Noise-tolerant zeroing neurodynamic algorithm for upper limb motion intention-based human–robot interaction control in non-ideal conditions

Abstract

Human–robot interaction control, a crucial technology in industrial manufacturing and rehabilitation robot, has been widely explored in recent years. However, there are still some problems in the existing strategy need to be solved for the rehabilitation robot: First, the human–robot confrontation (HRC) in motion control affects the information interaction. Second, in non-ideal conditions, the safety and accuracy of the control system will be disturbed by external unknown factors. Third, the existing continuous-time controllers are hard to be applied to hardware platforms. In this paper, an upper limb-exoskeleton coupling dynamics model based on human active motion intention is established to overcome HRC, and promote active and natural human–robot interaction control. Specifically, the Elman neural network (ENN) model is devised and utilized to estimate human active motion intention using the surface electromyography (sEMG) signals. On the basis of human active motion intention, the discrete-time controller predicated on noise-tolerant zeroing neurodynamic model (termed as NTZND controller) is proposed for constructing the discrete-time human–robot interaction control system (DTHRICS) under non-ideal conditions. In addition, the zeroing neurodynamic (ZND) controller, the gradient neural network (GNN) and proportional–derivative (PD) controllers are designed for comparisons. Theoretical analyses demonstrate that the 0-stability, consistency and convergence of the designed DTHRICS. The performance indicators root mean square errors (RMSEs) and coefficient of determination (R2) of the DTHRICS with different controllers are analyzed and compared under ideal and non-ideal conditions. Ultimately, numerical results illustrate that the NTZND-based DTHRICS (with the lowest performance RMSEs →0 and the highest performance R2→1, p<0.05) is significantly superior to the DTHRICS with the ZND, the GNN and the PD controllers in terms of control accuracy, noise-tolerant property and real time performance, while utilized for upper limb multi-joint motion control under non-ideal conditions.