A machine learning-based approach to assess impacts of autonomous vehicles on pavement roughness.

Abstract

Studies have been initiated to investigate the potential impact of connected and automated vehicles (CAVs) on transportation infrastructure. However, most existing research only focuses on the wandering patterns of CAVs. To bridge this gap, an apple-to-apple comparison is first performed to systematically reveal the behavioural differences between the human-driven vehicle (HDV) and CAV trajectory patterns for the first time, with the data collected fromthe camera-based next generation simulation dataset and autonomous driving co-simulation platform, CARLA and SUMO, respectively. A gradient boosting-based ensemble learning model for pavement performance (i.e. international roughness index) prediction is then developed with the input features including three driving pattern features, namely, lateral wandering deviation, longitudinal car-following distance and driving speed, plus 20 othercontext variables. A total of 1707 observations is extracted from the long-term pavement performance database for model training purposes. The result indicates that the trained model can accurately predict pavement deterioration and that CAV deteriorates pavement faster than HDV by 8.1% on average. According to the sensitivity analysis, CAV deployment will create a greater impact on the younger pavements, and the rate of pavement deterioration is found to be stable under light traffic, whereas it will increase under congested traffic. This article is part of the theme issue 'Artificial intelligence in failure analysis of transportation infrastructure and materials'.