Model Predictive Control With Learned Vehicle Dynamics for Autonomous Vehicle Path Tracking

Model Predictive Controller (MPC) is a capable technique for designing Path Tracking (PTC) of Autonomous Vehicles (AVs). The performance MPC can be significantly enhanced by adopting high-fidelity and accurate vehicle model. This model should capturing the full dynamics vehicle, including nonlinearities uncertainties, without imposing high computational cost MPC. A data-driven approach realised learning using operation data offer promising solution providing suitable trade-off between state predictions work proposes framework an with Neural Network (NN)-based learned dynamic plethora available from modern systems. objective to integrate NN-based higher accuracy than conventional models required prediction horizon into improved tracking performances. proposed highly approximating latent system states, which are difficult estimate, provides more in presence parametric uncertainties. observation results various road conditions shows that outperforms MPCs models.