A Novel Data-Driven DRNN-SMC Model for Redundant Manipulators

Abstract

The robot industry is developing rapidly, and how to control the redundant manipulators precisely and effectively has become a new hot topic in industry’s development. In recent years, many scholars in the industry have also proposed various control methods. However, most of these methods are proposed assuming that the Jacobian matrix is known. Actually, in practical applications, the detailed information of Jacobian matrix is often not precisely known. Therefore, this article develops a novel data-driven recurrent neural network (RNN) model that can update the Jacobian matrix and joint angles. By defining two dynamic error functions, two RNN designed formulas are used to obtain a continuous RNN (CRNN) model. Subsequently, the CRNN model is discretized by using Euler forward formula, and a discrete RNN (DRNN) model is generated. Then, a classic sliding mode control (SMC) algorithm is introduced, and DRNN-SMC model is further proposed. Moreover, the corresponding rigorous mathematical derivation and proof are carried out. In addition, simulation tests are carried out by using the Kinova Gen2 manipulator, comparing the DRNN model and PD controller, as well as the DRNN-SMC model and DRNN model, validating the precision of the DRNN-SMC model. Additionally, practical experiments using the Kinova Gen3 manipulator are performed to showcase the applicability and versatility of the DRNN-SMC model.