Road Friction Estimation Using an Artificial Neural Network in a Simulated Environment

Abstract:

With the transition of responsibilities from the driver to the automated driving systems in vehicles, the systems need to have been tested for an extensive list of test scenarios as the passengers require high trustworthiness.The friction coefficient for the tyre-road friction is of high importance for the control of the vehicle but the coefficient is dependent on the physically complexity and nonlinear behaviour of tyres and is difficult to measure. Hence, testing is performed in controlled environments which limits the systems exposure to different testing scenarios. The purpose of this thesis and the underlying work was to develop and evaluate a process for friction estimation using machine learning. The aim was to produce an estimation method using neural networks that are trained on data from a vehicle model implemented in a simulated environment using Unity 3D. The master thesis was produced at Combine Control Systems AB for Lund University in cooperation with National Electric Vehicle Sweden AB (NEVS).