Compatible Convex–Nonconvex Constrained QP-Based Dual Neural Networks for Motion Planning of Redundant Robot Manipulators

Abstract

Redundant robot manipulators possess huge potential of applications because of their superior flexibility and outstanding accuracy, but their real-time control is a challenging problem. In this brief, a novel compatible convex-nonconvex constrained quadratic programming (CCNC-QP)-based dual neural network (DNN) scheme is proposed for motion planning of redundant robot manipulators. The proposed CCNC-QP-DNN scheme not only has the advantages of DNN, e.g., parallel processing and real-time control, but also possesses the advantages of CCNC-QP, such as the zeroing initial error, considering convex or nonconvex constraints. Being different from most neural networks, the proposed approach is training-free and is able to track reference signals with superior accuracy and speedability. The detailed derivation process and theoretical analysis are presented. Computer simulations with five end-effector tasks verify the effectiveness and accuracy of the proposed control method in both the convex constraints condition and nonconvex constraints condition whether an initial error exists or not.