Optimizing Passenger On-Vehicle Experience through Simulation and Multi-Agent Multi-Criteria Mobility Planning

The rapid growth in urban population poses significant challenges to moving city dwellers in a fast and convenient manner. This thesis contributes to solving the challenges from the viewpoint of public transit passengers by improving their on-vehicle experience. Traditional transportation
research focuses on pursuing minimal travel time of vehicles on the road network, paying no attention to people inside the vehicles. In contrast, the research in this thesis is passenger-driven, concerning the role of the on-vehicle experience in mobility planning through the public transit
systems. The primary goal of the thesis is to address the following problem: Given an urban public transit network, how can we plan for the optimal experience of passengers in terms of their service preference? There are several challenges we have to address to meet this goal. First, a model or a simulator that captures not only the road traffic, but also the behaviors of passengers and other relevant
factors is a prerequisite for this research but has seldom been developed previously. Second, to plan for passengers’ mobility concerning the influence among passengers as well as multiple service preferences is computationally intensive, especially on a city scale. To achieve the research goal and overcome the challenges, this thesis develops a joint traffic passenger simulator, which simulates the road traffic, behaviors of passengers and on-vehicle environment dynamics. Specifically, the simulator combines the urban road traffic, the interactions among the passengers and the infrastructures that support certain on-vehicle services,
such as on-vehicle Wi-Fi, to provide a passenger-level simulation. A separate passenger behavior model and on-vehicle Wi-Fi service model are designed to run jointly with SUMO, a mature traffic simulator, for simulating the passenger behaviors and on-vehicle travel experience. A
joint simulator for the bus transit system in the city of Porto, Portugal has been implemented and tested by comparing the simulation to the real passenger data. To configure the background passenger flow in the simulation, real passenger data are used. The data were collected by an entry-only system and the destination information was missing.
This thesis contributes a machine learning algorithm, called semi-supervised self-training, to infer the missing destinations with a high inference confidence level.
Given the simulation platform, the passenger mobility planning problem can be formalized as a multi-agent path planning (MAPP) problem, where multiple passengers may interfere with each other when contending for service resources. The mobility planning operates on the client passengers (i.e., a subset of the overall passengers who request the planning service from our planner). State-of-the-art MAPP solvers, such as M*, do not scale well to such a MAPP problem. This thesis proposes the soft-collision-free M* (SC-M*), a generalized version of M*, to efficiently
handle the MAPP task under complex urban environments (i.e., with a large client passenger size and multiple types of client passengers requesting multiple types of service resources). We evaluate the performance of the SC-M* through a case study of the bus transit system in Porto,
Portugal and the experimental results show the advantages of the SC-M* in terms of path cost, collision-free constraint, and the scalability in run time and success rate.