Park, Hyun Su (University of Seoul), Shin, DongHwa (Kwangwoon University), Park, Shin Hyoung (University of Seoul), Cho, Shin-Hyung (University of Seoul)

Iterative Decision-Making in Reinforcement Learning for Ride-Sharing Optimization: A Joint Approach to Dispatching and Rebalancing

Scheduled for presentation during the Invited Session "S19c-Artificial Transportation Systems and Simulation" (TH-LA-T19), Thursday, November 20, 2025, 16:40−17:00, Surfers Paradise 1

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

This study proposes a reinforcement learning (RL)-based decision-making framework to optimize operations in ride-sharing (RS) systems. The framework jointly addresses two key challenges: passenger–vehicle dispatching and idle vehicle rebalancing under dynamically evolving spatial demand. A Deep Q-Network (DQN) algorithm is applied within a centralized architecture, enhanced by an iterative Q-value-based action assignment strategy that accounts for the interdependencies between vehicles. To mitigate computational complexity, an action space reduction mechanism is introduced using feasibility-based masking. The reward function integrates operational efficiency, service quality, rebalancing, and environmental sustainability to guide multi-objective learning. Experimental evaluations under four distinct spatial demand scenarios reveal that the proposed RL framework improves both system throughput and passenger experience compared to baseline settings. Furthermore, comparative analysis with and without rebalancing confirms its effectiveness in reducing detour time and improving request acceptance rates. Finally, single- and multi-vehicle experiments demonstrate the scalability of the framework, showing robust learning convergence and policy generalization in increasingly complex environments. These findings highlight the practical viability of the proposed approach for real-world RS system deployment, particularly in urban settings characterized by spatially asymmetric demand and limited supply.