The study of the kinematic structures of multi-link manipulators is not a trivial task. Difficulties appear in the study of kinematic schemes in which the number of moving independent parts exceeds the number of generalized independent degrees of mobility, since in such cases there is an overabundance of mobility of the system and more than two manipulator configurations can exist for one position. Such ambiguity greatly complicates the search for dependencies between the generalized coordinates of the system and the endpoint of movement of the working body. It is also necessary to take into account that in the future, when compiling dynamic equations of motion for such systems, such ambiguity complicates the study of problems of dynamics. The solutions to such problems are carried out by different methods, and this article discusses the possibility of using neural networks to study such ambiguous problems using the example of a kinematic study of an articulated manipulator. In the presented work, the possibilities of using the Tensor-Flow library from Google are considered with the help of which models of neural networks and neurons are created, as well as functions for finding the weight coefficients of setting the created network model. In the course of the study, a neural network was created to calculate the regression function of the dependencies between the geometric coordinates of the point of movement of the manipulator gripper and the generalized coordinates of its boom system.The purpose of this study is to demonstrate the capabilities of neural networks in solving engineering problems that require complex mathematical transformations.The use of neural network architectures in manipulator control systems makes it possible to create universal mechanisms for performing various technological procedures, while reducing the cost of developing such solutions.In general, neural networks are not universal means for solving all engineering problems, since they require a larger set of test samples to adjust (train) the parameters of their models however, there are a number of problems in solving which neural networks have great advantages.