On correcting systematic errors without analyzing them by performing a repetitive task

Describes a method for reducing systematic errors encountered between a true system behavior and the one predicted by a model. The structure of the model corresponds to the structure of the system only up to a certain limit. Error correcting is formulated as an optimization problem where the norm of the difference between the measured and the predicted system behavior is minimized. The solution is searched iteratively by doing the same task or experiment repeatedly and utilizing previously observed results. It is argued that the optimization approach may be useful in understanding the problems encountered in memory-based modeling, particularly in robot control. An iterative algorithm is given to correct the robot positioning errors. It is shown to converge to the right solution by making some general assumptions of the existing robot controller. An example is given with the PUMA robot where the precision of the arm movement is increased by repeatedly doing the movement task