Model-based vs Data-driven Estimation of Vehicle Sideslip Angle and Benefits of Tyre Force Measurements

TL;DR

This paper compares model-based and data-driven methods for vehicle sideslip angle estimation, highlighting the advantages of incorporating tyre force measurements and demonstrating superior accuracy of data-driven approaches on extensive experimental data.

Contribution

It introduces a comprehensive comparison between model-based and data-driven estimation methods, emphasizing the benefits of using tyre force measurements for improved accuracy.

Findings

Data-driven approaches outperform model-based methods in accuracy.

Tyre force measurements significantly enhance estimation performance.

Both approaches benefit from tyre force data, especially in non-linear tyre regimes.

Abstract

This paper provides a comprehensive comparison of model-based and data-driven approaches and analyses the benefits of using measured tyre forces for vehicle sideslip angle estimation. The model-based approaches are based on an extended Kalman filter and an unscented Kalman filter, in which the measured tyre forces are utilised in the observation model. An adaptive covariance matrix is introduced to minimise the tyre model mismatch during evasive manoeuvres. For data-driven approaches, feed forward and recurrent neural networks are evaluated. Both approaches use the standard inertial measurement unit and the tyre force measurements as inputs. Using the large-scale experimental dataset of 216 manoeuvres, we demonstrate a significant improvement in accuracy using data-driven vs. model-based approaches. Tyre force measurements improve the performance of both model-based and data-driven approaches, especially in the non-linear regime of tyres.