Routing Control for Automated Highway Systems

Abstract

Traffic congestion is one of the most important problems that have not been solved in the modern society for a long time. Transportation infrastructure, especially highways, is not easy to respond to changes in traffic demand. Constructing a new highway or extending existing highway requires significant cost and time. Therefore, it is important to utilize the current infrastructure efficiently. Automated Highway System (AHS) is one of the related future studies to improve the performance of the traffic system. In AHS, the central system can manage the decision of routing for all vehicles and distribution of the traffic volume. Once vehicles enter the AHS and specify their destination, drivers do not need to drive or select their route. The AHS can decide the routes of vehicles and the autonomous vehicles are operated by communicating with the AHS. The introduction of the AHS is expected to significantly improve the highway traffic performance. This study focuses on routing algorithms for AHS, named Automated Highway Routing Problem (AHRP). AHRP is to find the appropriate route for each vehicle in the AHS. The vehicle speed is changed by the number of vehicles on the road. Depending on how AHS selects the route of each vehicle, the road density of the road network changes and the travel time of each vehicle also changes. The first objective of AHRP is to increase the efficiency of a given AHS. However, to implement routing algorithms in the real world, equity must be considered, too. If several vehicles with the same destination depart at the same time, their arrival times should be as close as possible to each other even if their assigned routes are different. Thus, an efficient routing algorithm must improve the efficiency of the highway network as well as the equity of individual vehicles. In order to consider the equity of individual vehicles, an agent-based simulation model is used to develop and evaluate the routing algorithms. The efficiency objective such as average travel time can be obtained by both aggregated and disaggregated simulation models. However, the equity objective such as travel time difference between the vehicles with the same origin–destination (OD), needs a disaggregated model. Since the agent-based simulation model can be considered for both objectives, it is used in this study. This study proposes four solution approaches for AHRP. Suggested four approaches are based on shortest path algorithm, but they have different cost functions. The first approach uses length of highway link, which is static value in the road network. The second approach uses the average travel time of each link, which is updated periodically by AHS. The third approach uses predicted travel time of each link. The fourth approach also uses predicted travel time of each link, but it additionally considers predicted travel time difference between multiple routes. Simulation results of solution approaches on two highway network models with a couple of scenarios are also presented. The first model is a basic road network model with one origin node and two destination nodes, and the second model is complex road network model with three origin nodes and five destination nodes. Based on the analysis of the travel time of each vehicle by route, the performance of solution approaches is compared. The comparison results show that the while the prediction based algorithms perform well on the small model as expected, they do not perform well on the large network. From the careful analysis of prediction errors, we found out that the network structure caused travel time underestimation and accumulation of prediction error caused overestimation. In order to overcome the problems, we propose Q-learning based travel time prediction approaches. Simulation studies on both basic and complex road network models show that Q-learning based travel time prediction approaches are effective in increasing performance and decreasing prediction errors. This study focuses on AHP, which is future studies for the transportation system and basically requires autonomous vehicle systems. However, note that if the relevant traffic information is collected and appropriately shared, and human drivers follows the guidelines (suggested route selection) generated by the solution approaches, the proposed approaches can be applied to even today’s highway networks.