Overview of the application of neural networks in the motion control of unmanned vehicles

This paper aims to introduce the application of neural networks in the motion control of unmanned vehicles in recent years. With the breakthrough of computer, robot control, and sensing technology, the development of unmanned vehicles has entered a stage of rapid development. It can reduce driver mistakes, bring convenience to the daily travel of humans, and it is widely used in the military and dangerous fields. However, the unmanned vehicle itself has strong nonlinearity, signal delay, and parameter uncertainty and its control is affected by external factors such as the change of road adhesion coefficient and lateral wind. Therefore, traditional control methods often face challenges in controlling the vehicle stably and accurately. The learning, adaptive, and approximate nonlinear mapping abilities of neural networks provide an effective way to solve the problems of vehicle model parameter uncertainty change, external disturbance, and vehicle adaptive control. Therefore, it is increasingly applied to the motion control of unmanned vehicles. The self-learning and adaptive ability of neural networks enable them to calculate the direct output control quantity according to the state deviation of the vehicle, which can be used as the controller of the unmanned vehicle. The ability of the neural networks to approach a nonlinear mapping makes it possible to approach the unknown dynamic parts of the vehicle, such as the uncertain parameters and external disturbances, which improves the accuracy and robustness of the controller design. The neural networks can remember previous information that can be used to calculate the current output. Thus, the neural networks can be used as the vehicle state observer to estimate the vehicle state parameters. The adaptive ability of the neural networks enables them to optimize the parameters of other control algorithms online. From these aspects, this paper summarized and classified the achievements and progress made by domestic and foreign scholars in applying neural networks to the motion control of unmanned vehicles in recent years, introduced the application situation, and evaluated the advantages and disadvantages. Finally, the main problems of neural networks in the motion control of unmanned vehicles were summarized and the possible development direction was prospected.