Application of neural networks to design the controller for autonomous vehicles by learning driver's behavior

This paper presents an application of NNs to autonomous vehicles. The method is a result of linear quadratic regulator (LQR) used to design the controller for autonomous vehicles in the steady state. The NN structure is designed with the reference of the driver model. Two single neurons perform as the path filter where the previewed path is weighted. The vehicle model with nonlinear dynamics represents a real vehicle. The NN weights adapt to various road shapes. Finally, the trained NN are tested through the F1 Montreal circuit in order to evaluate the vehicle performance in navigation and speed control.