PTLC: Personalized Traffic Light Control Based on Deep Reinforcement Learning Approach

Traffic signal control is a critical aspect of improving urban traffic efficiency and reducing road accidents. Reinforcement learning (RL) has been applied to solve the intersection traffic control problem, which is flexible and can be adapted to dynamic traffic conditions. However, most existing RL methods only consider a single vehicle type, ignoring the impact of multiple vehicle types on traffic and safety. To address this gap, we propose a new RL method called pTLC, which aims to allow trucks to cross intersections quickly while minimizing the impact on other vehicles. In this method, vehicle type is added as a state variable to capture the influence of different vehicle types on traffic flow. At the same time, a new state representation method is proposed that divides the road into several road sections to balance the environmental information and the size of the state space. We design simulation experiments under different traffic demand scenarios, including cars and trucks, to verify the effectiveness of the method. The results show significant improvements in traffic efficiency and safety, especially for mixed vehicle types.