The Neural Network Control Of Robot Based On Genetic Algorithm

This paper uses robot (Googol GRB-400 SCARA) as researched object that is of strong coupling, nonlinear and multi-variable characters. The main task is to discuss the motion control based on neural network; the structure of the above neural network is optimized by basic genetic algorithm and messy genetic algorithm so that the motion control is better executed.After reviewing the present situation and developing trend of the robot technology and industrial robot, firstly, this paper summarizes some often used methods of establishing kinematics model of robot, and it puts emphasis on the method of the Homogeneous matrix and establishes the kinematics model of GRB-400 SCARA by this method; the inverse kinematics problemis solved. Secondly, this paper makes researches on traditional control methods and analyzes their merits and demerits, and it summarizes several typical intelligent control methods such as the network control and fuzzy control; moreover, it also summarizes composite methods combining traditional control and intelligent control. Thirdly, this paper summarizes the basic theories and its characters of neural network; according to the above theory, a method of neural network control is proposed and used to have control on the trajectory tracking of robot. Facts simulation shows that there are some degrees of tracking errors and tracking effect is not good. In fact, when neural network is used to solve the actual problems, the reason that solutions are not content with the expected effect is mostly unreasonable structure of neural network. Moreover, there are no uniform design rules of neural network but the researchers' experience. In order to improve the structure of neural network, two methods are put forward in this paper: one method is to use basic genetic algorithm to optimize the parameter of Sigmoid function; the other method is to utilize messy genetic algorithm to determine the numbers of hidden-layer node of neuralnetwork. Facts simulation shows that after using the optimized neural network, not only the convergent velocity becomes faster and tracking errors become smaller but also tracking effect is preferable. In the meantime, it shows composite control method is an important aspect in the techniques of future robot control.