Source code for the publication: Reinforcement Learning with Model Predictive Control for Highway Ramp Metering

Source code for the implementation and simulation of a learning-based ramp metering control strategy with the goal of improving highway traffic flow management, where the proposed solution embeds model-based Reinforcement Learning methodologies in a Model Predictive Control framework, thus enabling the adaptation of the controller in order to improve automatically its performance based solely on observed closed-loop data. Simulations on a highway network benchmark demonstrate significant reduction in congestion and improved constraint satisfaction compared to an imprecise, non-learning initial controller, showcasing the efficacy of the proposed methodology.