An Intelligent Framework to Maximize Individual Taxi Driver Income

Abstract

The ridesharing platform has significantly changed how taxis operate in recent years. Most previous works focus on improving the user experience and maximizing the revenue from the platform or system level. The individual driver benefits are rarely addressed. In this work, we propose a deep reinforcement learning-based framework to help the individual driver maximize their daily income via order selections and self-repositioning. We first formulated the taxi operation as a Markov Decision Process. Then we created a multi-agent simulation consisting of the taxi drivers that use different strategies. A deep Q network-based (DQN) framework is proposed for drivers to learn which orders to select and where to reposition. Our result shows the driver who adopts DQN framework outperforms all other drivers. Furthermore, we found that the optimal policy does not suggest the driver operating in particular areas but recommends selecting the order with $5 to $7.5 taxi fare.