A Congestion Pricing Design with Shared Autonomous Mobility using Deep Reinforcement Learning

Abstract

Shared autonomous mobility (SAM) may exacerbate congestion by increasing traveled distances, but dedicated congestion management strategies remain relatively underexplored. This study proposes a deep reinforcement learning framework to dynamically adjust congestion charges under SAM. To represent mixed traffic conditions (privately owned and shared autonomous vehicles), we integrate a multi-region Macroscopic Fundamental Diagram with a spatial dial-a-ride model. Building on this environment, a double deep Q-learning algorithm is applied to optimize congestion charges for a fixed cordon. The results show that the proposed framework can efficiently manage congestion. Furthermore, application results indicate that the introduction of SAM leads to higher congestion, as it increases vehicle total traveled distance.