Real-Time Distributed Taxi Ride Sharing

Taxicabs play an important role in urban public transportation. Analyzing taxi traffic of Shanghai, San Francisco, and New York City, we have found that the short trips within city are mostly of commuters during office hours and span a specific city area. Now, if the large number of commuters are ready to share their rides, that will have a huge impact on the 'super-commute' problem faced in various cities of USA and around the world. While ride-sharing can increase taxi occupancy and profit for drivers and savings for passengers, it reduces the overall on-road traffic and thereby the average commute time and carbon foot-print. While centralized ride-sharing services, like car-pooling, can address the problem to some extent, they lack scalability and power to dynamically adapt the taxi schedule for best results. In this paper, we propose a four-way model for the ride-sharing problem and develop a novel distributed taxi ride sharing (TRS) algorithm to address dynamic scheduling of ride sharing requests. Our algorithm shows the overall reduction in total distance travelled by taxis as a result of ride sharing. Empirical results using large scale taxi GPS traces from Shanghai, China show that TRS algorithm can grossly outperform a Taxi Distance Minimization (TDM) algorithm. TRS accommodates 33% higher ride share among passengers while dealing with 44,241 requests handled by 4,000 taxis on a single day in Shanghai.