Neural Network Based Approach to Positioning Task for the End-Effector of a Four-Joint Manipulator

One of the urgent problems of modular robotics is the lack of universal algorithms that ensure the recalculation of joint kinematics when changing the configuration of the entire system. This problem has a number of peculiarities: for each new configuration of heterogeneous modules, it is necessary to solve kinematics problems; the classical method of solving kinematics problems does not always have a solution, the achievement by the mechanism of one of the target points of space can be carried out through various positions of its links from which it is necessary to choose the best option. To find a solution for this problem, various alternative solutions for reaching the target point by the end-effector were analyzed. Several neural networks of direct propagation were tested, which are used in positioning systems for manipulator links for solving the forward (FKP) and inverse (IKP) kinematics problems.