Dilated Convolutional Neural Network-Based Modeling and Tracking Control Design for Intelligent Vehicles

This paper proposes a methodology for modeling an intelligent vehicle system utilizing dilated convolutional neural networks. This approach is then employed in designing a lateral tracking controller for vehicles to enhance both the accuracy of the vehicle model and the effectiveness of path tracking control. The dilated convolutional feedforward controller (DCFFC) for the vehicle is obtained by solving the nonlinear optimization problem of the designed dilated convolutional vehicle system model (DCVSM) online. Then, a lateral feedback controller (LFBC) based on the steady-state vehicle sideslip angle design is integrated to achieve precise vehicle tracking along the reference path Simulation results indicate that the optimization efficiency of DCVSM ranges from 86.26% to 99.76% compared to other vehicle models cited in this paper, markedly enhancing the predictive accuracy of the vehicle model. Likewise, the combined DCFFC+LFBC exhibits superior control performance over other controllers mentioned herein, manifesting in lower lateral error peaks, superior tracking results, and increased stability in various simulated environments. Copyright (c) 2024 The Authors.