Cheng, Hanyu (Tsinghua University), Tian, Qingyun (Nanyang Technological University), Wang, Zhiwei (Nanyang Technological University)

Adding Runs of Modular Autonomous Vehicles with Skip-Stop Strategy Using Deep Reinforcement Learning

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

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 April 2, 2026

Abstract

Coordinating modular autonomous vehicles (MAVs) with regular buses offers opportunities to enhance the flexibility and efficiency of urban transit services. However, most studies on MAVs do not consider such an integrated operation, primarily because of the difficulty in handling the resulting complexity and dynamics. Methodologically, traditional optimization methods (e.g., rolling horizon heuristics) lack long-term optimization and adaptability to demand fluctuations, while deep reinforcement learning (DRL) remains underutilized for such hybrid systems. To bridge these gaps, we propose an integer programming model to capture the complexity of coordinated MAVs and regular buses, with a deep reinforcement learning framework integrating Soft Actor-Critic (SAC) with for high-level scheduling. SAC determines key parameters including dispatched MAV formations and headway, which are then fed into a refined integer programming model to obtain detailed optimized skip-stop operation strategy. Experiments show that our method outperforms heuristic baselines in cost efficiency. Additionally, the incorporation of skip-stop operation can further improve the system performance. This framework demonstrates the potential of combining domain-specific modeling with modern DRL techniques for scalable and adaptive transit design.