Journal of Intelligent Transportation Systems最新文献

In this work we present a multiagent deep reinforcement learning approach for autonomous driving vehicles that is able to operate in traffic networks with unsignalized intersections. The key aspects of the proposed study are the introduction of route-agents as the main building block of the system, as well as a collision term that allows the cooperation among vehicles and the construction of an efficient reward function. These have the advantage of establishing an enhanced collaborative multiagent deep reinforcement learning scheme that manages to control multiple vehicles and navigate them safely and efficiently-economically to their destination. In addition, it provides the beneficial flexibility to lay down a platform for transfer learning and reusing knowledge from the agents’ policies in handling unknown traffic scenarios. We provide several experimental results in simulated road traffic networks of variable complexity and diverse characteristics using the SUMO environment that empirically illustrate the efficiency of the proposed multiagent framework.

The global positioning system (GPS) trajectory data are extensively utilized in various fields, such as driving behavior analysis, vehicle navigation systems, and traffic management. GPS sensors installed in numerous driving recorders and smartphones facilitate data collection on a large-scale in a high-frequency manner. Therefore, map-matching algorithms are indispensable to identify the GPS trajectories on a road network. Although the local map-matching algorithm reduces computation time, it lacks sufficient accuracy. Conversely, the global map-matching algorithm enhances matching accuracy; however, the computations are time consuming in the case of large-scale data. Therefore, this study proposes a method to improve the accuracy of the local map-matching algorithm without affecting its efficiency. The proposed method first executes the incremental map-matching algorithm. It then identifies the mismatching links in the results based on the connectivity of the links. Finally, the shortest path algorithm and the longest common subsequence are used to correct these error links. An elderly driver’s driving recorder data were used to conduct the experiment to compare the proposed method with four state-of-the-art map-matching algorithms in terms of accuracy and efficiency. The experimental results indicate that the proposed method can significantly increase the accuracy and efficiency of the map-matching process when considering high-frequency and large-scale data. Particularly, compared with the two-benchmark global map-matching algorithms, the proposed method can reduce the error rate of map-matching by nearly half, only consuming 18% and 58% of the computation time of the two global algorithms, respectively.

This paper presents a new ridesharing simulation model that accounts for dynamic driver supply and passenger demand, and complex interactions between drivers and passengers. The proposed simulation model explicitly considers driver and passenger acceptance/rejection on the matching options, and cancelation before/after being matched. New simulation events, procedures and modules have been developed to handle these realistic interactions. Ridesharing pricing bounds that result in high matching option accept rate are derived. The capabilities of the simulation model are illustrated using numerical experiments. The experiments confirm the importance of considering supply and demand interactions and provide new insights to ridesharing operations. Results show that higher prices are needed to attract drivers with short trip durations to participate in ridesharing, and larger matching window could have negative impacts on overall ridesharing success rate. Comparison results further illustrate that the proposed simulation model is able to replicate the predefined “true” success rate, in the cases that driver and passenger interactions occur.

Driving behavior monitoring by smartphone sensors is one of the most investigated approaches to ameliorate road safety. Various methods are adopted in the literature; however, to the best of our knowledge, their robustness to the prediction of new unseen data from different drivers and road conditions is not explored. In this paper, a two-phase Machine Learning (ML) method with taking advantage of high-pass, low-pass, and wavelet filters is developed to detect driving brakes and turns. In the first phase, accelerometer and gyroscope filtered time series are fed into Random Forest and Artificial Neural Network classifiers, and the suspicious intervals are extracted by a high recall. Following that, in the next phase, statistical features calculated based on the obtained intervals are used to determine the false and true positive events. To compare the predicted and real labels of the recorded events and calculate the accuracy, a method that covers the limitations of previous sliding windows is also employed. Real-world experimental result shows that the proposed method can predict new unseen datasets with average F1-scores of 71% in brake detection and 82% in turn detection which is comparable with previous works. Moreover, by sensitivity analysis of our proposed model, it is proven that implementing high-pass and low-pass filters can affect the accuracy for turn detection up to 30%.

Dedicated Lanes (DLs) have become prevalent on highways and arterial roads as they help accelerate carpooling vehicles or buses. However, capacity is wasted if the penetration rates of these vehicles with priority are low. Wasted capacity can be utilized optimally by implementing Vehicle-to-Everything (V2X) technology and granting General-Purpose (GP) vehicles the ability to traverse on DLs. However, existing research on flexible DLs has mostly focused on preset, static operating rules. In this study, we propose a true, active DL management strategy named Dedicated Lane with Intermittent Priority (DLIP) that operates at the vehicle level. An Optimal Right of Way Allocation (ORWA) model is proposed that maximizes the benefits of allowing GP vehicles into the DLs. To validate the proposed strategy, a simulation model based on VISSIM was developed. Results under various demand scenarios demonstrate that the proposed strategy outperforms traditional DL management strategies in terms of overall productivity, with improvements ranging from 10% to 25%.

Ephemeral incidents, or events in traffic or on the roadside that have only local and short-term impact on road safety and road capacity, are noteworthy for vehicles nearby—especially those approaching and planning to pass by. We study ways to communicate detected ephemeral incidents between connected vehicles, comparing various decentralized (vehicle-to-vehicle) communication strategies and weighing with established centralized mechanisms with regard to efficiency and broadcasting redundancy. The strategies are implemented in a simulation using realistic road networks, travel routes and traffic. We identify the strategy that achieves up to 100% success rate in transmitting incident messages to the affected vehicles under each scenario, while minimizing broadcast redundancy. In general, decentralized vehicle-to-vehicle communication strategies show strong potential to transmit incident messages efficiently and effectively.

This work is about the generation of driving styles for autonomous cars. Here, we propose a definition of driving style based on the partition of controller parameters for self-driving vehicles. The main contributions of this work are the following. 1) A methodology based on the controllers’ parameters for creating comfortable driving styles that can be used as autonomous cars’ operation modes. 2) A proposal to use virtual reality as a testbed for the evaluation of driving styles by users. 3) As an illustration of our methodology, we determine and evaluate distinguishable driving styles by partitioning the time-to-collision parameter of the Intelligent Driver Model (IDM) controller using the Just Noticeable Difference (JND). 4) A proposal of four driving styles that are equally preferable among passengers.

Real-time prediction of crash risk can support traffic incident management by generating critical information for practitioners to allocate resources for responding to anticipated traffic crashes proactively. Unlike previous studies using archived traffic data covering a limited highway environment such as a segment or corridor, this study uses a statewide live traffic database from HERE to develop real-time traffic crash prediction models. This database provides crowdsourced probe vehicle data that are high-resolution real-time traffic speed for the entire freeway network (nearly 2,000 miles) in Alabama. This study aims to use machine learning models to predict crash risk on freeways according to pre-crash traffic dynamics (e.g., mean speed, speed reduction) along with static freeway attributes. Traffic speed characteristics were extracted from the HERE database for both pre-crash and crash-free traffic conditions. Random Forest (RF), Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost) were developed and compared. Separate models were estimated for three major crash types: single-vehicle, rear-end, and sideswipe crashes. The model prediction accuracy indicated that the RF models outperform other models. Models for rear-end crashes are found to have greater accuracy than other models, which implies that rear-end crashes have a significant relationship with pre-crash traffic dynamics and are more predictable. The traffic speed factors that are ranked high in terms of feature importance are the speed variance and speed reduction prior to crashes. According to partial dependence plots, the rear-end crash risk is positively related to the speed variance and speed reductions. More results are discussed in the paper.