On route peer-to-peer dynamic ridesharing system: an agent-based modeling approach

Abstract

Over the past decades, the urban mobility sector has explored ridesharing systems (RSs) as sustainable solutions to reduce private car ownership and improve mobility for transit-dependent users. However, simultaneously optimizing both matching of drivers and passengers and routing given the dynamic and real-time complexity of urban environments is still a challenge to overcome. This research presents a novel framework that combines Agent-based Modeling (ABM) with Reinforcement Learning (RL) to address this joint optimization problem. The ABM component simulates interactions among individual agents—drivers and passengers—capturing the spatial and temporal complexities inherent in urban mobility patterns. Coupled with an RL algorithm, the framework learns adaptive policies that optimize driver-passenger assignments and routing decisions in a realistic, simulated environment. By jointly addressing matching and routing, the model can reduce idle time, improve driver-passenger proximity, and dynamically adjust routes based on traffic conditions and demand fluctuations. Furthermore, the model provides insights to inform government policies and business models to support sustainable urban transit solutions.