Personalized Routing using Crowdsourced Connected Vehicle Data

This paper develops an algorithm for personalized route recommendations in traffic networks, using crowdsourced connected vehicle data. Current policies usually consider minimal travel and/or minimum energy paths for planning a route between a given origin-destination pair in a road network. However, individual driving preferences may involve a combination, in varying proportions, of the time and energy aspects while choosing a route, and may also depend on additional features such as the type of vehicle, amount of expected speed variations along the routes, turns, etc. These additional factors need to be considered to provide individualized route recommendations for different drivers. This paper uses individual driving histories to, i) create a generalized probabilistic model of the driver-specific features given certain macroscopic traffic conditions for each road segment between a chosen origin-destination pair, and, ii) learn a personal cost function based on the predicted features. The algorithm for recommending routes for different drivers is validated using Simulation of Urban MObility (SUMO)-based simulation of an urban road network.