A dual neural network applied to drift-free resolution of five-link planar robot arm

Abstract

In this paper, a recurrent neural network (termed, dual neural network) is revisited and applied to the online joint angle drift-free redundancy-resolution of a five-link planar robot manipulator. To do this, a drift-free criterion is exploited in the form of a quadratic function. In addition, the drift-free scheme could incorporate multiple joint physical limits such as joint limits and joint velocity limits simultaneously. Such a scheme is finally reformulated as a quadratic-programming (QP) problem. Similar to other new types of recurrent neural networks, the dual neural network is piecewise-linear as well and has a simple architecture of only one layer of neurons. As a QP real-time solver, the dual neural network could globally exponentially converge to the optimal solution of a strictly-convex quadratic program. This suits well our scheme formulation on drift-free redundancy-resolution of robots. The dual neural network is then simulated based on the five-link planar robot manipulator, which substantiates the effectiveness of the joint-angle-drift-free neural resolution scheme.