Policy-Based Reinforcement Learning for Intelligent and Sustainable Urban Mobility Systems: A Framework Aligned with SDG 11

Policy-Based Reinforcement Learning (PBRL) is a strong branch of reinforcement learning that aims at maximizing the output of decision-making policies in direct interaction with dynamic environments. PBRL, within the framework of urban movement, provides effective solutions to complex and adaptive transportation problems. Intelligent sustainable urban mobility systems, supported by PBRL, are directly aligned with the goals of Sustainable Development Goal 11 (SDG 11), which promotes inclusive, safe, resilient, and sustainable cities by enhancing real-time transportation decision-making. This paper addresses the use of PBRL in the context of intelligent sustainable urban mobility systems, which is consistent with one of the objectives of the United Nations Sustainable Development Goal 11. The study seeks to contrast and compare PBRL algorithms, namely, REINFORCE, Actor-Critic, Proximal Policy Optimization (PPO), and Trust Region Policy Optimization (TRPO), in a simulated urban mobility space. Based on performance measures such as expectations, sample efficiency, and convergence stability, the paper concludes that PPO and Actor-Critic approaches provide the most stable and robust results, balancing computational requirements and learning performance. TRPO demonstrates high reliability in terms of convergence, but its computational cost is high, whereas REINFORCE has been reported to exhibit high variance and low sample efficiency. The results emphasize the propensity of policy-based techniques to benefit intelligent transportation systems like traffic light control and vehicle assignment. This study reflects on the development of AI that contributes to urban sustainability because it can inform practitioners in selecting suitable RL frameworks for performing socially significant, real-time decision-making.