Recently several learning and adaptive control schemes for robotic manipulators have been proposed. However, they are no better than perturbation learning schemes. So, experience obtained during learning cannot be used for the execution of a quite different movements. In this thesis, an adaptive learning control scheme for manipulator is proposed. The proposed control scheme enables robots to move faster and to adapt themselves to various environments through learning. The control architecture is divided into two parts. One is the inverse dynamics model learning part, the other is the trajectory planning part. The former is the same as Kawato's feedback error learning scheme which requires neither an accurate model nor parameter estimation but can generalize learned movements, the latter is accomplished by the stochastic automation which updates its own performance during operation. Simulations attempting to control a 2 DOF planar arm and a 3 DOF PUMA arm were each accomplished and the selected results are presented.