Detection of Driving Dynamics Anomalies Using Deep Learning

The push towards automated and autonomous driving is enabled in part by the recent advances in machine learning. Various algorithms are entrusted with perception, planning and control of vehicles to perform the desired driving task. However, monitoring the state of the vehicle, i.e. the driving dynamics, is mandatory in such a scenario. To improve accuracy and reduce computation effort, a deep learning approach is chosen to model the driving dynamics of a vehicle using the Stuttgart Handling Roadway test bench. Different rear wheel steering controls are used to realize subtle differences in the vehicle dynamics. The resulting data is used to train different neural networks that are capable of predicting the driving dynamics of each configuration. It is shown, that the neural networks are able to differentiate between the different rear wheel steering controls.