Strategic Control for Autonomous Car on Freeway Merging Sections

抄録

With the expansion of the freeway network, traffic congestion at freeway merging points is expected to increase. While ordinary drivers focus only on safe merging for their own vehicles, automated vehicles can apply a variety of merging behaviors as long as safety and comfort are ensured. In such cases, it would be possible to strategically implement merging behaviors that contribute to the overall optimization of traffic flow. Therefore, in this study, we model merging behavior by reinforcement learning, assuming an automated vehicle. In addition, four reward settings (safety, comfort, efficiency, and overall efficiency) are proposed and analyzed for differences in vehicle behavior and their effects on surrounding vehicles. As a result, we were able to improve the safety, comfort, and efficiency of merging vehicles and reduce their average trip time.