Transformer-based Sideslip Angle Estimation for Autonomous Vehicle Applications

This paper presents a preliminary study on sideslip angle estimation for autonomous vehicles (AVs) based on the Transformer architecture. Specifically, considering the decisive role of the dataset in the network's inference results, a high-coverage dataset is constructed from three inputs: vehicle speed, steering wheel angle, and road surface coefficient, which is used for offline training of the network. Additionally, the model-based Extended Kalman Filter (EKF) algorithm and data-driven Long Short-Term Memory (LSTM) network are designed as baseline methods for comparison. Co-simulation results under varying vehicle speeds demonstrate that the neural network-based approach achieves significantly higher accuracy in sideslip angle estimation compared to the model-based method. Furthermore, the Transformer-based approach exhibits lower mean absolute error (MAE) and root mean square error (RMSE) in sideslip angle estimation compared to LSTM. These preliminary results validate the effectiveness of using the Transformer for solving vehicle state estimation problems.