A neural-networks scheme for robot positioning control

As various robot manipulators and controllers are designed and built, an intelligent device-independent robot manipulator control scheme must be developed for an unmanned manufacturing cell. In this study, a neural-networks based approach has been adopted to control a robot's point-to-point positioning capability. This control scheme lets a robot learn and store the knowledge and adjust itself to maintain its process capability. The approach includes using a modified two-layer counterpropagation network (MTL-CPN) algorithm and efficient training method. Such an architecture can accommodate different robot systems, and is suitable for a variety of tasks and working envelopes.