Energy Optimized Green Light Assist in Varying Traffic Scenarios Using Reinforcement Learning

Abstract

Optimizing traffic flow at signalized intersections is a central challenge in road networks. Existing approaches address this problem through measures such as adaptive traffic signal control to decrease travel time and reduce vehicle energy consumption. A complementary approach is the optimization of vehicle-level driving behaviour, which directly influence queue formation and stop-and-go dynamics. This paper evaluates deterministic and reinforcement learning approaches for recommending an approach speed towards a traffic light in a simple SUMO scenario. Using energy consumption and travel time as primary metrics, we compare both methods across traffic volumes. Baseline runs with SUMO’s default driving behaviour were conducted as reference, followed by experiments with two, six, and 20 vehicles. We achieve a decrease of energy consumption by 22.83% over the distance travelled in a scenario with 20 vehicles while roughly maintaining the reference travel time (+1.73%).The reinforcement learning agent demonstrated improved consistency and reduced deviation compared to the baseline runs without any external control, especially in high traffic volume scenarios.