High precision control of a flexible arm by neural networks

A special feature of the back-error propagation algorithm on neural networks is shown as a key component of a control system for ruling a flexible arm which carries a payload quickly along a circular trajectory with almost zero-stroke end acceleration and vibration. The leading idea is to commit the network to learning those aspects of the control law which are not strictly available from the theory of optimal control. Therefore, special care was taken both to define the province of the network and to supply it with all the above theoretical results and with the available information relevant for solving the problem. The resulting control system guarantees a tip velocity equal to about 1 m/s and a vibration width of less than 0.1 mm