Neural Network Control of an Undersea Manipulator Constrained to Planar Motion

摘要

This report discusses the development of an artificial neural network controllerfor a manipulator moving in two dimensional space. The motion was limited to the shoulder and elbow joints. Two tasks were defined within this effort as follows: (1) teach the manipulator to move to a fixed target position from an arbitrary initial position, and (2) teach the manipulator to move to any arbitrary target position from any arbitrary initial position. The back propagation learning/algorithm was implemented for both subtasks. The trained neural network was initially tested using a graphics simulation program and later transferred to the manipulator hardware.