Traffic lights play a crucial role in managing traffic flow, ensuring safety and efficient transportation systems. Traditional fixed-time traffic light control systems have been widely used for decades, but they have limitations in adapting to changing traffic patterns. This can lead to increased waiting times, congestion, and inefficient use of road networks, resulting in reduced overall transportation efficiency. To address these issues, researchers have proposed dynamic traffic light control systems based on reinforcement learning (RL) algorithms. These systems have shown promise in learning and adapting traffic light timings based on real-time traffic conditions, leading to improved traffic flow and reduced waiting times. Several RL algorithms have been proposed for traffic light control, including SARSA, Q-learning (QL), and Deep Q-network (DQN). These algorithms use different methods to learn and adjust traffic light timings. This work compares the performance of three different RL algorithms (QL, Sarsa and DQN) over traffic light control and evaluates their effectiveness in reducing waiting times and congestion. Our results demonstrate that RL-based traffic light control systems outperform traditional fixed-time systems in managing traffic flow and reducing waiting times. Specifically, our study shows that the proposed RL-based system, based on Deep Q-network, performs better than traditional fixed-timing systems and the other RL algorithms.
Traffic Lights Control Using Reinforcement Learning: A Comparative Study
Lect. Notes in Networks, Syst.
The International Conference on Artificial Intelligence and Smart Environment ; 2023 ; Errachidia, Morocco November 23, 2023 - November 25, 2023
2024-01-30
6 pages
Article/Chapter (Book)
Electronic Resource
English
European Patent Office | 2023
|Engineering Index Backfile | 1930
|European Patent Office | 2020
|