A Novel Ride Discovery Scheme for Ride Sharing Systems: a DQN Approach

Ride-sharing systems are a cornerstone of modern urban mobility, addressing transportation demands through efficient resource utilization and real-time decision-making. This paper introduces a distributed grid-based ride discovery framework that employs a localized proxy vehicle model which enhances taxi assignment and routing efficiency. Each grid in the geographic region hosts a unique deep reinforcement learning model which is downloaded by taxis upon entry. The proxy vehicle model enables the vehicles to make dynamic adjustments to their paths while taking the inconvenience to the passengers into account. The proxy vehicle model depends on the number of taxis in the grid, pending requests of the neighbouring grids, passenger occupancy and their respective delay tolerances. The objective of the proxy vehicle model is to maximize the total expected reward of any taxi that uses the proxy vehicle model. The proxy vehicle model is implemented using the deep Q network which is trained jointly by the experiences of all the vehicles across the grids. The proposed work is compared to some benchmark ride discovery models. Performance analysis reveals better passenger occupancy and lower passenger inconvenience under the proposed work compared to the benchmark models.