A multitask neuromorphic controller for redundant robots

In this article, we propose a multitask neuromorphic controller with a hierarchical architecture, which consists of two artificial neural network (ANN) sub-systems. Based on Hopfield model, the higher level neural network system is designed to solve kinematics problems for redundant robots with several constraints in an environment of collision-free. The lower neural network system at servolevel, built on backpropagation (BP) algorithm, is employed to control joints of the manipulator with approximate dynamic model to track the reference trajectory accurately. The stability characteristics of the subcontroller and the convergence property of the ANNs are mathematically analyzed. Furthermore, improvements on learning of the proposed ANNs are also addressed in this paper