Direct learning of feedforward control for manipulator path tracking

The author considers a motor control task such as manipulator path tracking that could be solved by applying an appropriate feedforward control program. This program depends on a vector of the task parameters. A direct feedforward program learning control approach that is based on iteratively learning and storing the control programs for some values of the parameter vector is presented. To implement the concept several subproblems need to be solved. One is discretization to obtain a compact parametric representation of the task input and output data. Another is development of an iterative adaptive learning procedure. An important step is to find a method for control program approximation over the task parameter domain. These subproblems are discussed. This approach is applied to a manipulator path tracking problem. The results of its experimental implementation are examined.<>