Understanding and Optimizing Ride-Sourcing Drivers' Learning Dynamics

点击翻译按钮获取中文摘要

This project will generate behavioral insights and algorithmic tools for ride-sourcing drivers to enable effective driver organization and eventual market efficiency and enhanced societal welfare. It will work with drivers to obtain location and operational data and generate optimized and coordinated guidance on tactical and operational decisions to maximize their welfare. Transportation agencies and local communities could partner with drivers to gain insights into the market and optimize policies to meet societal goals. The independence of ride-sourcing drivers comes at the price of social isolation and anxiety. This project will chart a technology-inspired pathway to better connection and organization of the diffusive workforce, and increase its cohesion, vitality, and social contribution. Ride-sourcing drivers are more likely from the lower income and other vulnerable parts of the population, and the project prototype serves as an outreach tool to improve their well-being, in addition to being a test bed for the research program. The PI will incorporate the research results in her graduate/upper-class undergraduate courses on transportation systems analysis and economics, where traditionally the supplier side of the transportation market is not treated in the same depth as the consumer side. There are three major research objectives: 1) Understand drivers' learning and choice behaviors using dynamic discrete choice models grounded on psychologically sound learning theories; 2) Develop model-based and model-free algorithms to optimize decisions on when to start and end working, where to search for passengers and whether to accept a ride request, scalable to the level of driver participation; 3) Generate behaviorally informed driver guidance synthesizing results from previous two objectives. The project is innovative in three major aspects. First, it provides an alternative approach to enhancing the ride-sourcing market by directly working with drivers instead of platforms as commonly done in the literature and practice. The driver-centered approach is potentially more socially efficient due to the avoidance of a platform's tendency for over-supply and fairer due to its collaborative nature. Secondly, it advances understandings of drivers' learning and choice making dynamics under uncertainty at various temporal scales: routing and order acceptance at the minute-to-minute level, and scheduling at the hour-to-hour level. This integrated approach will fill the knowledge gap of reasons and dynamics behind the low retention rate of ride-sourcing drivers. Thirdly, it contributes to the development of high-performance optimization algorithms for ride-sourcing operations with an emphasis on the scalability to the level of driver participation and data availability.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.