Reinforcement Learning-Based Freeway traffic Control Concerning Emissions

This study presents a reinforcement learning based framework involving the integrated use of ramp metering (RM) and variable speed limit (VSL) control towards the ultimate aim of mitigating traffic congestion and emissions. Traditional freeway traffic control strategies often fail to adapt dynamically to evolving traffic conditions, resulting in suboptimal performance. The proposed framework seeks, through simulation, the optimal setting of VSL and RM actions by leveraging RL. The learning-based architecture we have designed is trained and tested using data from a hypothetical freeway network piece and synthetic demand profiles. The performance of the framework is evaluated by considering multiple traffic demand levels and connected and automated vehicle penetration rates.