Application of neural networks in robotic control and design of mechanisms

Balasubramanian, Raghu (1993) Application of neural networks in robotic control and design of mechanisms. Masters thesis, Memorial University of Newfoundland.

[English] PDF (Migrated (PDF/A Conversion) from original format: (application/pdf)) - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (19MB) [English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. (Original Version)

Abstract

Neural network has been widely used in various fields of robotics. In this work, the neural network analysis using backpropagation algorithm is applied to the inverse velocity analysis of robotic manipulators near the singularity points accounting for the tracking error and feasibility of joint velocities. The inverse computations using the pseudo-inverse of the Jacobian matrix are compared with those obtained by the neural network analysis. The results illustrated using examples of two well known manipulators show the advantages of using the present work. A new learning algorithm called LP-neuro method is then developed to solve neural network problems, in this algorithm, the weights are obtained by a combination of Linear Programming having a sparse coefficient matrix and a single variable non-linear optimization method. The results are illustrated by solving three different problems, two of which are useful in the on-line control of robotic manipulators. The designs of a function generator and a four-bar mechanism whose coupler curve passes through nine specified points, have been carried out using neural network methods. The design problem has been solved using non-linear techniques which yield a weight matrix in each of the cases. The accuracy of the methods is also discussed. Finally, gain parameters required for the trajectory control are evaluated using non- linear optimization method. Neural network is then trained to evaluate the gain parameters based on error history of different trajectories.

Actions (login required)

View Item