Solution Of The Manipulator Inverse Based On Neural Network

Manipulator is a kind of robot, it can be used in all aspects of social life, such as machine control, explosive assembly, handling or remove, and fire extinguishing, anti-terrorist altitude dangerous applications replace or auxiliary people finish high-risk work, has the important practical significance.Currently, the robot research includes three aspects: Firstly, research in robot kinematics; Secondly, the mechanical dynamics of hand; thirdly manipulator trajectory planning research. In these three areas of study, the robot kinematics is the foundation. Manipulator Kinematics includes the establishment of the manipulator motions model and the positive and inverse solutions of the manipulator motion equation. The inverse solution of the manipulator motion equations is directly related to the robot motion analysis, offline programming, trajectory planning and real-time control and so on. Therefore, the inverse solution of the manipulator motion equation is a very important issue of manipulator Kinematics.SCARA-type manipulator is researched in this paper. Its manipulator kinematics is analyzed, the manipulator motions model is established by using DH method and the positive kinematics equations are deduced. The researches of SCARA-type manipulator provide the theoretical basis for the simulation experiments in the later chapters.Manipulator motion equation is a nonlinear system and neural network has powerful approximation ability for nonlinear system, so the neural network which realizes nonlinear mapping from the work variable space to the joint variables space is used to solve manipulator motion equation and obtain the inverse solution. Using neural network, non-linear function approximation ability, we can obtain the inverse robot movement equation.The BP algorithm and the RBF neural network are used widely in neural networks. Because traditional BP algorithm in training neural networks has some drawbacks, such as slow search speed, low-solving accuracy y, and easy to fall into local minimum, the genetic algorithm is proposed based on BP neural network to learn the connection weights of neural networks, solving the inverse solution of manipulator motion equation. Theoretical analysis and numerical simulation results show that BP neural network optimized by genetic algorithm is practical and feasible to solve the inverse solution of manipulator motion equation and increases convergence speed and solution accuracy compared with the traditional BP algorithm. For the RBF neural network, the focus is to determine the structure of hidden layer. In this paper, the central value of the hidden layer nodes can be selected quickly and efficiently and avoided local minimum problems by using the genetic algorithm. Finally, on the SCARA-type robot simulation results show that the principle of genetic algorithm based RBF neural network convergence speed, generalization ability, can improve the accuracy of the inverse solution of manipulator motion equation.