Minimum-Time Control of Robotic Manipulators using a Back Propagation Neural Network

Abstract

Algorithmic control system designs may deteriorate considerably in the presence of model uncertainty. In contrast, neural network controllers provide a non-algorithmic approach that does not depend on a mathematical model of the controlled system. Here, we utilize a back propagation neural network for the minimum-time control of a robotic manipulator assuming a bang-bang solution. This implies that the solutions obtained may be suboptimal in some cases, but will be easier to obtain and implement than true minimum-time solutions. The approach is applied to a single-link manipulator with nonlinear gravitational term where the minimum-time control is bang-bang. Generalization to the n-link manipulator case is also discussed.