Conference Agenda

We consider a mixed-operation ride-hailing system that offers customers the option to request a ride on-demand or to pre-book it in advance. For time-sensitive customers, pre-booking provides a service guarantee at a price premium. From the operator’s perspective, pre-booking allows for planning ahead with higher certainty but may incur the duty to operate unfavorable trips that may even induce a shift in the demand distribution, e.g., in low-demand suburban neighborhoods. So far, the literature on ride-hailing systems mainly focused on the pure on-demand setting, and only a few works investigated the effects of pre-booking, assuming pre-booked rides follow the same distribution as on-demand rides. Against this background, we develop an optimization framework for a mixed-operation ride-hailing system, allowing us to study the trade-offs between higher planning certainty and the rise of unfavorable rides due to shifts in the demand distribution. Accordingly, we propose a two-stage stochastic optimization formulation in which the first-stage problem consists of deciding which pre-booking requests to accept, while the second-stage problem involves assigning vehicles to requests and planning routes in the face of uncertain on-demand requests. We introduce a solution algorithm that combines Bender’s decomposition and column generation. We conduct experiments based on the New York City network using historical yellow taxi trip data. We show that greedily accepting all pre-booking requests decreases the operator’s profit and increases dead mileage compared to the pure on-demand baseline. In contrast, optimal solutions to our formulation lead to increased profit and higher fleet utilization while maintaining customer service quality.

We consider the problem of many on-demand transportation start-ups: how to establish themselves in a new market. When starting, such companies often have limited fleet resources to serve demand in a city. Dependent on the use of the fleet, different service quality is observed in different areas of the city. The service quality impacts the respective growth of demand in each area. Thus, operational fulfillment decisions drive the longer-term demand development. To integrate strategical demand development in the real-time fulfillment operations, we propose two steps. First, we derive analytical insights in the optimal allocation decisions for a stylized problem. Second, we use the insights to shape the training data of a reinforcement learning strategy for operational real-time fulfillment. We show that the combined consideration of real-time effectiveness and long-term strategy is very beneficial. We further show that the careful shaping of training data is essential for successful development of demand.

Due to the required reduction of emissions, modern mobility concepts are rapidly evolving. Ridepooling is one of these concepts. Beside the reduction of emissions due to electric vehicles, ridepooling services promise to reduce traffic due to pooling and to increase mobility access especially in suburban areas. In practice, ridepooling services receive customer orders dynamically and thus have to integrate them in the vehicles’ tours. In this talk, we discuss an efficient procedure to insert new customer requests into given tours while incorporating possible future customers with the objective to serve as many customer requests as possible over the time horizon.