Drivers’ Attitudes Toward Rerouting
Drivers’ Attitudes Toward Rerouting: Impacts on Network Congestion
Principal Investigator: Jorge A. Laval, Professor, School of Civil and Environmental Engineering
Project Duration: 12 months
Project Budget (Federal UTC Funds): $114,815
Project Budget (Cost-share): $80,006
Institution: Georgia Institute of Technology
Abstract
This project aims to answer the following questions: (1) What machine learning (ML) approaches are useful to help people make rerouting decisions in congestion? (2) What properties of congested urban networks influence the ML result? (3) How do the rerouting decisions influence the bifurcation phenomena in macroscopic fundamental diagrams (MFDs)? Deep reinforcement learning (DRL), one of the advanced RL methods, is identified to analyze rerouting behavior. In congested urban networks, some factors are found to have a huge impact on the DRL result. First, the density of the network influences the performance of DRL significantly. Especially when the network is close to jam density, DRL cannot learn in such situations. Second, the DRL result is affected by the driver’s rerouting probabilities, or rerouting intention. Third, the blocking area distribution is an important value for the demand matrix because it is the reward in each iteration. The goal of this project is to distribute the blocking area evenly across the network. Because the project’s focus is on the impact of adaptive driving behavior, a more realistic network is built in SUMO to perform simulations. This helps assess how each of the above factors affects the results and whether there are any other factors that influence the results.