A Lateral Control Based on Physics Informed Neural Networks for Autonomous Vehicles
摘要
In the paper, a lateral control strategy is presented using Physics-Informed Neural Network (PINN) for automated vehicles. The main idea is that the physics information is incorporated into the training process, which leads to an improvement in the performance level of the control algorithm. Moreover, in the highly nonlinear range of the lateral dynamics, which is not properly covered by the training dataset, the stability of the vehicle is guaranteed. The results are compared to a conventional neural network trained to control the vehicle.