Neural network models based on regularization techniques for off-line robot manipulator path planning

A novel approach for continuous function approximation using a two-stage neural network model, involving regularization techniques, is herein presented. The suggested method can be applied to real functions of many variables as in robot path planning problems. It involves a regularized Kohonen feature map (SOFM) in the first stage which aims at quantizing the input variable space into smaller regions representative of the input space probability distribution and preserving its original topology, while increasing, on the other hand, cluster distances. This is achieved through adapting not only the winning neuron and its neighboring neurons weights but, also, losing neurons weights during map's convergence phase. Losing neurons weights are adapted in a manner similar to that of LVQ, by increasing the distance between these weights vectors and the corresponding input data vectors. During convergence phase of the map a group of Support Vector Machines (SVM), associated with its codebook vectors, is simultaneously trained in an online fashion so that each SVM learns to respond when the input data belongs to the topological space represented by its corresponding codebook vector. Moreover, these SVMs follow a task specific regularization strategy which aims at incorporating additional information in their training process. The proposed methodology is applied to the design of a neural-adaptive controller, by involving the computer-torque approach, which combines the regularized two-stage neural network model with a servo PD feedback controller. For this task, the regularization technique aims at filtering SVMs outputs so that their values become closer to that of a PD feedback controller, while compensating the nonlinear terms of the error, as regards the estimated torque, introduced in the robotic manipulator by employing the PD controller.