Data-driven vehicle modeling of longitudinal dynamics based on a multibody model and deep neural networks

The vehicle dynamics simulation and preview control require a dedicated vehicle model, such as multibody dynamics model. However, the multibody model has higher computational complexity which affects the response time of the vehicle controller. This issue can be alleviated by using a data-driven vehicle dynamics model due to its effective generalization and computational speed. In this work, we propose a data-driven modeling approach based on deep neural networks (DNNs) for computing and predicting the vehicle characteristics. The high-fidelity simulations of a validated vehicle multibody model are performed for data acquisition. This data is then used for training and testing the proposed model. The DNN inputs comprise the initial speed of the vehicle and the torque applied on front wheels to imitate vehicle acceleration and deceleration. The DNN outputs comprise the driving distance and the longitudinal velocity of the vehicle. The dynamics characteristics resulting from both the data-driven model and the multibody model are investigated and compared. Furthermore, the accuracy of the data-driven model is analyzed in terms of various error functions. The data-driven model is verified by using the results obtained from a commercial software package. The simulation results show that the data-driven vehicle model predicts the accurate velocity and driving distance in real-time. The data-driven model can be used for real-time simulation and preview control in autonomous vehicles.