Realistic urban traffic simulation with ride-hailing services: a revisit to network kernel density estimation (systems paper).

App-based ride-hailing services, such as Uber and Lyft, have become popular thanks to technology advancements including smartphones and 4G/5G network. However, little is known about to what degree their operations impact urban traffic since Transportation Network Companies seldom share their ride data due to business and user privacy reasons. Recently, transportation engineering researchers began to collect data in large cities trying to understand the transportation impacts of ride-hailing services, so as to assist transport planning and policy making. However, (1) there does not exist a general data collection approach applicable to any city, and (2) the studies were based on historical data and cannot project the future easily even though ride-hailing services are developing quickly. In this paper, we introduce our approach to building a digital twin of the transportation in the medium-sized city of Birmingham, Alabama. This digital twin is a transportation simulation model that incorporates transportation modes such as public transits and ride-hailing services, in addition to private vehicles that constitute the majority of Birmingham's traffic. With this digital twin, transportation engineers can flexibly analyze the impact of ride-hailing services under different scenarios, such as "if the number of Uber drivers doubles" which could happen in the near future. The digital twin may also enable new opportunities, such as being an environment for learning policies with reinforcement learning. To enable realistic transportation simulation, we propose a novel approach to collect Uber ride data that is easy to carry out in any city. This approach collects app screenshots about Uber rides from Uber drivers, and uses crowdsourcing to postprocess these screenshots to extract detailed ride information. We then fit a spatiotemporal distribution of Uber rides using the collected data via network kernel density estimation (KDE). The existing network KDE method is flawed in that the contributions of different data samples are not the same, so we propose a new formulation to solve this issue. The distribution combined with population statistics from census data enable the generation of realistic Uber rides for agent-based simulation. Our project is shared at https://github.com/jalal1/UberSim.