Real-time re-optimization for generalized ridesharing feeder service with mixed scheduled and on-demand riders

Ridesharing revolutionizes feeder services by seamlessly connecting dispersed passengers to common destinations, such as train stations, bus terminals, airports, special event locations, and post-disaster shelters. This paper develops real-time re-optimization methodologies for generalized dynamic ridesharing feeder service operation. Such ridesharing feeder service typically encompasses a mix of scheduled and on-demand service, in which riders can schedule the request early in advance or send on-demand request for immediate service, and both types of riders are possible to share the ride. The developed methodologies enable the system to continuously re-optimize and re-adjust the vehicle routing plan that achieves the primary objective of maximizing the total number of served riders and the secondary objective of minimizing the total vehicle miles or hours traveled, simultaneously accounting for mixed types of riders' mobility requirements. This paper develops a re-optimization model implemented by a rolling horizon planning approach. An efficient heuristic algorithm, namely Large Neighborhood Search by Tabu Search algorithm (LNS-TS), is developed to solve large-scale problems. To validate the methodology, a simulation is developed to model rider activity, as well as the ridesharing process. This paper presents two case studies in Houston, Texas: a first-mile ridesharing service improving transit connectivity and a post-disaster cooling shelter access service addressing emergency transport under extreme heat. These cases represent two distinct operational contexts—everyday urban mobility and emergency disaster response—demonstrating the model's adaptability across diverse scenarios. The results generated by the re-optimization methodologies are compared with those from a periodic optimization approach, where matching and routing optimizations are conducted in isolation at different time intervals. The simulation results demonstrate that the routing plan obtained by the real-time re-optimization methodologies can serve more riders and save more vehicle miles or hours traveled compared with the periodic optimization methods. The proposed real-time, dynamic re-optimization approach for ridesharing feeder services not only demonstrates practical benefits but also holds promise for applications in autonomous vehicle systems.