Smart Fleet Management for Shared Micro-mobility: Balanced demand, Redistribution and Charging via Deep Reinforcement Learning

Shared electric micro-mobility, as an emerging mode of urban transportation, has been booming worldwide in recent years. Although it provides sustainable, eco-friendly, and cost-effective mobility, it also faces several challenges, particularly due to existing inefficient feet management strategies. These typically rely on fixed redistribution schedules that fail to adapt to highly dynamic user demand and overlook practical factors such as time-varying patterns and charging requirements. To address this problem, this paper presents a user-involved deep reinforcement learning framework for real-time feet management of shared electric micro-mobility, utilizing the Soft Actor-Critic network. It ensures system balancing through dynamic redistribution, taking into account charging constraints while reducing costs by aggregating discharged micro-vehicles. Our framework is demand-aware, incorporating predicted user demand through online learning using an LSTM network, thereby ensuring effective system balancing that meets user needs. Experimental results demonstrate the effectiveness of the proposed framework, showing robust performance for real-time applications.