A Virtual Value-guided Collaborative Method for Autonomous Taxi Route Planning and Charging

Autonomous electric taxi fleets possess inherent advantages in complete dispatch compliance and continuous operational efficiency, making their scheduling pivotal for enhancing urban transport systems and ensuring grid stability. Most existing studies, however, separately optimize order matching, idle cruising, and charging decisions, leading to short-sighted planning and suboptimal long-term profitability. To address this issue, a virtual-value-guided collaborative scheduling method is proposed. By quantifying the potential future value of charging (per kWh) and idle cruising (per km) as virtual values, a unified decision-making framework integrating order acceptance, cruising, and charging is established. Within this framework, a linear programming model determines real-time dispatching, while a reinforcement learning-based value function approximator evaluates the long-term return of decisions, enabling synergistic optimization of operational and charging strategies. Simulations based on real-world road networks and order data from Shenzhen demonstrate that the proposed method increases the order acceptance rate by 23.87% and improves fleet profit by 27.47% compared with baseline models, while maintaining low unit charging costs and passenger waiting time. Furthermore, the method exhibits strong scalability and adaptability under varying fleet sizes and charging station densities. By coordinating charging with operational decisions, this approach not only ensures reliable energy supply but also effectively controls charging costs, thereby enhancing overall operational efficiency and maximizing long-term fleet revenue.