On non-myopic internal transfers in large-scale ride-pooling systems

With advancements in technology, digital ride-pooling has become a popular on-demand shared mobility service. However, maintaining system efficiency while achieving high ridership remains a challenging task in such services. Allowing passengers to be transferred between vehicles within the system is a potential way to improve service availability for customers and increase fleet efficiency. This study proposes a non-myopic state-of-the-art dynamic routing algorithm for operating a ride-pooling service with synchronized so-called intramodal transfers. A non-myopic cost function approximation policy is proposed to control for the opportunity cost of transfer decisions in offering a route with or without transfer to a passenger. The policy approximates this cost with the “time until the potential transfer” as online transfers without such consideration are shown to significantly underperform, while the approximation makes transfers at least as competitive as non-myopic ride-pooling without transfers. The policy is tested in a newly developed simulation platform to investigate various scenarios. Experimental results on real-world trip data in Hamburg, Germany, demonstrate the importance of taking a non-myopic approach in making sequential decisions. Although allowing transfers can potentially improve vehicle occupancy, it can be challenging to be implemented in practice, given the added uncertainties in real-world operations.