Annotation

Annotation: Building a multivariate control system for a manipulation robot based on solving its inverse dynamics problem is an effective approach to eliminating the interactions between degrees of freedom. One of the main factors limiting its application is the high computational complexity associated with the significant degree of nonlinearity of the manipulator's dynamic model. In this regard, it is important to consider intelligent knowledge processing approaches capable of approximating nonlinear functions of many variables. Problems of this kind are effectively solved by neuro-fuzzy systems, of which adaptive neuro-fuzzy inference systems (ANFIS) are currently the most widely used. This paper is devoted to the development of a multivariate control system for a manipulation robot based on these systems. A three-link manipulator with PUMA (Programmable Universal Manipulation Arm) kinematics is considered as the control object, since this kinematics has become widespread in industrial robotics. Ten ANFIS structures, which together approximate the solution to the inverse dynamics problem for this manipulator, are trained using analytical data. This paper presents two types of multivariate control systems: one with direct dynamic compensation for the mutual influence of degrees of freedom and one with feedback-based linearization of the plant. The approximation of the solution to the inverse dynamics problem required in both systems is provided by the implemented ANFIS structures. The developed neuro-fuzzy control systems are compared with a system based on proportional-integral-differential controllers (PID controllers). Experimental studies to estimate the tracking error during the execution of a given trajectory were conducted using a mathematical modeling environment. According to the obtained results, direct dynamic compensation enabled a tenfold increase in accuracy, while feedback-based linearization enabled a thousandfold increase in accuracy compared to a control system based on PID controllers.