One of the biggest challenges in operating on-demand mobility services is the need to dynamically reposition idle vehicles, whether they are taxis, shared vehicles/bikes, or empty shuttles. This latest research with Dr. Hamid R. Sayarshad at Cornell University proposes new models and algorithms to anticipate future demand for the problem by approximating future opportunity costs with queue delay. In addition, we formulated a lower bound of the queueing-based location model from Marianov & Serra that can be solved much more computationally efficiently. Simulation tests in a controlled study area with NYC taxi data suggests the feasibility of nearly 30% improvement over myopic positioning techniques that do not use data to look ahead.
This work was initially undertaken when Hamid was a PhD student with funding support from the Canada Research Chairs program. Resources from C2SMART are also acknowledged.