Autonomous electric vehicles are set to replace most conventional vehicles in the near future. Extensive research is being done to improve efficiency at the individual and fleet level. There is much potential benefit in optimizing the deployment and rebalancing of Autonomous Electric Taxi Fleets (AETF) in cities with dynamic demand and limited charging infrastructure. We propose a Fleet Management System with an Online Optimization Model to assign idle taxis to either a region or a charging station considering the current demand and charging station availability. Our system uses real-time information such as demand in regions, taxi locations and state of charge (SoC), and charging station availability to make optimal decisions in satisfying the dynamic demand considering the range-based constraints of electric taxis. We integrate our Fleet Management System with MATSim, an agent-based transport simulator, to simulate taxis serving real on-demand requests extracted from the San Francisco taxi mobility dataset. We found our system to be effective in rebalancing and ensuring efficient taxi operation by assigning them to charging stations when depleted. We evaluate this system using different performance metrics such as passenger waiting time, fleet efficiency (taxi empty driving time) and charging station utilization by varying initial SoC of taxis, frequency of optimization and charging station capacity and power.

Autonomous electric vehicles are set to replace most conventional vehicles in the near future. Extensive research is being done to improve efficiency at the individual and fleet level. There is much potential benefit in optimizing the deployment and rebalancing of Autonomous Electric Taxi Fleets (AETF) in cities with dynamic demand and limited charging infrastructure. We propose a Fleet Management System with an Online Optimization Model to assign idle taxis to either a region or a charging station considering the current demand and charging station availability. Our system uses real-time information such as demand in regions, taxi locations and state of charge (SoC), and charging station availability to make optimal decisions in satisfying the dynamic demand considering the range-based constraints of electric taxis. We integrate our Fleet Management System with MATSim, an agent-based transport simulator, to simulate taxis serving real on-demand requests extracted from the San Francisco taxi mobility dataset. We found our system to be effective in rebalancing and ensuring efficient taxi operation by assigning them to charging stations when depleted. We evaluate this system using different performance metrics such as passenger waiting time, fleet efficiency (taxi empty driving time) and charging station utilization by varying initial SoC of taxis, frequency of optimization and charging station capacity and power.