Towards Optimized Dynamic Ridesharing System Through Multi-Objective Reinforcement Learning

As cities grow, addressing traffic congestion, environmental impact, and transportation efficiency becomes crucial. Dynamic ridesharing is a new paradigm for smart cities that aims to reduce traffic congestion and emissions by dynamically forming passengers” trips at short notice. It involves matching drivers and passengers in real-time. The utilization of ridesharing is highly volatile and undergoes significant changes over time, making it important to model and forecast its potential ahead of time. This work presents a novel approach to enhance ridesharing systems within the context of smart cities, where an intelligent and adaptive transportation infrastructure is paramount. Unlike conventional ridesharing platforms that typically prioritize either a singular objective or multiple objectives that are aligned, our approach focuses on the optimization of conflicting objectives by leveraging Multi-Objective Reinforcement Learning (MORL) principles. Firstly, it targets minimizing passengers' waiting times, which is crucial for enhancing user experience and satisfaction. Secondly, it focuses on reducing drivers' detour times, thus optimizing their efficiency and reducing operational costs. Thirdly, the system aims to maximize vehicle utilization, ensuring that resources are utilized optimally. By leveraging MORL, the ridesharing platform adapts to real-time demands of the urban environment, accounting for spatio-temporal constraints and diverse user preferences. The results obtained from our experiments demonstrate the system's effectiveness, robustness, and adaptability to various constraints.