Autonomous Two-wheeled Balancing Vehicle Intelligent Control Research

For the past few years, congestion of urban traffic is getting from bad to worse. The tail gas of motor vehicles results in more serious environmental pollution. And noise pollution causes a growing influence on people’s life. Facing the competitive urban life, designing a new-type means of transportation appears very immediate. Two-wheeled self-balancing vehicle is a novel transportation with the advantage of portable, low power consumption, no pollution and no noise, which becomes more and more popular since its introduction. It has been used in many fields and brought us great convenience in our daily work and life. But the safety problems can’t be ignored.Two-wheeled self-balancing vehicle has attracted a great deal of attention in recent years.Control technique is not only the most important part for the vehicle to keep the balance, but also the core technology of the entire system. To solve the safety problems, it’s obligatorily need to do further research in control. For the two-wheeled self-balancing vehicle with under-actuation, nonlinearity, instability, uncertainty, it’s not very easy to design the controller.In early research, there has been much work in state-space feedback control based on accurate model, robust control based on nonlinear model, and LQR control, sliding mode control, fuzzy control, neural network control and so on. Although great achievements have been made, researchers did not stop the racing ahead.The main content of this paper is control algorithm of the system. Stable controller has been designed, which is proposed to improve the robust performance of the system. Fist of all,the dynamical equation of the system was established using Lagrangian, and the controller are designed based on the subsystem. PID control as an traditional technique has been verified in the Two-wheeled self-balancing vehicle system. Then we present an adaptive robust controller and an adaptive neural control scheme. The neural controller is designed using back-stepping approach with full state feedback, and the extreme learning machine is utilized to adjust the neural controller. All of the methods are validated by MATLAB simulation, and then are applied on the self-balanced vehicle.