sharpness of curb radii, and presence of pedestrians affect driver stopping decision and position, and speed choice. Approximately 12 h of video (7:00 a.m.–7:00 p.m.) were collected at nine study sites that closely matched the designs tested in the driving simulator. Comparisons of drivers’ speed in a similar scenario obtained from field and simulator data were performed to benchmark the observations from the driving simulator against field data. Stop line speeds were found to be consistent between experiments and turning speeds were found to be slightly higher in the driving simulator experiment. The research results suggest that curb radius should be smaller [less than 9 m (30 ft)] to control driver speed. Additionally, a setback distance of 6 m (20 ft) is a suitable upper bound for crosswalk placement when reconstructing intersections.

Dynamic systems modeling (DSM) provides an opportunity to systematically investigate traffic system dynamics at the network scale. However, simulation-based DSM approaches are often criticized for their heavy computational burden and low efficiency, as well as the difficulty in calibrating and validating their results. In this study, we propose an analytical modeling approach that integrates the analysis of both demand and supply sides using the polynomial arrival queue (PAQ) model. This model approximates the arrival rate by a polynomial function and the discharge rate by a constant, offering an efficient and accurate representation of traffic dynamics. Our proposed approach allows for the joint optimization of demand regulation policies and infrastructure capacity-building efforts, addressing the shortcomings of locally oriented congestion reduction strategies in isolation. This integrated approach is of great importance for transportation network modeling, management, and control. Overall, our study presents a novel and effective approach to transportation engineering, utilizing DSM and analytical modeling to optimize both demand and supply sides, ultimately improving transportation network performance.

This paper implements a two-staged ensemble learning model for traffic forecasting, focusing on the interpretability of predictions. The stacking model leverages the advantages of its diverse component learning models. Experiments on high-dimensional and sparse data validate the stacking model’s superior predictive accuracy compared to baseline models, including LightGBM and XGBoost. In addition to validating the stacking model’s outstanding predictive performance, this paper emphasizes the interpretability of its predictions by proposing an innovative explanation model based on feature contributions. This explanation model addresses the high dimension and sparsity in data prevalent in transportation engineering with its integration of resampling and consensus clustering, offering a scalable, stable, and computationally efficient solution ideal for real-time and large-scale applications. The paper presents theoretical justification, experimental results, and empirical validation of the interpretation model. Extensive experiments demonstrate the model’s enhanced stability compared to traditional shapley additive explanations (SHAP) implementations such as kernel SHAP. Investigating trade-offs between stability and computational efficiency of resampling provides insights for optimal configuration choices. This paper contributes to traffic flow prediction with broad applicability in real-time and large-scale traffic management scenarios, underscoring the vital role of ensemble learning and interpretable machine learning in contemporary data-driven decision making processes.

Abstract

Connected and autonomous vehicles (CAVs), with their potential to enhance the interactive perception of vehicle behavior, are expected to benefit traffic congestion and travel efficiency. However, the research scenarios in most current literature are oversimplified and limited, such as a road section or an intersection. To address this issue, this paper proposes a congestion avoidance routing strategy for CAVs to reduce the occurrence and propagation of congestion at the network level. Unlike rerouting after detecting the congestion downstream, the floating-car data are utilized to extract the network congestion patterns, based on which the routes of CAVs are optimized and updated. A simulation framework was built to model the network consisting of CAVs and human-driven vehicles (HDVs). Cooperative adaptive cruise control (CACC) and intelligent driver model (IDM) car-following models were set to characterize the driving behavior of CAVs and HDVs. Simulation experiments were conducted to examine the performance of the proposed routing strategy. The results indicate that the proposed CAV routing strategy can significantly improve the overall congestion state of the network. Compared with the full HDV environment, the vehicles’ average delay can be reduced by up to 46.7% and the travel time by up to 28.2% if all vehicles are switched to CAVs. The sensitivity analysis on CAV penetration rate and vehicle inflow rate shows that the vehicles’ average delay and travel time decreases with the CAV penetration rate increase, and the travel efficiency of CAVs outperforms HDV users sufficiently. Moreover, the benefits of CAVs would be weakened with the increase in vehicle inflow rates. Finally, the findings also provide a reference for CAVs’ centralized control strategy in urban intelligent transportation construction.

Practical Applications

Congestion has always been a problem in traffic networks, and causes substantial economic loss and environmental pollution. However, the emergence of connected and autonomous vehicles brings new potential for improving traffic efficiency via the technologies of autonomous driving, vehicular communication, collaborative perception, and so on. This paper evaluates the impact of CAVs on road congestion under the network mixed with HDVs. A routing control strategy for CAVs based on the road congestion patterns extracted from the taxi trajectory data is proposed, while HDVs are assumed to follow the user equilibrium (UE) principle and choose the fastest path. We obtain that CAVs under the proposed routing strategy demonstrate significant benefits in improving the congestion state of the road network. Regardless of the vehicle inflow rate, the higher the CAV penetration rate, the more significant the mitigation in the average delay and travel time. Also, the increasing CAV penetration rate can promote the reduction of HDVs’ average delay and travel time. Moreover, traveling via CAVs is much more stable and time-saving than HDVs. Finally, with the same model parameters setting and higher vehicle inflow rate, the CAVs’ advantage in mitigating the overall road congestion and improving HDVs’ traveling efficiency will be weakened. These findings provide a reference for CAVs’ centralized control strategy in urban intelligent transportation construction.

Finding effective routes is critical to support the decision-making of first responders to save lives in emergency and accident situations. Effective routing planning requires accurate traffic information including railroad blockage that could lead to significant delays. This study aims to improve the decision-making efficiency of emergency responding by integrating railroad crossing blockage and dynamic traffic information into routing planning process. Specifically, we propose a data-driven algorithm that considers real-time train blockage time windows in the route planning process. Through a series of first responder vehicle dispatching scenarios in Columbia, South Carolina, we demonstrate that the framework can assist first responders in finding the best route with the minimum response time in an emergency. Results from the scenarios show that the method can save as much as 61.6% of response time compared with the time calculated based on existing practices. This study demonstrates the benefit and necessity of integrating train operation information into emergency responding vehicle route planning to save lives.

Parking spots are a premium commodity, especially in dense downtown settings, so this study examines the service impacts of shared autonomous vehicles (SAVs) parking in legal on- or off-street locations when idle across Travis County in Austin, Texas. Using an agent-based activity-based travel demand model with dynamic traffic simulation, two restricted-parking strategies for SAVs were simulated. SAVs either found the nearest available parking spot or the lowest-cost spot (via a tradeoff of parking fees and distance-based costs). Two comparisons were conducted to analyze the impacts of these strategies. First, two restricted parking strategies were compared, where SAVs park without competition with private human-driven vehicles (HVs) for parking locations. Second, a more realistic analysis compared two SAV parking strategies with a scenario where SAVs remain idle in place. Private HVs in all scenarios and strategies of this comparison park at the closest designated location unless they opt for private parking. Using a supply of 8,400 aggregated parking locations in Austin, this study simulated fleet performance under different trip demands, with SAV fares of $0.62 per kilometer ($1 per mile) plus a $1 fixed pickup fee with dynamic ridesharing permitted. Parking costs were negligible in both SAV parking search strategies applied to the Austin network because of the region’s provision of mostly free parking. Requiring SAVs to park on designated on- and off-street parking locations and parking lots (restricted parking) also increased parking costs for HV drivers by up to 22% since SAVs occupied some free parking spaces, especially in the least-cost parking search strategy.

Tracking trajectories of the unconnected vehicles contributes to the improvement of traffic efficiency and safety. However, the effects of occlusions on the accuracy and reliability of the tracking results are nonnegligible. To address this issue, a modified multiple objects tracking algorithm was proposed to reduce the loss of trajectories caused by occlusions. The proposed algorithm was based on the multiple objects detection results, in which the motion states of the detected objects were determined. Further, the Kalman filter was employed to predict the trajectories, and each trajectory was uniquely matched to the detected object labeled with a tracking identity (ID) using the Hungarian algorithm. Afterward, a flexible life cycle strategy was proposed in terms of the speeds and accelerations of the detected objects, which controls the life cycle of the labeled trajectories when the occlusion occurred and guarantees the continuity of trajectories. The proposed tracking algorithm was tested on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) data set, and the classification of events, activities, and relationships workshops, multiple object tracking (CLEAR-MOT) metrics was introduced to evaluate the performance of the proposed algorithm. The results indicated that the proposed algorithm with appropriate life cycle apparently increased the precision and accuracy and reduced the influence of occlusions on multiple objects tracking. Future work will concentrate on field implementations of the algorithm, and various scenarios and weather conditions will be taken into account.

This paper optimizes the assignment of shared automated vehicles under users’ uncertain departure times. Automated vehicles can drive themselves, so no staff are needed to relocate vehicles in the one-way carsharing system. To optimize fleet placement and use, a two-phase solution method was established. Phase 1 strategically distributes vehicles across stations using a system optimization approach, while Phase 2 tracks vehicle movements via an agent-based simulation model. Phase 1 solutions serve as inputs to Phase 2 simulations. Using a fleet size of roughly 10,000 vehicles, case study applications were run across the Austin, Texas region's six-county network. In the base case setting, results suggest that system profits are optimized when vehicle rental is priced at $\$1.28/\mathrm{km}$ ($\$0.8/\mathrm{mi}$). The number of proactive vehicle relocations falls 9.8% if the relocation operation cost rises from $\$0.096/\mathrm{km}$ ($\$0.06/\mathrm{mi}$) to $\$0.32/\mathrm{km}$ ($\$0.2/\mathrm{mi}$). Average per-trip profit is $10.60 when using high-cost vehicles, and $11.60 when using low-cost vehicles. Results from a 3-h simulation show an average person-trip length of 25 km (15.6 mi), with 29.6 min of average driving time. When a 24-h day was simulated, the vehicle-occupied time and vehicle-distance traveled were 4 h and 200 km (125 mi) per vehicle-day, respectively. The low coefficient of variation of satisfied demand across 30 demand scenarios suggests the robustness of the two-phase solution method.

At present, autonomous driving decision-making solutions take few elements into account while ignoring the unpredictable nature of driving behavior, which makes it challenging to manage complicated traffic situations. To this end, we present a decision-making architecture in this paper that enhances the existing reinforcement learning methodology by combining the bootstrapped technique and the random prior network (RPN). The RPN can give each learner a neural network with unique weights to avoid the contingency created by the artificially built prior functions, while the Bootstrapped technique can balance out the exploration and exploitation. The ego vehicle was trained by three algorithms and verified in random environments to evaluate the effectiveness of our method. The results show that our algorithm outperformed the current reinforcement learning algorithms.