Structured neural-network approach to robot motion control

It is shown that neural network techniques can be used to control the motion of a robot. This is done by applying a structured approach, i.e., by decomposing the overall control system into its main components (linearizer, acceleration controller, and inverse kinematics) and describing each component by several smaller networks. It is then possible to train each network effectively and interlink them to produce smooth control of the robot. A principal advantage of using a neural network as the controller is that it can be used to specify any controller behavior. For the given task of moving the robot arm from initial rest position to a final specified position, the position profiles showed that, using neural networks, the manipulator arm could be moved smoothly. The torque profiles clearly revealed the robustness of the neural network