Levenberg-Marquardt optimised neural networks for trajectory tracking of autonomous ground vehicles

Abstract

Trajectory tracking is an essential capability of robotics operation in industrial automation. In this article, an artificial neural controller is proposed to tackle trajectory-tracking problem of an autonomous ground vehicle (AGV). The controller is implemented based on fractional order proportional integral derivative (FOPID) control that was already designed in an earlier work. A non-holonomic model type of AGV is analysed and presented. The model includes the kinematic, dynamic characteristics and the actuation system of the AGV. The artificial neural controller consists of two artificial neural networks (ANNs) that are designed to control the inputs of the AGV. In order to train the two artificial neural networks, Levenberg-Marquardt (LM) algorithm was used to obtain the parameters of the ANNs. The validation of the proposed controller has been verified through a given reference trajectory. The obtained results show a considerable improvement in term of minimising trajectory tracking error over the FOPID controller.