Skip to main content Skip to main navigation


Learned steering feel by a neural network for a steer-by-wire system

Patrick Krupka; Paul Lukowicz; Christopher Kreis; Bastian Boßdorf-Zimmer
In: Peter E. Pfeffer (Hrsg.). 10th International Munich Chassis Symposium 2019. International Munich Chassis Symposium, Pages 449-464, Proceedings (PROCEE), ISBN 978-3-658-26434-5, Springer, 11/2019.


High available steering systems for autonomous driving enable the development of Steer-by-Wire systems. Due to the missing mechanical linkage the steering wheel torque which depends on several inputs with different influences needs to be artificially generated by a Force-Feedback-Actuator. The present publication describes how machine learning methods and especially artificial neural networks can be used to provide a steering feel for a Steer-by-Wire system. Therefore training data consisting of synthetic driving maneuvers is recorded to train a feedforward neural network. Measurement signals of the driver input and the vehicle reaction were selected to be used as inputs for estimating the steering wheel torque as the output of the model. Networks of different sizes are trained and evaluated on the basis of their training and test error to examine how complex the model must be to calculate the output sufficiently. To extract more information from the training data sliding window features are used in addition to the current signal values. A trained network has been integrated into the software of a Steer-by-Wire system in a prototype vehicle to provide the steering wheel torque to the Force-Feedback-Actuator. In this vehicle the steering feel generated by the model could be subjectively evaluated on a test site.

Weitere Links