On-demand high-capacity ride-sharing via dynamic trip-vehicle assignment with rebalancing

Abstract

On-demand ride-sharing systems with autonomous vehicles have the potential to enhance the efficiency and reliability of urban mobility. However, existing ride-sharing algorithms are unable to accommodate high capacity vehicles and do not incorporate future predicted demand. This thesis presents a real-time method for high-capacity ride-sharing that scales to a large number of passengers and trips, dynamically generates optimal routes with respect to online demand and vehicle locations, and incorporates predictions of anticipated requests to improve the performance of a network of taxis. We experimentally assess the trade off between fleet size, capacity, waiting time, travel delay, and amount of predictions for low to medium capacity vehicles. We validated the algorithm with over three million taxi rides from the New York City taxi dataset and demonstrate that our approach can service nearly 99% of Manhattan taxi demand using a fleet of only 3000 vehicles (less than 25% of the active taxis in Manhattan).