Iterative self-improvement of force feedback control in contour tracking

A very general three-level learning method for self-improvement of the parameters of a force feedback controller is demonstrated in contour tracking tasks. It is assumed that no model is known a priori, either of the robot or of the contour to be tracked. The system identifies such a model, including information about its reliability. The model and estimated noise were used to generate optimal control actions for the sample trajectory. They were then used for estimation of the parameters of the controller. This controller then produces a new trajectory, which in turn could be optimized and trained. Kalman filter techniques were applied in all adaptation levels involved. Learning was possible off-line or online. The model and controller may be based on linear difference equations or include nonlinear mappings as associative or tabular memories or neural networks. It was shown that even for a linear controller substantial improvements could be attained as no assumptions were needed about the bandwidth