Kinematic model for manipulator robot control based on neural networks

In this paper, we consider the problem of controlling robotic manipula-tors based on trained artificial neural networks. It is based on the tasks of positioning manipulators, depending on the coordinates of the target point in Cartesian space. The direct and inverse kinematics problems have different methods, techniques, and algorithms for solving them. The authors propose calculating the coordinates of the manipulators, i.e. solving the tasks of the direct and inverse kinematics problems us-ing a trained neural networks based on the created kinematic model of the robot ma-nipulator. The mathematical model provides calculated data for neural networks training and is based on the Denavit–Hartenberg representation (DH representation), which allows us to obtain a homogeneous transformation matrix with a dimension of 4 x 4 describing the position of the coordinate system of each link relative to the coor-dinate system of the previous link. The kinematic model of the robot is implemented in the Matlab program using the Robotics System Toolbox. To do this, a function has been created that sets the structure of the manipulator and its parameters using the DH representation. Experimental studies of the model and various types of neural networks have been conducted, and appropriate algorithms and programs have been written for all processes. The conducted experimental studies allow us to judge the possibility of using the developed methods for solving kinematic problems of multi-link manipulators based on neural networks. The use of this approach is also relevant in aerospace for manipulator control in production and in space.