Effect of Number of Lanes on Traffic Characteristics of Reinforcement Learning Based Autonomous Driving

Abstract

Traffic characteristics such as signalized intersections and high vehicle density combined with low vehicle speeds account for 12-55% increase in commute time. Increasing the number of lanes in the highway infrastructure can help increase the highway capacity and consequently reduce the associated commute delay. However, due to human-related features such as driver behavior and vehicle interaction as well as induced demand, it is recommended to limit the number of highway lanes to four. Recently, in view of the rising discussion regarding the deployment of Autonomous Vehicles (AVs), it is important to study the effect of the number of highway lanes on traffic characteristics with respect to AVs. This will provide insights into the full potential of AVs in terms of reducing commute time as well as provide crucial insights into the design of future road networks. Therefore, in this study, RL-based AV frameworks are developed to investigate the effect of the number of highway lanes on traffic characteristics. Specifically, we study the effect of the number of lanes on traffic characteristics in terms of travel speed, collision, and driving on the right-most lane. Results with two well-known RL-based AV frameworks simulated on highways with lanes ranging from 2 to 8 and increasing vehicle count respectively show improved traffic characteristics as the number of lanes increases. However, improvement depends on the RL algorithm employed.

Keywords

• Number of highway lanes
• autonomous vehicles
• reinforcement learning
• traffic characteristics