A Neural Control Architecture for Joint-Drift-Free and Fault-Tolerant Redundant Robot Manipulators

Abstract

Fault tolerance is important for a redundant robot manipulator, which endows the robot with the capability of finishing the end-effector task even when one or some of joints’ motion fails. In this paper, a varying-parameter neural control architecture is designed to achieve fault tolerance for redundant robot manipulators. Specifically, a quadratic programming (QP)-based fault-tolerant motion planning scheme is formulated. Second, a varying parameter recurrent neural network (VP-RNN) is proposed to resolve the standard QP problem, which can make the remaining healthy joints to remedy the whole system which is effected by faulty joints and complete the expected end-effector path. Theoretical analysis based on Lyapunov stability theory proves that the proposed VP-RNN solver can globally converge to the optimal solution to the fault-tolerant motion planning scheme, and the joint motion failure problems are solved successfully. Computer simulations and physical experiments based on a 6 degrees-of-freedom Kinova Jaco2 robot substantiate the effectiveness of the proposed varying-parameter neural control architecture for fault-tolerant motion planning scheme to redundant robot manipulators.