Tire—Road Friction Characteristics Estimation Based on ESO Observer and LSTM Neural Networks

To enhance the perception of road surface information during autonomous driving, this paper proposes a Tire-Road Friction Characteristics (TRFC) estimation algorithm that integrates a 7-DOF vehicle dynamics model, the Magic Formula, an improved adaptive Extended State Observer (ESO), and Long Short-Term Memory (LSTM) neural networks. The vehicle's key state parameters are obtained through the dynamics model, while the tire forces and the ESO observer provide accurate estimates of the utilized TRFC and slip ratio. Furthermore, an improved LSTM model, incorporating the Physics-Informed Neural Network (PINN) approach based on physical constraints, is employed to estimate the TRFC and critical vehicle state parameters. The algorithm's reliability and robustness are validated through MATLAB/Simulink and CarSim co-simulation experiments. The results demonstrate that the proposed method maintains high accuracy even under highly dynamic and uncertain conditions. This approach can serve as a complementary technique for fusion with vision and LiDAR-based TRFC identification methods, providing theoretical support for improving the safety and stability of autonomous vehicles.