Adaptive Motion Control For Two-Wheeled Robot

Two-wheeled robot is one of the most active research areas involving disciplines such as robotics, computer technology, automation, artificial intelligence, and due to its simple structure and flexible control, it is widely used in industrial production and household services.Because two-wheeled robot has characteristics of kinematics non-holonomic constraint, dynamics nonlinear saturation and two drive system mutual coupled on its structure, and exists uncertain factors such as system time-varying and strong interference in the movement process, which brings great difficulties to motion control for two-wheeled robot. Therefore, the motion control for two-wheeled robot is the important and difficult problem for robotics research.This paper is focus on the point stabilization control. Most point stabilization control methods carry out in off-line condition. These methods design control algorithms by controlled object model which is obtained from established model, use some optimized algorithms to adjust controller parameters, and at last realize motion control for two-wheeled robot. However, offline optimized controller cann‘t adapt well the change of work condition and performance of the robot itself during movement process, which will result in decrease of control quality.To solve these problems, the main research of this paper is as follows:(1) put forward a two-level structure point stabilization adaptive controllerThe controller consists of kinematic parameters adaptive level and dynamic model reference adaptive level.Aiming at the characteristics of uncertain factors and strong interference in movement process, Kinematic parameters adaptive level uses human-simulated intelligent controller to set parameters online based on BP neural network. This controller perceives different motion state by the detection of linear velocity and angular velocity for robot, divides the motion state into different modes with different controllers, and then uses BP neural network setting controller parameters, to achieve the velocity control and drive current control of left and right motor for system.Aiming at the dynamics characteristics of multivariable, dynamics model reference adaptive level designs dynamic controller based on model reference adaptive for two-wheeled robot. Make quasi-equivalent model of motor system as reference model for control system of the robot system each, and design the online state observer based on Kalman filter to online estimate characteristic state of dynamic model. And then, online identify motor model by neural network identifier and transmit gradient information to neural network controller. At last, online adjust weights of the neural network controller by making use of error between reference model output and motor system output, to make the actual output for controlled object consistent with reference model output. It can realize the adjustment of dynamic characteristic for two-wheeled robot and improve the control performance of the system by this method.(2) build simulation platform and physical platform based on quasi-equivalent model for two-wheeled robot motion, and carry out point stabilization control simulation experiment and physical experiment for two-wheeled robot.It proves two-level structure point stabilization controller can keep good control performance, ensure the stability and robustness of the system and achieve the point stabilization control for two-wheeled robot by simulation experiment and physical experiment.