Highly Efficient Intelligent Learning Control for Manipulation Robots by Feedforward Neural Networks

Abstract A major objective in this paper is the application of feedfonranl connectionist architectures for efficient on-line learning of dynamic relations used in the synthesis of adaptive learning robot control. The new fast learning algorithms using deterministic single layer approach "black box" multilayer approach and hybrid structure approach are proposed. The important feature of new learning control structures is fast convergence properties, because the problem of adjusting the weight of internal hidden units in multilayer perceptrons is considered as a problem of estimating parameters by well-known identification methods. The efficiency of proposed approach is shown on the simulation examples of robot trajectory tracking.